“Building reliable industry-specific AI across a wide range of use cases for a highly regulated industry is hard. It takes time, focus, and the right skills. It integrates proprietary datasets, sophisticated logic, validated processes, and depends on specialized domain expertise and safeguards to maintain compliance and data integrity. If done well, our agents will provide significant value for customers and Veeva.”

— Veeva’s Q4’FY26 prepared remarks

To date, Veeva has generated little, if any, revenue directly tied to AI. But while Veeva has been perhaps a little slow to capitalize on AI vs. certain other “applied AI” winners (e.g. Cursor, OpenEvidence, Harvey, Sierra, etc.), I think the opportunity is still ripe for AI in life sciences and that Veeva is well positioned to capitalize on it.

Veeva being slower to capture the AI opportunity is not necessarily surprising: (1) there is extreme regulatory complexity in life sciences, and thus AI solutions that touch clinical/post-clinical processes need extra time to bake, (2) as you’ll see below, the degree of difficulty on their AI opportunities I think is a tick (or two) higher than perhaps some of the “lower hanging fruit” captured by the AI-natives listed above, and (3) Veeva was also a bit slow to react on AI initially.

Veeva’s AI Vision

Veeva’s AI vision has come a lot more into focus over the last 12 months. In Nov-26, Veeva will “early release” Falcon, their agentic platform. Falcon looks like it will initially have 4 key areas of “applied AI,” which I detail below. Each area seems primed for an AI solution, where Veeva’s agentic offering should be “better, faster & cheaper” than the human labor alternative (at least over time, if not immediately). And all of these areas are at least reasonably sizable.

• Safety Case Intake & Processing. I touched on this opportunity in my last post on Veeva. It remains the cleanest and largest agentic opportunity for Veeva for now. Pharma companies are required to meticulously track and report on “adverse events” that are reported (through a range of different channels) for the drugs they sell. The process of collecting, summarizing, triaging and even making causation determinations around adverse events (a process also known as “pharmacovigilance”) is a ~$10bn market that is ripe for automation. Further, the surrounding Safety application suite is an area in which Veeva is taking share despite an incredibly risk-averse set of buyers. As we can see below, Veeva has increased its count of top 20 pharma from 2 to 6 over just the last ~1.5 years.

  

  Given the trajectory, perhaps Veeva can ramp to at least half of the top-20s in the coming years, which will allow Veeva to service the majority of this market with its agentic case processing offering. If the AI case processing opportunity amounts to just 20% of this $10bn market, that’s a $2bn opportunity.

• Agentic TMF12 (clinical trial document intake & quality control). I also discussed the opportunity in TMF in my last Veeva post. Clinical trials generate troves and troves of documents (in the tens if not hundreds of millions annually, according to management). These documents need to be checked for quality and properly categorized into an eTMF (electronic Trial Master File). This is highly rote work, much of which is outsourced today (the outsourced TMF market alone is estimated at ~$1.5-2.5bn annually). Veeva is exceptionally well positioned to serve this market, with all 20 of the top 20 pharma co’s using Veeva’s eTMF application. I assume the size of this market is ~$500mm.

• Regulatory Writer (interaction intake & response + agentic authoring). As discussed in my original deep dive, Veeva has the opportunity to partially automate correspondence with health authorities on behalf of its biopharma customers. This would include both the drafting of major submissions (Veeva is calling this its “agentic authoring” product, separate from Falcon) and also the more regular correspondence with health authorities (which Veeva is including under its Falcon umbrella). Veeva again dominates this category at the application layer, with 19 of the top 20 pharma co’s being customers of its RIM (Regulatory Information Management) suite. One group sizes the “AI in Medical Writing” market (specific to pharma) at $1.2bn growing to $2.0bn by 2030. McKinsey has estimated that AI-driven savings related to clinical ops will amount to $13-25bn, with an agentic regulatory writer being a major contributor (though not broken out separately). AbbVie (a top 5 pharma) has actually gone ahead and built an in-house solution already, which seems bullish for the chances that Veeva’s writer can be immediately useful. I estimate the size of this market is perhaps ~$750mm.

• Agentic MLR.3 Also as discussed in my original Veeva deep dive, AI provides pharma companies with the ability fast track the MLR (Medical, Legal & Regulatory) review process when approving new commercial (aka marketing) content. Veeva management believes a speedy MLR approval process is increasingly important in the AI era, where someone like OpenEvidence (or a pharma company’s own website) may need to offer targeted collateral instantly say while a doctor is researching a drug. Instead of sending over a 60-page slide deck on a drug and saying “look in here,” an MLR agent can help extract and approve the relevant content in a compliant manner. Veeva is targeting a 70%+ reduction in manual labor related to the MLR review process within 5 years. Veeva acquired Copli in Jun-26 to help bolster this offering. We’ll say MLR review is a $250mm opportunity.

Combined, these four initial areas represent in the ballpark of ~$3.5bn of incremental TAM for Veeva, according to my loose estimates. Further, if these agentic offerings indeed prove to be “better, faster & cheaper” than the human labor equivalent, one would expect them to be rapidly adopted by the pharma industry. How could you defend not adopting AI for safety case processing if say your next closest competitor is saving 50%+ (and potentially multiple tens of millions of dollars) with an agentic solution? I think it’s fair to say that economy-wide, for AI offerings that do truly resonate with customers (see Cursor, OpenEvidence, Harvey, Sierra), we’ve seen that demand is basically insatiable.

Longer-term, there are many additional areas of life science where AI functionality will make a great deal of sense. To name a few:

• CRM “co-pilot.” By all accounts, Veeva’s CRM/commercial suite will evolve over time from essentially static record keeping to more of a “system of action.” The goal will be for CRM to look more like an intelligent assistant (e.g. taking notes/prepping for calls, recommending actions, even providing feedback/coaching).

• Clinical trial “co-pilot.” AI can help with trial activities like optimizing trial site selection (to maximize enrollment/trial success based on candidate data like health records, biopsies and body scans), creating trial documentation and protocols (which necessitate thousands of pages of documents), drafting communications with trial sites, smartly adapting trials on the fly (e.g. changing dosages in response to early indications from the field), etc.

• Safety signal detection. Mining adverse event data for safety “signals” is a perfect use-case for AI.

• Brand engagement. Increasingly drug brands need to be able to engage patients/HCPs (healthcare providers) in an “AI chat” format (through a brand’s website, OpenEvidence, or other channels) in a compliant manner. Veeva acquired Ostro4, a leader in the space, in Mar-26 to bolster this capability.

Encouragingly, Veeva is increasingly pumping out AI capabilities of all different flavors, as we can see on the AI section of their website.

In all, McKinsey estimates the savings that can be generated across life sciences from AI (in just the areas that Veeva touches — i.e. post “research and early discovery”) could amount to $38-67bn. If we assume Veeva can capture 20% of that, that would expand Veeva’s TAM by ~$8-13bn (vs. an estimated TAM of ~$20bn pre AI and total revenue of ~$3bn today).

Thrive, Sequoia, a16z and Sam Altman have all backed an “AI-native pharma company” called Formation Bio, who’s goal is to drastically speed up the drug development process. Formation doesn’t actually do any drug discovery work, but rather acquires drug candidates and helps speed them through the clinical trial process. Formation’s contention is that they can ultimately reduce the cost of clinical trials by ~50% while also speeding trials up by ~50%. Much of this, it seems, Formation believes can be achieved by cutting out CROs (contract research organizations, or outsourced managers of clinical trials, regulatory compliance, etc.) through, in part, AI-driven processes. Here is CEO Ben Liu from a Nov-25 a16z interview:

“One of the big challenges is most biotech and pharma companies today hire a CRO. The CROs are a bill-per-hour, cost-plus business…So the business model means the more billable hours, the more revenue they earn, right?…So there is, not for any other reason than just the financial incentives of cost-plus business models, usually means there’s cost inflation over time. So let’s say you’re running a phase two trial using 50 different trial sites. All those trial sites operate a little bit differently. Some collect data using one system, you know, at the site level, then they put it into an electronic data capture system. These things don’t talk with each other. But for a CRO, that’s positive because it’s a ‘bill-per-hour event’ to address that variability. So, a big way we’re able to run trials a lot faster and more cost effectively is we built our own tech-enabled CRO where we built all the software technology to disintermediate a lot of those cost-plus, bill-per-hour activities…

You could probably do trials for half the cost, half the time. Because, you know, the cost-plus of a CRO is like a 40% margin…One heuristic is for every $3, $1 goes to the site, like the site investigator, and $2 goes to the CRO, right? So there’s a lot of opportunities on the, like, human capital intensive type of activities. If you begin to also think about study design, then you can get really interesting order-of-magnitude kind of differences, but that requires some regulatory [intervention/re-design].”

Much of Veeva’s agentic offerings will be targeting the same categories of efficiencies (that often fall to a CRO), but instead making them widely available across the pharma industry. Thus, I view Formation’s thesis as bullish re Veeva’s AI efforts.

As I’ve mentioned before, Veeva CEO Peter Gassner has also stated that he believes AI (from Veeva alone) could lead to a ~15% increase in efficiency across all of life sciences within ~5 years.5 Across a $2 trillion industry, that would amount to ~$300bn in savings. This seems like a prediction he might want to have back at some point, but even if this misses by a significant margin it still indicates a large opportunity set for Veeva.

AI’s “Pull-Through” Potential

An important side note to the AI discussion is that if Veeva’s agentic offerings are indeed compelling, this should accelerate the uptake of its “legacy” software offerings through which this AI is delivered.

For example, from the chart above, only 6 top-20 pharma co’s have adopted Veeva Safety. If Veeva’s agentic safety offering is ultimately industry-leading, it would not be surprising to see this number ratchet up. Veeva management would tell you they already have the best application in the space, with the top 2 competitors supposedly offering “legacy” on-premise solutions that have grown long in the tooth (take with a grain of salt).67

Similarly, while Veeva owns 19 of the top-20s in its RIM suite, a great AI regulatory writer could help spur adoption of other suites that an AI writer might ideally want to access. For example, in order to draft a clinical trial submission, an AI writer will need clean access to the clinical trial data. If this data is not housed in Veeva’s CDB application, where Veeva counts ~8 top-20s as customers, perhaps the writer won’t work as well. The same could probably be said for the eTMF and CTMS. Another area that would make a lot of sense for an AI writer to access would be the safety suite. Pharma co’s need to submit regular reporting on safety activity, as well as ad hoc reporting when adverse events build/cluster.

Perhaps the most interesting “pull through” opportunity from AI, however, is in Veeva’s data products (more specifically its Compass suite). Veeva’s Compass suite is a “sleeping giant,” representing the largest single product opportunity for Veeva with an estimated TAM of ~$4bn+. The opportunity in Compass could be unlocked, in part, through a more “agentic” CRM/commercial paradigm.

An agentic co-pilot in CRM will need access to real-time, market-wide patient- and prescriber-level data to function well. In order to recommend actions (which HCPs to prioritize? what’s changed since my last visit? what message should I use?), for example, the CRM will need detailed, up-to-date commercial/market data (e.g. patient diagnoses/procedures, HCP prescribing behaviors, market share trends).

In fact, Veeva already powered the first iteration of its “CRM bot,” which it released in Dec-25, with data from Compass. From Veeva’s Q1’FY24 prepared remarks:

“CRM Bot is an AI application for Vault CRM. You can think of it as ChatGPT for field teams. We will start by powering it with Veeva data, specifically Compass. We’ll learn and evolve from there, with a focus on partnering with customers to get AI right for the industry.”

It turns out Veeva is one of only two players who can provide this type of data (alongside market leader IQVIA). For a decade+, IQVIA was the only game in town. However, after much effort and 7+ years of development, Veeva recently launched a set of products which can viably compete against IQVIA. Veeva management, with a straight face, will also tell you its Compass products are better than IQVIA’s along certain lines, providing: (1) richer patient-level data (which is better for today’s more modern, complex and targeted medicines) and (2) faster data (refreshed daily, which will be increasingly important in an agentic commercial world). Veeva’s data products will also, of course, plug very neatly into Veeva’s commercial application suite, which provides a nice distribution advantage (Veeva still dominates in CRM/commercial).

Here is management from a Jun-25 Baird conference on Compass’ progress:

“In many areas, our product is already the very best in the marketplace. And I say it that way because there’s some — it varies by therapeutic area. But there’s an incumbent in place and there’s a lot of effort to switch out the incumbent even though you have a better data product, and you can prove that you have a better data product. So we have to prove this brand by brand. It’s the buying cycle in life sciences that takes time. You buy one brand, you establish success and you move on to the next one. And that’s the cycle that we’re going through right now. And we’ve seen some enterprise customers start small with one and then expand to multiple brands and then expand to an ELA. That’s the pattern we’re looking for, and I think that’s what will play out over the next several years.”

Below is commentary from an Apr-26 AlphaSense transcript with Veeva’s Former Head of Commercial & Medical Sales & Strategy. He states that while getting customers to move off of IQVIA is extremely difficult, that Veeva actually has the better data product at this point.

“I think that their Compass product is very well-positioned in the sense where the data business is going to become more and more valuable because these LLMs need data. If they don’t have the proprietary data to train the model appropriately, then they’re pretty much useless models at that point. Veeva does have a data moat should they continue to grow within their Compass business, which I don’t think is necessarily grown to the extent where they anticipated at this point, which is not due to inferior products by any means.

It’s just really hard to rip out your primary data source. If someone’s using IQVIA, to move fully onto Veeva Compass for minimal uplift, it’s hard to justify that change. It’s a lot of work. People don’t want to do it. I think they need to get over that hump, and they will, as they make their data set better. That could be also the next flame for them to ignite, from a revenue perspective, quite honestly…

In the data product itself, you asked about the Compass product too, I think what they did initially and why it hasn’t captured as much market share as they thought is they launched the product too early. It wasn’t ready. The product was not robust enough to compete with the IQVIAs of the world when it came out. Now, I think it’s actually better, and it’s going to continue to get even better. It’s more so of convincing the customers to move away entirely from IQVIA and move on to Veeva, which I think some will do…

The data business, I think, is worth more than the software business. It’s going to be even more lucrative as well, like I said, with LLMs needing this data to actually operate.”

This sentiment was echoed on a May-26 Third Bridge transcript with a former Veeva Senior Strategy Exec:

“Where I think Peter is going strategically right now, and I think it creates a much bigger moat…In 2019, they acquired Crossix. That’s data on 300 million lives. You can look it up, maybe you already have, so you got real-world data…

Peter was really frustrated in the beginning, and the acquisition was to start to build this moat to say, ‘We have the data that we’re generating on the commercial side. We’re now acquiring really insightful, psychographic, demographic purchasing behavior on 300 million lives.’ Now, here’s the key. The Compass product is now integrating all of that data. Other companies can’t do it. Komodo has the data on the lives, but they don’t have all the clinical data that Veeva is generating.

Where the moat becomes strategically, and I took it to another level just beyond the Vault, is that I believe the strategy now is to become, with a strong moat, the definitive data source for the life sciences industry for the years to come by buying data from IQVIA, which they do. It’s an interesting frenemy situation, generating your own clinical data more and more and more over time. Your Crossix data keeps expanding. Then they made a significant shift with the Compass product, where Veeva, as you know, has always been upper right-hand quadrant, premium price, premium product, 24/7, we’ll give you whatever you need, just ask for it…

When you do your data comparison at your company, if you work at Pfizer, let’s say, and you’re like, ‘Man, the Veeva data is just far superior,’ the ClinOps teams, the sales ops teams, market research, everybody is like, ‘This is the data.’ I don’t know how anybody can usurp that.

Veeva was already seeing healthy uptake of Compass prior to any type of pull from AI. As we can see below, over the past ~2.5 years, the Compass Patient product has quickly scaled from around zero brands to ~140 brands (including brands with 2 top 20 pharmas).

Veeva also signed ELAs with 2 top 50 pharma groups over this period (one in the ~Q2’FY25 quarter and one in the ~Q2’FY26 quarter). If we assume there are ~4k brands that are potential Compass customers (Veeva has stated that they have data on 4k brands), that would suggest Veeva has ~3-4% market share currently (on a logo basis).

Compass’ progress is something to monitor closely. The data suite addresses a very large TAM and is only growing in importance. If Veeva starts seeing accelerating traction in Compass, that would be a very significant development.

To recap, my hope is that sometime in 2027-2028 we begin to see a reacceleration of sorts as a result of (1) growing AI consumption revenue (from Falcon), (2) the continued and hopefully somewhat accelerating adoption of Veeva’s “legacy” applications, in part spurred by the promise of Veeva’s AI products and (3) the continued and hopefully somewhat accelerating adoption of Veeva’s Compass suite, in part spurred by the industry-wide shift to an agentic commercial suite.

Valuation Refresh

In Nov-24 at their 2024 investor day, Veeva set a target of $6bn in revenue by FY31 (aka CY30), a target they continue to reaffirm.

For illustrative purposes, let’s say AI hits, and that this somewhat accelerates Veeva’s legacy software business and data offerings. As a result, we’ll say Veeva hits $7bn in revenue by FY31 instead of $6bn.

If we also assume Veeva generates a 50% non-GAAP EBIT margin (vs. 45% today), and trades for ~15x LTM non-GAAP EBIT (vs. ~22x today)8, this gets us to a ~1.7x MOI over 5 years, or an IRR of ~11% (based on a $230 share price).

Here are sensitivity tables showing MOIs and IRRs based on a range of entry share prices and exit multiples.

Based on the above, Veeva’s share price getting as low as ~$150 as part of the “Saaspocalypse” seems like a gift in retrospect. As things stand, I think I would start adding again in the ~$180 range (but this of course could go up/down depending on Veeva’s evolving trajectory or how one feels about different parts of the story).

Updates to Selected Risks

• DIY Risk. Veeva’s revenue is fairly concentrated. According to Jared Sleeper, Veeva’s top 10 customers represent ~28% of revenue (and generate almost ~$1bn in ARR). Thus, with some customers paying Veeva $100mm+, there is certainly incentive to rip Veeva out and replace with a homegrown solution. If we do end up in a world where even complex software like Veeva’s can be easily replicated, this puts Veeva at risk. Mitigants include: (1) the industry is highly regulated and Veeva has one of the most risk-averse sets of buyers in all of software (making DIY a lot less likely) and (2) if Veeva can transition into an “applied AI” winner, it will add an AI/data layer of defensibility that couldn’t be “vibe coded” and should be quite formidable. I feel comfortable with this risk for now.

• Frontier Lab Competition. Anthropic, in particular, is attempting to expand its reach in life sciences to clinical and post-clinical applications. Anthropic has added “connectors” to a host of partners including Veeva, Medidata and many others. It’s also added certain life science-specific “skills” like the ones listed below:

  

  Anthropic signed Bristol Myers Squibb (~8-10th largest pharma) to an enterprise-wide agreement in May-26 where Anthropic will assist with, amongst other things, drug commercialization, drafting clinical trial reports and driving better quality/compliance in manufacturing. Anthropic also appointed Novartis (~6-8th largest pharma) CEO Vas Narasinhan to its board in Apr-26. Merck (~4th largest pharma), for its part, signed an enterprise agreement with Google in Apr-26 reportedly worth up to $1bn (across multiple years) which calls for the deployment of an enterprise-wide agentic platform.
  
  Veeva (and the rest of SaaS) has received good news on this front of late, however. The first piece of good news has been that open-source models have kept pace (and even gained ground in certain capabilities) with the closed models. The second is that we’ve seen “applied AI” companies like Harvey and Cursor use these open-source models to train smaller models with their own data that actually outperform closed models in domain-specific areas (and do so at a far lower cost). Here is Gavin Baker from an Aug-26 Invest Like the Best podcast:
  
  “I do think open-source getting closer to the frontier and companies like Fireworks making it really easy to customize a model such that you can get, in some cases, better than frontier for performance for meaningfully lower cost. That is a godsend for the software industry and it’s also a godsend for all these AI natives…A lot of people would dismiss them as ChatGPT wrappers. Well, now with open source, you’ve actually generated some data that’s unique to your use case, whatever vertical you’re going after…And that’s why you saw, you know, Harvey, before it was acquired Cursor, lean so heavily into this. Harvey, Legora — all of them. Because if you can go from just using one, two, or three frontier models for whatever it is, to 30 to 60% of your token consumption. And then use your own RL model. All of a sudden, you’re not a wrapper. You’re way more defensible.”
  
  For all the Falcon use-cases highlighted above, I assume Veeva follows the path articulated by Baker if they haven’t already. This gives Veeva a real shot to actually outperform the frontier labs in these targeted areas with proprietary models (and do so at a far lower cost).

Other Notes

• LIMS.9 Veeva added its first top-20 LIMS customer. This was a nice milestone, as the LIMS opportunity is one of Veeva’s largest (~$1bn in size by my estimate)

• EDC.10 After showing very strong momentum in EDC, Veeva hasn’t added a single net new customers for ~2.5 years. While management is fairly adamant that growth will return (and that the last couple years have simply been an “air pocket”), this is obviously not ideal and something to watch.

• Clinical trial tailwind. More clinical trials are good for Veeva. While Veeva’s pricing is pretty convoluted and varies by product area, in general on the R&D side, the more clinical trials, the more revenue Veeva generates, all things equal. Many expect the pace of trials to increase in coming years boosted by AI-driven drug discovery (after flattish trial activity the past few years).

• Complex therapeutics. The share of complex, large molecule drugs in drug pipelines is growing rapidly (now accounting for ~50% of late-stage assets and ~65% of projected pipeline value). As discussed above, Veeva’s Compass suite is especially well suited for this class of drugs.

• CRM/commercial referendum. In Dec-22, Veeva announced its widely discussed move to take its CRM off of the Salesforce platform and onto its own internally developed “Vault” platform. This meant that Veeva would have to effectively “re-win” all of its CRM customers. Salesforce responded with a life science-specific CRM of its own, and surprised Veeva (and the industry at large) by taking a handful of top 20 customers away from Veeva (6 thus far). In hindsight, Salesforce benefited from (1) the fact that many Veeva CRM customers used Salesforce products for CRM-adjacent applications like marketing and service (Veeva wasn’t allowed to develop these applications under its prior agreement with Salesforce) which enhanced Salesforce ecosystem stickiness, (2) going all-in on an AI vision at a time when Veeva was downplaying AI’s impact, (3) that Veeva was entrenched in a legal battle with IQVIA, the top data provider in the space and critical partner for enabling an agentic commercial suite (IQVIA and Veeva ended their legal battle and entered into a commercial partnership in Aug-25), and (4) that Veeva chose to force its customers to pay for the service cost of transitioning CRMs even though its customers weren’t ultimately migrating off of Veeva (which annoyed all of Veeva’s customers to varying degrees). As of Q1’FY27, 10 top-20s had selected Veeva and 6 had selected Salesforce with 4 remaining undecided. Veeva stated on the Q1’FY27 call that they expected to win “the majority” of the remaining four. I view this as an interesting referendum on the state of Veeva’s commercial suite. Veeva has now had a chance to fully articulate its AI vision (as well as firm up a partnership with IQVIA), after a slow start re AI in which they “fell behind” Salesforce. If Veeva can win 3 or 4 of the remaining top-20s, that would seem to indicate that the market believes in Veeva’s commercial vision. If Salesforce instead surprises on wins, that might tell a different story. Veeva received good news on 8/11 when Eli Lilly (the #1 overall pharma and one of the remaining “undecideds”) announced that it had committed to Veeva.

Autonomous vehicles (AVs) are coming to Uber soon.*

We’re talking about non-Waymo, US-based, unsupervised (no safety drivers) AVs with large ODDs (operational design domains, or the areas in which the AVs are permitted/safe to operate).

What is soon? In theory, H2’26. In H2’26, we’ll have Nuro-powered Lucid AVs (the Lucid Gravity SUV) operating without safety drivers1 throughout the SF Bay Area (ex-highways).23

Full-stack NVIDIA-powered AVs are slated to arrive in the Bay Area not long behind Nuro/Lucid, in H1’27 (car OEM TBD). In fact, Uber is targeting the launch of NVIDIA-powered robotaxis (using NVIDIA’s “full stack” — chips, systems and software) across 28 cities by 2028.

Beyond Nuro and NVIDIA, there are a slew of other AV providers (and Uber partners) currently piloting AVs (or close to it), with plans to operate commercially over the next couple years. These partners include: Apollo Go (owned by Baidu), WeRide, Pony.ai, Wayve, Waabi, Zoox, Motional, Volkswagen, Rivian, May Mobility, Avride and Momenta.

By the end of 2028, in fact, Uber will likely have two or more AV providers operating commercially and unsupervised within the same city across, say, 20+ major cities.

*In actuality, Uber already has AVs operating commercially in a handful of international markets. They also do in the US through Waymo in two markets (Austin and Atlanta).

Whether or not Uber can get AVs on its platform is not the only risk that AVs present, however (in fact, I’d argue that’s probably not much of a risk at this point). I’d say the main risk, and what remains to be seen, is what increased competition (from Waymo, Tesla, and Zoox primarily) will do to the growth and margins of Uber’s mobility business.

An Uber ride is ultimately a commodity product or thereabouts. When someone wants to get from A to B, it doesn’t really matter if it’s an Uber, a Lyft, a Waymo or a Tesla. They mostly just want the best price, and a low-ish (<3-5 min) ETA. Having more options (e.g. Waymo, Tesla) in the market that can provide this is not good for Uber. Even worse is the specter that these AV offerings will get much cheaper over time, eventually displacing human drivers altogether.

Uber, for its part, is well aware of and trying to get ahead of this shift. For one, Uber has its own AV partners, as discussed above, that should allow it to keep pace with the likes of Waymo and Tesla (to what degree it can keep pace is discussed later). Another key initiative is the Uber One membership program. The company has ~50mm Uber One members (growing ~50% YoY) who represent ~50% of all bookings on the platform. It’s a hugely important competitive differentiator in the industry. Uber also benefits from the fact that half of its business does not actually come from mobility (it comes from delivery). And of the half that comes from mobility, only about 40% is generated in the US (the market most prone to AV disruption).

Ultimately, however, the reason Uber is interesting is that, even if Uber were to cede market share in its mobility business, there’s a pretty good chance it could still grow the business strongly for another decade+. This is because, as AVs proliferate, the cost of rideshare will increasingly approach the cost of personal car ownership (before eventually dipping below it, many believe). This will unlock a massive portion of the mobility market, of which Uber/Lyft represent less than 1% of today (in terms of total miles traveled). Once rideshare costs less than personal car ownership (in addition to offering the convenience of not having to park your car, charge it, maintain it, insure it, etc.), where does rideshare’s overall market share land? 10% of all miles traveled? 20%? 30%? More? As we’ll see below, there are not unreasonable scenarios in which Uber ultimately cedes market share (from ~70% today to say 50% or less) but still grows its US mobility business by 15%+ for a decade+.

Outside of the AV risk, there’s a lot to like about Uber.

The company has a lot of momentum. Revenue has grown by 17%, 18% and 18% over 2023-2025, respectively. EBITDA has grown by 137%, 60% and 35% over 2023-2025, respectively. Monthly active users have grown by 15%, 14% and 18% over 2023-2025, respectively.

As mentioned above, their membership offering is growing rapidly. Uber is the only group that can offer benefits on both mobility and delivery simultaneously — a key structural advantage they just extended to 58 countries (from 34) through the acquisition of Delivery Hero. And the Uber One program continues to get better. As of Apr-26, Uber One members can earn 10% cash back (in the form of Uber credits) on hotels/Vrbo’s booked through Uber (via their partnership with Expedia). This is on top of an additional to 20% off a rolling list of 10,000 hotels — even The Points Guy thinks this is a pretty good deal.

Further, Uber continues to have a deep moat. It operates in essentially duopolies (if not borderline monopolies) in just about every one of its markets around the world (across mobility and delivery). It is the leader or clear leader in most of these markets.

Lastly, despite rapidly inflecting profitability, the stock has traded sideways over the last ~2.5 years. Its multiple stands at just ~10x ‘27e adj EBITDA — quite “undemanding” I would say, if you are a believer in the business.

Hopefully the following write-up sheds good light on the opportunity. Between the two sides of the business (mobility + delivery), the global nature of Uber, and the rise of AVs, frankly it’s not an easy one to connect the dots on. And given how rapidly the industry is changing, this write-up could have a short self-life. However, we do seem on the cusp of a giant “s-curve” in autonomous vehicles. With Uber still the 500-lb gorilla in the space, at a minimum the company should prove to be a very interesting follow over the coming decade(s).

Business Overview

Just about everyone knows Uber. It’s an app where you press a button, and then a car shows up and takes you where you want to go. It’s also, through Uber Eats, an app where you press a button, and food shows up (or increasingly, other goods).

Uber currently serves (1) ~58 countries with both “Uber” and “Uber Eats” (i.e. “cross-platform” markets) (2) ~12 countries with “Uber” only, and (3) ~23 countries with “Uber Eats” only.

Prior to the Delivery Hero acquisition, just announced in Jul-26, Uber operated (1) ~34 “cross-platform” markets, (2) ~36 “Uber”-only markets, and (3) no “Uber Eats”-only markets.

Pro forma for the Delivery Hero acquisition, the US represents ~37% of Uber’s total revenue. On just the mobility side, the US represents ~40% of bookings.

The company was founded in 2009 by Travis Kalanick, Garrett Camp, and Ryan Graves. The idea for Uber supposedly came to Camp while, of all things, he was watching this scene from the 2006 James Bond film, Casino Royale:

After building the initial app, the three founders canvassed SF signing up drivers. In exchange for free iPhones, through which drivers could access the Uber driver app, drivers joined the platform (it turned out many drivers had plenty of free time between the occasional airport pickup, etc.). Once driver supply was live, and after a little marketing, they watched as demand flooded in. By all accounts there was just about instantaneous and overwhelming product-market fit.

Many will not remember prior to Uber, how comparatively bad the experience was of hailing a cab. You had to call the dispatch, a 1-800 number. Dispatch told you it was sending a cab, but you didn’t know if one would show up in 1 minute or 15 minutes (or if one would show up at all). Payment after the ride was a process like paying for the bill at a restaurant. With Uber, you could hail a cab by touching a button. You could see the cab’s ETA, even where it was on a map. Payment was automatic.

With clear product-market fit, Kalanick and team set about aggressively growing the business. Uber parachuted “launchers” into new cities who would hire a local team, scour Yelp for and cold call potential drivers (and initially pay them by the hour just to sit on the platform), and generate demand through mostly referrals and word-of-mouth. They often ignored complaints from taxi commissions and/or city regulators — launching first, and asking for forgiveness later. Uber Eats was launched in 2014, becoming its own dedicated app in 2015.

As competition escalated, Uber became increasingly aggressive with incentives on both the driver and rider sides. The company became one of the poster boys of ZIRP-era spending, burning prodigious sums of cash. Per the chart below, the company generated around ~$30bn in cumulative operating loses before turning EBIT positive in 2023.

In an effort to reduce its cash burn, over 2016-2020, Uber sold off its China, Russia, SE Asia and India (delivery) businesses, its self-driving technology division (“ATG”) and its eVTOL technology division (“Uber Elevate”). As these cash burning segments were clipped off, as demand pressure from COVID abated, and as ZIRP-era spending faded, the company has been able to rapidly inflect its profitability.

Despite this, however, shares have traded sideways over the last ~2.5 years. The rise of robotaxis has opened up the playing field to new entrants (e.g. Waymo, Tesla), a frightening development which seems to be keeping a lid on the share price.

This isn’t necessarily unwarranted — the rise of AV technology increases uncertainty. Waymo is the clear leader, and has a “1P” offering that completely circumvents the Uber platform. Further, Tesla is a scary specter of a competitor that is seemingly on the verge of L4 autonomy.

Investment Merits

A Natural Monopoly

I’d argue Uber has historically operated a natural monopoly. It’s a natural monopoly because in markets in which it operates, it has been “rational” for one firm to occupy the entire market.

Drivers mostly care about where they can generate the most earnings. The most earnings are generated by the most liquid marketplace, where the most riders are and the downtime is lowest. Riders care most about low pricing and reliability (ETAs and coverage). The marketplace with the most drivers has the best prices (most supply), the lowest ETAs and the best coverage.4

Thus, we have a virtuous flywheel, where the Uber service improves as more drivers/couriers and riders/consumers enter the ecosystem. Any platforms that take drivers/riders or couriers/consumers away from Uber, actually diminish the overall rideshare/delivery experience in a market.

The proof is in the proverbial pudding in that Uber has vanquished its main competitor in mobility, Lyft, basically twice (in both the Kalanick era and then again in the post Kalanick era after Lyft was given a second life). Similarly, mobility groups like Didi (rideshare in China) and Grab (rideshare + delivery in SE Asia) have also come to dominate their respective geos with huge market shares. These markets, once all players are forced to compete rationally (i.e. ex subsidies), have shown that they eventually converge to one dominant player over a long-enough timeframe.

The same, to a lesser degree, has happened in food delivery. Current delivery markets tend to have ~2-3 players, with the #1 player representing ~60%+ of a market (to oversimplify). Breaking into one of these markets today from a standing start is highly difficult. DoorDash, maybe the most sophisticated of all the food delivery operators, has been investing heavily in international expansion and the experience has not exactly been like a hot knife through butter. It’s been a slog rather, and DoorDash doesn’t seem to be making significant progress.

Autonomous vehicles pose a clear risk to the status quo. However, the post-AV world will share many of the same characteristics of the pre-AV world. Riders will care most about cost and ETAs. The owners of AVs will care most about maximizing earnings. The largest, most liquid marketplace will likely continue to win on both fronts.

Highly Attractive Financial Model

Uber is a marketplace business that takes a ~20% cut across all the transactions it helps orchestrate.

Marketplaces can be exceptional businesses. They require relatively little capital to grow and can generate high operating leverage and high margins. Uber is by far the largest “taxi” company in the world, yet it doesn’t own any cars (for now, at least).

Uber has almost 7x’d revenue since 2017 (a 27% CAGR). It has generated a 18% revenue CAGR over the last three years.

In the meantime, adj EBITDA margins have inflected dramatically — from -33% in 2017 to 17% in 2025. Uber generated ~$9bn in adj EBITDA in 2025.

Further margin expansion is expected. Much of this is the result of Uber’s low variable costs. Since 2022 the company has grown revenue by a CAGR of 18% while it has grown opex by a CAGR of just 3%.

I believe Uber will be able to wring out even more efficiencies from its cost structure through AI in the coming years.

With substantial excess free cash flow, Uber now finds itself with the capacity to invest in organic/inorganic growth and AV strategies, while also returning significant sums of capital to shareholders in the form of buybacks.

Uber One > DashPass > Lyft Pink > Instacart+

Uber seemingly has the best membership program of all the delivery/mobility players. And the advantage is structural. While DoorDash/Instacart can offer discounts on delivery, and Lyft can offer discounts on mobility, Uber can offer discounts on both. Uber is also ~4.5x the size of DoorDash, ~10x the size of Lyft and ~17x the size of Instacart (by revenue, including Delivery Hero), so at this point Uber can use its greater size and profitability to swat away attempts from competitors to grab share. These peer groups are public now as well, and attempts to ramp cash burn to take share (a highly dubious proposition at this point in its own right) would be swiftly punished by the market.

For $9.99 per month, Uber One members benefit from a $0 delivery fee (there is still a “service” fee), 6% back on rides, up to 10% off on delivery/pickup orders over certain thresholds, automatic surge savings and other exclusive offers.

For the same $9.99, a DashPass member receives $0 delivery fees, 5% back on eligible pickup orders, and certain member-exclusive deals. But of course there’s no mobility discount.5 So unless you believe DoorDash’s coverage/ETAs are materially better than Uber’s, and/or never use Uber on the mobility side, Uber One seems the better choice.

Here is Dara on a Mar-26 Semafor podcast highlighting Uber One’s differentiated position:

“Uber One has been an enormous success. It’s one of the fastest growing parts of our business. We’ve got over 46 million members now, growing well over 50% on a year-on-year basis. And close to 50% of the overall spend on Uber now comes from members.

And the way I look at membership to some extent is we compete against players who are rides-only players or delivery-only players. We have both. And our membership typically is priced at the same level [as our] competitors. And to some extent, I think we look like Netflix, which is you pay a certain amount and we just have more content than anyone else. And the player who has more content is the player who, over a long, long period of time, wins. And winning means getting you to use my services on a frequent basis hopefully for the rest of your life.

And membership, it is a trade-off for the consumer, which is I’m going to pay some money up front. It creates stickiness for you in that ecosystem because you’ve already paid. But what we see out of membership is that the use of our services, both in terms of the amount you spend — members spend about three times more than non-members — but also the cross-platform usage of members. ‘Hey, I’ve got an Uber One membership because I use Uber for rides all the time. Let me try Uber Eats because there’s no delivery fee. Why not?’ That kind of activity, the cross-platform activity also accelerates.”

The growth in Uber’s membership base backs up the attractiveness of the program. The membership program has rapidly expanded over the past four years. From 6 million members in 2021, the program exited Q1’26 with 50 million members growing ~50%. Today, Uber One members represent ~50% of total bookings on the Uber platform (across both mobility and delivery), a surprisingly large figure (which continues to grow strongly).

Of late, Uber has been doubling down on its membership program, further enhancing its differentiation. In Q2’24 Uber introduced Uber One for Students at a price of $4.99 per month across the US. And in Q4’25 Uber began offering Family Sharing, “allowing members to extend benefits to another adult and their teens at no additional cost.”

Further in Apr-26, Uber announced that users could now book hotels/Vrbo’s on Uber (through a partnership with Expedia). Uber One members will get at least a 20% discount on a rolling list of 10,000 hotels, as well as an additional 10% back in Uber credits on every booking. Uber also hinted that Uber One members will soon have the opportunity to benefit from discounts on parking through SpotHero, after acquiring the business in Feb-26. In grocery, Uber announced a “no fees above $60” benefit for Uber One members on eligible orders in the US.

More benefits will likely come. In theory, as the subscriber base grows, it should be easier for Uber to strike partnerships, spinning the flywheel faster. There’s no doubt Uber has circled flights/air travel as a key potential opportunity, amongst other categories.

Here’s a look at Uber One membership vs. other large consumer membership offerings:

This chart, as well as Uber One’s growth rate, would seem to imply that there is still significant room for growth. Uber One looks well on its way to surpassing Costco membership given its 50%+ growth rate, by far the highest of these comps (other than DoorDash). Uber added ~20mm members in 2025 alone. Uber also stands to add perhaps 10-20mm new members from Delivery Hero’s base of roughly ~49mm MAPCs.

Are all these membership giveaways actually a good deal for Uber? you may be asking. Well, according to Uber — yes. While the membership program lowers a member’s profitability initially, over time this is made up for through increased stickiness and greater spend (i.e. customer lifetime value increases).6

AVs Should Expand Uber’s Mobility TAM Significantly

According to the FHA, US drivers travel 3tn miles annually (excluding large trucks, which represent ~10% of miles driven).

Today, the rideshare companies (Uber & Lyft) represent <1% of all of these miles traveled:

The cost of rideshare, however, is set to fall significantly (1) once the driver is removed from the car, and (2) the cost of AV hardware drops sufficiently. It’s expected that the cost of rideshare will drop below the cost of personal car ownership sometime over the next decade.

Here is a forecast from Goldman Sachs from Jul-25:

To date, in fact, I think it’s fair to say costs have fallen more quickly than expected. Waymo’s upfitted Jaguar I-Pace is rumored to have cost ~$150k per vehicle, while its newer Ojai/Zeekr might be less than half that, in the $60-75k range. Tesla claims its Robotaxi will have a starting price in the $30k range. Baidu’s RT6 robotaxi has had an all-in cost of less than $30k since 2024.7

A fully functional L4 AV for less than $30k would already put rideshare around price parity with personally-owned vehicles in many developed markets (discussed later).

Innovation around vehicle form factor will likely take another big chunk of costs out of the system. Cars today are built to handle extreme use cases. I personally own a car that seats 7, but the vast majority of our trips only transport 1-3 people. Many EVs come with ranges of 400+ miles, but our average trip is just a few miles around the neighborhood.8 Different form factors will emerge to handle different types of trips. For example, if you are traveling solo for a few miles in a residential area, in the future you’ll likely get matched with a super cheap 2-seater (that can’t even go on highways).9

Assuming 3tn miles traveled, here is how large the US rideshare market could be under a range of price per mile and market share scenarios.

If say the cost of rideshare dropped to $1.50, and rideshare as a % of miles traveled hit 15%, the total market size in the US would be $675bn, as an example.

Note that more total miles traveled could also result from autonomy. People could likely stand living a bit farther from work if they don’t have to drive. Instead of flying from SF to LA, a car could pick you and the family up at your front door and drive you straight to your destination while you nap, watch a show, work, etc. No airports, no hauling around luggage, no dealing with cars to/from the airport, and a door-to-door travel time that would be comparable. You could even book last minute for no extra cost. Further, at some point eVTOLs will be a thing, and likely bookable on Uber, which could be another source of miles on the platform.

Below are some potential growth scenarios for Uber in US mobility, assuming different rideshare growth scenarios and Uber market share scenarios. If we assume the market grows a bit (5%) to $3.15bn, a price per mile of $1.50, that AV penetration hits 12.5%, and that Uber’s mkt share falls to 50%, Uber’s US mobility business would ~7.6x (a 14% CAGR if we assume this takes 15 years).

This is a highly “back-of-envelope” analysis, but it gives us an idea of the potential that AVs could unlock for Uber. Much of the investment thesis for Uber depends on where you land on (1) overall robotaxi market share and (2) what piece of that Uber can retain.

Uber, as you can imagine, has spun this as a $1tn opportunity in US mobility alone:

It’s possible they are right, but to hit $1tn in US mobility, it will likely require 25%+ of all miles traveled to be rideshare.

Internationally, Uber Eats is DoorDash

I actually began my work in this space looking at DoorDash. I was a loyal DoorDash user and had seen a few compelling theses on the business. In doing my research, however, one thing that surprised me was that DoorDash seemed to be struggling taking market share internationally. DoorDash is considered by many to be the best operator in the space, and that dovetailed with my experience as a user. I enjoyed the product, and my feeling from past experience was that Uber Eats was not nearly as good (more expensive, less selection, longer ETAs, no “DoubleDash”). Further, as someone who lives in the suburbs, I’m not a heavy rideshare user. Thus, I wasn’t necessarily a prime target for Uber One.

To my surprise, however, internationally Uber Eats is DoorDash (i.e. Uber Eats is the dominant provider in many of the largest non-US markets). This dynamic is only further enhanced by Uber’s acquisition of Delivery Hero. See below my highly unofficial (i.e. pieced together through various one-off datapoints) table of the leading food delivery players by geo.

As we can see, Uber Eats leads in Japan, the UK, France, Canada, Australia, Mexico, Turkey and Taiwan. By way of the Delivery Hero deal, Uber Eats also now leads in South Korea, Saudi Arabia, Argentina, Egypt, the UAE and more. Amongst the major, non-US markets, in fact, DoorDash only leads in Italy, and only by way of its recent acquisition of Deliveroo.

As a result of its higher market share, Uber Eats should naturally have an edge when it comes to generating lower ETAs, better prices, and more selection in these countries. Further, Uber provides mobility in every one of its top Uber Eats’ countries (other than the Philippines), tying the two offerings together through Uber One. Based on this, it makes a lot of sense that DoorDash was struggling taking share. DoorDash is structurally disadvantaged in two ways (less market share + no mobility offering) in almost all international markets where it competes with Uber.

A telling development occurred in Feb-26 when Uber announced it was launching delivery efforts in seven European markets, including Finland, Denmark and Norway. This was surprising because these are some of Wolt’s (the European player that DoorDash acquired in Nov-21 at a massive GMV multiple) oldest and strongest markets. Wolt was actually founded in Finland and is still HQ’d in Helsinki. Finland is not a large market either. So why would Uber seemingly waste its time/money attacking these Wolt strongholds? Uber is not of the mindset of launching into markets in which it can’t compete (over the years it has pulled out of China, Russia, and SE Asia completely and India, Brazil, Italy, and Israel in food delivery). Well, Uber must feel that it has a real shot in these markets, which in and of itself I think is a meaningful datapoint.

DoorDash also recently announced it was exiting Japan. DoorDash launched in Japan in 2021 and Wolt was actually there before DoorDash, planting a flag in 2020. Neither DoorDash nor Wolt were able to ultimately make inroads while Uber (there since 2016) has continued to grow and now accounts for ~60% of the market.

Strong Potential for AI-Driven Efficiency Gains

Uber’s top three most important capabilities might arguably be: (1) its pricing/matching/prediction engines that sit at the core of its marketplaces, (2) developing great apps (which includes recommendation engines, advertising) for both the consumer as well as the driver/courier, and (3) customer/driver/merchant support.

All three of these areas are primed for AI-driven enhancement. And based on quotes from management, Uber is likely only scratching the surface at this point re AI actually impacting its financial performance. Here is commentary from Dara on the opportunity in customer service10, the opportunity in software development11, and both (from a Mar-25 Morgan Stanley conference):

“So I think every company in the world is going to tell you that they’re using AI and it’s awesome and they’re going to save lots of money. Actually getting AI to work in an enterprise at scale, I’m finding, is really, really hard. It’s very easy to have little — call it, little optimizations that increase some efficiency about 5%, but we want much more than that, right? We want 20%, 30%, 40% increases in efficiency.

And the good news for us is like I’m finally seeing real signs of it. The scale areas where you’re going to see material differences in terms of AI and these larger models and having big impact on our business are with customer service. Now it’s not just easy. You don’t take these models and train them on a bunch of customer service interactions and expect results. It’s smaller parts of the customer service, your history as a customer, are you a good customer, a bad customer, your claim, you said that your food was getting in late. Did it actually get in late? What’s the policy? What’s the communication, et cetera? We are kind of — we have AI models optimizing each of those. Maybe we’ll have one giant end-to-end model at some point, but we don’t [today]. That activity is coming together, which is going to translate into hundreds of millions of dollars of customer service savings, at, we think, better outcomes.

Same thing as it relates to developer productivity. We’ve kind of had, I’d say, level 1 of dev productivity. Earlier uses of products that got us, call it, 10% to 15% improvements in productivity. We’re now hitting the next level. And it takes — it’s hard to get to these next levels, but I think that, over the next 2 to 3 years, you will see — we expect to see benefits in the hundreds of millions of dollars with better outcomes, which is what we’re demanding of our teams.

And I would tell you that it is — it has taken multiple optimizations for us to get there. It is not easy, but it’s absolutely worth the effort that we’re putting into it. And I’m a lot more optimistic today than I was 6 months ago. Like there’s — I’m seeing breakthroughs, and obviously, with the amount of capital coming into the marketplace, the ecosystem is on our side.”

Here is Dara on the opportunity in better recommendation engines, search and discovery (from a Dec-25 episode of On with Kara Swisher):

“But we have long been driven by AI in all things pricing, matching, etc.…But the models, the ability to just build larger models now, right? So for example, if you’re on Uber Eats, you have a ‘sort’ that’s unique to you based on your past behavior and just patterns that we know about you — what your favorite restaurants are, etc. And we don’t want to just serve you those favorite restaurants. We also want to help you explore new places, etc. These models, as they get larger and as we get better silicon are just getting more and more powerful and capable…If you’re shopping with us and you choose oat milk, what’s the next thing that we show you? These larger models are enormously more effective than the last generation of models.”

Steve Jang, an early investor in Uber, in a Nov-24 appearance on Yahoo Finance, discusses the opportunity in the company’s core prediction engines:

“Uber has historically been one of the largest real-time real world applications of AI/ML in the world. The marketplace engine makes 10 million predictions per second on behalf of its drivers and riders and delivery couriers. At its core, it is a giant, global prediction engine orchestrating all this movement. Now, have they tapped into generative AI and all the cost efficiencies and all the productivity benefits there? Not yet. It’s a large scale operation. They’re starting to make those transitions. And when they transform, I think that the cost and the software-like scalability of the company will really come out. You’re seeing today a 40% gross margin with a 10% operating profit margin. That’s starting to look — that’s transitioning into a software company. And so I think with generative AI and all the work that they’re putting in right now to transform their internal operations and their network across the world, I think that’s the second or third area of opportunity for them.”

Uber seems well positioned to capture any AI-driven efficiency gains given it will still enjoy strong competitive moats across both mobility/delivery for many years to come (i.e. gains won’t be competed away).

Uber’s Ad Business is a Monster

Uber Eats is a highly attractive and impactful place to advertise for merchants. Uber Eats generates a ton of traffic and sits right up against the “point of purchase,” where more and more ad spend is flowing (vs. say “top-of-funnel” awareness).12 Uber also serves an attractive demographic — the relatively wealthy and the relatively young.

The performance of Uber’s ad business reflects the attractiveness of the offering. In Q4’25, Uber generated “well over” $2bn of ARR from its ads business, up “over 50%” year over year.

While Uber doesn’t disclose the profitability of this segment, at least one peer (Delivery Hero, actually) has achieved a ~70% EBITDA margin in its respective ads segment (see yellow box below, from the company’s Sep-25 investor presentation).

I haven’t seen any similar disclosures from other delivery peers.

There are good reasons to believe that if Delivery Hero can generate this type of margin profile, that Uber should be able to as well. Uber has: (1) greater scale, (2) a richer ad targeting toolset (e.g. target certain dayparts, keywords, only new customers, etc.) which helps attract, importantly, enterprise-scale advertisers, and (3) more advertisers (if not now, eventually), which will result in richer auctions and higher CPCs.

If we assume the same margin profile as Delivery Hero, Uber would be generating $1.4bn of EBITDA from the segment, or ~16% of Uber’s total EBITDA.

While this business represents a little over 2% of delivery gross bookings ex Delivery Hero (the target Uber originally set a few years ago for its ad business), management recently stated that this business could, in fact, be much larger as a percentage of bookings (from Q4’25):

“So on ads, we’re very, very pleased with the momentum that we are seeing. As you rightly pointed out, we had many years ago talked about 2% as the potential ceiling for penetration with delivery advertising. What we are seeing is that the opportunity size here is potentially much larger.

And as we think through where we are on the journey with enterprises versus SMBs, SMB ad penetration is a lot higher than 2% and enterprise year-on-year growth is now outpacing SMBs by a lot more. So in a way, enterprise advertising is playing catch-up, and that means there’s going to be a lot of runway here. At the same time, our products on grocery and retail and mobility are a lot more nascent, and there’s going to be opportunity for us to grow there as well.”

The fact that the segment is growing 50%+ would also suggest that there is still much room to grow (and that 2% is not the ceiling after all).

Many of Uber’s delivery peers, in fact, do well north of 2%:

Delivery Hero actually set a minimum long-term target of 4% (see slide above).

Further, it seems we are still relatively early in the ad monetization journey for food delivery. Here is the CEO of talabat making this point (granted this is from Nov-23, or ~2.5 years ago):

Tomaso Rodriguez: “We have a lot of advertising products that you could buy on talabat if you want to grow your business on talabat. The great thing about our advertising product is that we look at ROI for the restaurant that is immediate…We have a cost per click product like Google has, right?…

So those products, we price them in a way that you make your money back immediately. Two or three times the money you make, so that’s the target we have. And for me I mean as a vendor — so if I had a restaurant with something that gives me back two or three times the money that I spend today, the same month, for me it’s a no-brainer. Because the alternative is spending that money on Instagram, right? And maybe I acquire a customer that will come to me and I will get the money back in one year or two years…

And you know the cost of a click versus the order that you get from the customer is is very low. So I would argue that doing advertising on talabat or on in general on aggregator platforms is the absolute most efficient marketing spend that a vendor, a restaurant, a grocery store or whatever could have.”

Interviewer: “So does the cost per click go up for example if there’s more demand in a specific area?”

TR: “No, not today. Today it’s, I would say, because we’re just getting started, and probably our tech is also not so evolved. And so it’s a fixed fee. It depends on the time of the day, whatever. But at the end of the day, it’s also — when Google was started, when Facebook was started — like the first ones that started doing advertising [on these platforms] built huge businesses out of it because algorithms were not so sophisticated and whatever. So I think when it comes to food delivery, we’re a bit at that stage. So it’s a great time for a vendor to invest in advertising right now.”

In other words, CPCs should continue to rise as the industry matures and auctions become more robust. This should provide a meaningful continued tailwind for Uber’s ad business.

If Uber can grow ad revenue to 4% of delivery bookings, and if the overall delivery business doubles (including DH), you’d be looking at ~$10bn of ad revenue and ~$7bn of ad EBITDA (assuming a 70% margin). That would represent ~$5bn of incremental EBITDA for a business that will generate ~$16bn of EBITDA in 2027. This seems doable over, say, the next 4-5 years.

Interestingly, the ads business increasingly doubles as a barrier to entry in food delivery: (1) you can only turn it on once you have massive scale, and (2) it allows you to run the rest of the delivery business at a very thin margin.13

Uber’s AV Future: Everywhere all at Once?

It may not appear like it today, but Uber seems on the cusp of bringing AVs to let’s say 30+ major cities across the globe in relatively quick succession. It’s an “everywhere all at once” strategy, employing various AV tech providers, various car OEMs, and various fleet operators (and potentially various fleet financiers) all “at once.”

Let’s look at the partnerships that Uber has lined up in already specified cities across the US:

Note the charts above/below blatantly copy the formatting of these charts from Jackson Lester of AV Map.

We can see that Uber has partnerships announced (or live in some cases) across eight US cities. This includes three partners each for the SF Bay Area (Nuro, NVIDIA and Rivian) and Los Angeles (VW/MOIA, Zoox and NVIDIA).

Uber has even more partnerships lined up internationally:

Uber has another nine international cities announced (or live in some cases) with launch dates all anticipated before the end of 2026. Four of these cities have multiple AV partners lined up, and one city — Dubai — actually has three (WeRide, Apollo Go and Pony.ai).

Uber has its sights on many more cities than just these ~17 combined, however. They clearly intend to expand their Nuro partnership well beyond just the SF Bay Area and Houston, given they’ve committed to purchase 35,000 vehicles from Nuro/Lucid over the next ~5 years. Uber has committed to at least 10,000 AVs from Rivian by 2028. Further, Uber announced with NVIDIA an intention to be in “28 cities by 2028” with “full-stack” NVIDIA-powered AVs.

Below is commentary from Ali Kani, VP of Automotive at NVIDIA, from a Mar-26 Turing Post podcast. He states that NVIDIA has fleets in 20+ cities today gathering data and laying the groundwork for AV introduction. Once NVIDIA’s self-driving software is ready (targeting initial rollouts in H1’27), NVIDIA anticipates going live across the remaining cities quickly thereafter.

“We’re sending a fleet, like, everywhere right now. So what we did was we identified 20 cities which are relatively unique. So it’s technical, it’s like what are the most unique cities? And then we also included where are most of the customers going to be? Because that’s another area where it’s like, ‘Hey, we care about this. Don’t just think it’s the same and let’s skip that town.’ Then we send a fleet of cars there to test and also get data. And then based on that — the model is very generalist. So it’s pretty good but we make sure that it’s as good everywhere. And then we’ll slowly release it as our metrics tell us that we’re ready, but it will be kind of nationwide.”

In all, Uber has laid out a target of being the largest facilitator of AV trips by 2029. From its Q4’25 “AV Spotlight” and prepared remarks:

“By the end of 2026, we expect to be facilitating AV trips in as many as 15 cities globally, with a roughly even split of U.S. and international cities. And by 2029, we intend to be the largest facilitator of AV trips in the world. AVs will change how trips are supplied, but not how demand is aggregated. History suggests that over time as supply fragments and technology commoditizes, the platform that can bring the highest utilization to assets, and superior reliability to customers, will capture a large share of value. That is the role Uber is set up to play.”

Even if Uber “misses” this target, it should be well on its way to a significant amount of AV supply live on the platform by 2029.

Overall, I think an underrated aspect of the Uber story is that, in certain ways, Uber is well ahead of the pack in terms of AVs. Yes, Uber doesn’t have the cars and the technology yet (though this is coming). However, in the meantime, they are laying the groundwork on much of “the rest” that will be needed for AVs. And “the rest” is really important and a really significant lift.

AVs need to be maintained, cleaned, charged, parked over night/during off-peak hours, and insured. They potentially could use help being financed. AVs need to be positioned well in a city to maximize earnings. The operator needs to know when an AV is about to run out of battery, or when the air pressure in its tires are too low. AVs will need on-road support (i.e. an asset may need to be recovered in the field, etc.). Lost items will need to be managed. The AVs need to be integrated into a hybrid platform that understands when to dispatch an AV and when not to (is a pick-up/drop-off within the ODD, is a human needed to enter a gate code to a private community). In-vehicle experience needs to be managed (e.g. access to sound, temperature, rider support).

All of these tasks that once fell mostly to the driver, will need to be subsumed by other players in the ecosystem. Uber’s answer to this is (1) its Autonomous Solutions offering (discussed later on), introduced in Feb-26. It’s a diverse set of capabilities that take on much of the burden of actually operating a commercial rideshare service (and that no other players can match today). (2) To handle fleet operations Uber will leverage the relationships it’s developed with just about every rideshare fleet operator in existence, from the very small to the very large. These are the same fleet operators that will be servicing (and in some cases buying) many of the AVs that will go onto the Uber marketplace. (3) Lastly Uber is prepared to step in to develop charging infrastructure and depots where its partners may not have the resources or desire.

Here is management from a Mar-26 Morgan Stanley conference, commenting on how Uber is engaged in leasing conversations for AV depots across 50 (50?!) different sites related to just their 2027 deployments:

“From our perspective, when we think through this deployment curve over the next few years, once we start getting conviction that the software is going to hit the mark and it’s going to get to that L4 deployment with superhuman safety, our efforts start shifting to a couple of other priorities.

The first one is securing OEM supply, right? We need to ensure that any of our partners who can deliver that kind of software have enough scale volumes coming to them, and then that volume can get deployed on Uber’s network.

And then the other thing we have to do is to ensure that we are bringing AV infrastructure on the ground. And for instance, for our 2027 deployments, we’re already scouting sites across our footprint. We already have about 50 different sites that we are beginning to get into leasing conversations, and those deployments will start once we have the software there.”

Of course we haven’t even mentioned the fact that Uber already has a 1P app with tons of aggregated demand. Everyone else (Waymo, Tesla) still need to get customers on their app. They need to provide customer support. They need to handle payments/fraud. They need different “products” like “reserve,” “shared routes,” XL, etc. Uber, of course, has had these things dialed in for awhile now, and these aren’t trivial things.

So while Uber may seem behind on autonomy today, I believe they’re actually well ahead in many important ways. Uber is laying the groundwork to scale AVs. They are just waiting on the tech and the cars.

Attractive Valuation

If we adjust for the impact of the Delivery Hero acquisition, Uber trades for just ~10x ‘27e non-GAAP EBITDA and ~12x ‘27e non-GAAP EBIT (burdened for SBC).

Others

• Delivery + Mobility Synergies. Having both delivery and mobility offerings creates synergies for Uber beyond simply the Uber One membership.

  • Lead Gen. The Uber app is the largest source of new users for Uber Eats — larger than Google, Instagram and Facebook, combined.14 About 30% of Eats’ first time orders come from Rides.15 Only ~1 in 5 Uber MAPCs are active across both segments today, a clear opportunity for Uber (they are highly focused on pushing this north of 50%).16 Users that are active across both businesses spend roughly 3x those that aren’t.17

  • Driver Recruitment. Uber Eats couriers are easier to pull into the Uber ecosystem than Uber mobility drivers (mobility drivers require background checks, car inspections, etc. while couriers do not). Once couriers are in the Uber ecosystem, however, they are a much easier “cross sell” for Uber mobility. Thus, Uber Eats is an important source of new driver recruits for Uber mobility.18

  • Labor utilization. Uber should be able to keep its drivers/couriers more utilized as a result of its multiple offerings. Better utilized labor should result in lower prices, all things equal. This might be especially critical in suburban mobility. The suburbs would normally be very expensive to cover from a mobility standpoint due to a lack of density. However, because Uber drivers can stay occupied on the delivery side, it allows Uber to cover the occasional mobility ride (and do it at a reasonable price).
    
    The same concept might apply in the AV world, where having both offerings could drive better utilization of AVs (from Q4’25):
    
    “If you really think about the long, long game, one of the key elements is going to be what companies are able to utilize these cars in the troughs where they will largely not be busy. And having delivery and freight as part of our logistics ecosystem gives us an opportunity to actually use these vehicles at a structurally higher utilization than anyone else. We have the network utilization. We’ve already demonstrated that our network is driving higher utilization even in a circumstance where supply is really low. As supply increases, the utilization advantage that we have, both on the mobility-only network, but then for delivery, for freight, for last mile delivery is going to be super interesting. And it’s a structural advantage that we have that’s not available to any other player.”

• Strong Management. I’m a believer in Uber management (discussed later on). They are “doing the basics” well: (1) keeping take rates low, (2) highly focused on building marketplace “supply,” (3) building an attractive Uber One membership program, and (4) squeezing more and more efficiency/profitability out of the marketplace. I think they are doing satisfactory work re capital allocation as well.

• Hybrid AV/Human “Workforce.” Of course Uber has something else that Waymo, Tesla, and Amazon don’t — human drivers/couriers. This, for a period of time, will allow Uber to both meet peak demand (by flexing its “supply” base up with human drivers) while also keeping AVs maximally utilized (they can be reserved for “baseload” supply). If your platform is AV-only, like Waymo’s, you will have to make a trade-off in one direction or another (better coverage vs. higher utilization).

Investment Risks

Waymo

Waymo is doing just fine. It’s the only group in the US with a real commercial robotaxi service. They are live in ~15 cities, with another 15+ lined up for near-term expansion. In the markets in which Waymo has launched 1P, it’s clearly taking share. It’s fresh off raising $16bn at a $126bn valuation from a laundry list of top investors including Sequoia, Dragoneer, a16z, Silver Lake, Kleiner Perkins, etc. They just launched the Ojai vehicle, their first vehicle purpose-built for autonomy (i.e. not a full retrofit).

As we are seeing with Waymo, when you remove the driver from the car, the market opens up to additional players. Waymo doesn’t need to recruit drivers to a platform with no demand on it (a money incinerating proposition), because they can just seed the platform with many thousands of robotaxis — no human drivers necessary.

And users generally love the Waymo “product.” It’s safer than a human driver. Most riders prefer to be alone in a car. The cars are brand new and clean, unlike many Ubers. It’s also a futuristic experience. While ETAs are generally higher, coverage is spottier and pricing is higher, Waymo has taken material share relatively quickly. Here is a look at SF rideshare market share since Waymo’s launch in Aug-23 through Apr-25:

Here is additional market share data across a broader set of Waymo markets. Over time Waymo has climbed into the 15-20% range, for now, in Phoenix, SF and LA (its most mature markets).

And this is with Waymo choosing to be “constrained” on the supply-side. They are not flooding any given market with AVs, but rather seem to be aiming for more “baseload” capacity. Peak-to-trough demand over the course of a day is ~4x, so if Waymo were to put enough AVs in a market to handle peak demand, they’d be stuck with many of their cars sitting idle during troughs.

What is worrisome about Waymo for Uber:

• Leading AV Tech. Waymo is the undisputed leader in AV tech. They have real, commercial, driverless deployments all across the country. Where I live — the SF Bay Area — they are ubiquitous (even in the suburbs). With more miles comes more data and more edge cases “solved.” Even when new AV offerings come into the fold, for those especially concerned with safety, the Waymo “Driver” will likely continue to be the safest option for at least the next few years. It’s hard to say exactly how far Waymo is ahead of the pack, and when other providers will have robust enough solutions to safely operate large, flexible ODDs.

• Physical World Build-Out. First out of the gate with robotaxis, Waymo is securing depots/charging infrastructure as well as partnering with leading fleet operators all across the US (and soon internationally). While Uber can and will do the same, it’s got to be a little depressing to see Waymo settling in like this for the long haul. Within the next couple years, Waymo will likely have infra up and running across 30+ major cities.

• Certain Cost Advantages. I’d assume Waymo can train its models at below market rates on Google hardware (Waymo’s in-car chips are NVIDIA GPUs, for now). In its 1P markets where Waymo goes direct to consumer, Waymo pays no “Uber tax.” Also Waymo has its Waymo “Driver,” of course. Thus, it does not have to pay a AV software “licensing fee” (vs. Uber which has to pay this indirectly, like in the case of its Nuro/Lucid robotaxis). These are cost advantages for Waymo.
  
  However, of course Waymo also needs to invest considerable sums to develop its “Driver” as well as its “1P” service. Thus these cost advantages only really kick in at massive scale. Further, if we take NVIDIA’s Alpamayo-powered vehicles (where the software is open sourced), Uber can also enjoy a cut-rate fee for AV software.

• Google-Backed. Waymo is backed by Google, a monster of a company with a market cap of ~$4tn. Google can afford to be loss-making in its AV/rideshare efforts longer than just about anyone.

What is not so worrisome about Waymo for Uber:

• Structural ETA Issues. Waymo can’t really build enough AVs to handle peak traffic hours — for now. Peak-to-trough is ~4x over the course of a day (and the size of a peak also varies over the course of a year in any given market). If Waymo had enough cars to handle peaks, then they’d have a lot of cars sitting idle during troughs. You can see them running into this issue already today. Their wait times in California have flatlined at around 18 minutes, according to the Driverless Digest:

  

  Wait times for Waymo are not nearly this bad in other geos, it should be noted. However, it seems it will likely continue to be challenging for Waymo to offer low ETAs during peak hours. Many will not want to have to bother with switching between apps (Uber for peak hours, Waymo for off-peak), especially when Uber eventually has sufficient AV supply (and where presumably rates are comparable).

• AV Alternatives Are Coming. As alluded to in the intro, a number of other AV providers are going to launch offerings in the coming years. Waymo being the sole provider of self-driving likely isn’t going to last much longer. More on this later.

• No Membership / Lock-In. Waymo doesn’t have a “comprehensive” membership program like Uber One. For many, it will give Uber the edge, especially once Uber has sufficient AV supply.

• Still Limited Geo Coverage. When you leave your home city, or home country, Waymo might not have you covered. This will be the case for awhile. While not necessarily the end of the world, this will be a headwind for a membership product, a business product, etc. It will also make it a little harder to build stickiness with some users, when these users can’t just open one app all the time, no matter what. Around 40% of US-based Uber mobility users took rides outside their home cities in 2025 (and in total, ~1.5bn trips were taken outside of home cities, or ~10% of total trips).

• Lack of “Products.” Waymo doesn’t yet do shared routes, “wait & save,” “reserve,” elite, XXL, pets, car seat, etc. They don’t have “Waymo for business.” Uber has you covered no matter what you need.

• Coverage of Suburbs. Uber management has stated a couple times that they believe that AVs will have a difficult time serving the suburbs for some time (though I haven’t seen them explain exactly why).19 My assumption is simply that utilization will be that much lower, and deadheading trips that much longer on average. Uber doesn’t have to absorb the costs of idle drivers (they only pay per trip). However, Uber does need to compensate drivers for this lack of utilization with, I assume, higher rates, which provides air cover for Waymo to also charge more. It could be the case that Uber drivers being able to toggle between driver/courier mode in the suburbs is a big advantage when it comes to offering mobility in the suburbs. Uber can keep its couriers utilized making deliveries in the suburbs, and when the occasional ride is requested, Uber will have the coverage (and can provide the ride at a low enough cost to make it interesting for the rider). I’ve never seen Uber confirm this advantage, however — if you’ve seen them do so, please let me know in the comments.

• Fleet Financing. Waymo owns all of its own fleet today. It may have a harder time bringing in financial partners to own cars (and thereby offload some of its capital intensity). Partners want high returns and low risk. If you have a less liquid marketplace (i.e. lower revenue generation per vehicle), and one that is less established (and therefore more risky), it may prove to be harder to lure in financial partners. There’s no guarantee Uber won’t have similar issues, but Uber seems much better positioned on this front.

I do wonder what differentiation Waymo brings in a world where Uber has sufficient AV supply. Waymo’s advantage right now is its lack of drivers, cleaner/newer cars, and a futuristic experience, from what I can tell. Do these not all go away in a world where Uber has plenty of AV supply, say by 2028/2029 (and the user can set an AV preference)? Further, it seems reasonable to assume that Uber should still have lower costs, better coverage and lower wait times in all the intervening years between now and then. If I’m a resident of SF (or LA, Las Vegas, etc.) in 2027, when Uber has 3+ AV offerings on its platform, I’m not sure why I’m using Waymo over Uber (other than Waymo perhaps subsidizing fares).

Maybe Waymo sees a path to being the low cost operator over time. Maybe they are willing to lose money (and subsidize rides) over many, many years in order to build the type of scale that could achieve this. The Google backing makes this perhaps a viable option.

Unfortunately, it’s hard to know at this point how Uber’s competition with Waymo evolves. Waymo could keep investing with the intention of building a rideshare network that rivals Uber in its geographic coverage and density. Waymo could also decide that, if it’s not gaining enough ground quickly enough, that it is better off simply licensing the Waymo Driver to car OEMs (as well as licensing “driver” offerings for other applications, like delivery).

Tesla

Tesla poses a risk that, in some ways, is even scarier than Waymo.

Tesla at one point was staking much of the future of the company (a ~$1.5tn company by market cap) on self-driving and the Cybercab. It has been on the cusp of autonomy for years, seemingly. Tesla is dangling the idea of a $30k Cybercab, which would undercut Uber (and Waymo) significantly on price (and of which Tesla could manufacture 2mm annually, they claim). Tesla has a huge number of vehicles on the roads today (~9mm) that they claim could be instantly flipped into Tesla’s Robotaxi fleet.

Furthermore, Tesla has a clear desire to go 1P. It has also signaled a likely aggressive approach to pricing, charging far less than Uber and Waymo in its Robotaxi pilots in SF and Austin. Tesla claims to have driven 1.7mm paid Robotaxi miles (across SF and Austin) as of Q1’26:

In Apr-26, Tesla announced the expansion of its Robotaxi pilot program into Dallas and Houston. In Jul-26, they added the Miami market (giving Tesla 5 markets in total).

Perhaps there is a world where Tesla can “flip a switch” on its Robotaxi initiative more broadly once the tech is ready. In such a world, Uber would suddenly have to deal with aggressive competition from a very large and fully integrated car OEM with structural pricing advantages.

What is worrisome about Tesla for Uber:

• Cost Advantages. Tesla seems to have the lowest cost hardware “bill of materials” in the US for robotaxis. It’s Cybercab is purpose built for autonomy. It has no steering wheel, pedals, or rearview mirrors. It’s “sensor suite” is the lowest cost in the industry (no LiDAR, no radar, just eight cameras). It employs custom-built Tesla silicon for its in-car compute needs. It’s manufactured in a factory — Giga Texas — that seems truly innovative.
  
  However, Tesla is far from the most scaled car OEM. It has respectable sales in the US with around the #9th and #17th best selling cars (and if looking just at EVs, the top two models by far).

  

  But this is not exactly a dominant position. Further Tesla vehicle deliveries actually fell by 9% YoY in 2025.
  
  In terms of the top selling car brands in the world by units, Tesla does not rank in the top 15:

  

  If we look out a few years, Toyota and BYD seem like they will be the clear leaders (BYD grew units sold by 31% in H1’25). It doesn’t strike me like there’s a path for Tesla to build cars at a materially lower cost than these top players longer term, given the scale differential (even considering some of the innovation Tesla is bringing).
  
  In fact, Tesla is already, by far, not the low cost EV producer globally — that crown belongs to the Chinese manufacturers, most notably BYD and Geely. BYD’s latest Seagull EV comes equipped with LiDAR and a “mid-tier” smart driving system (aka Level 2+/2++) for just ~$13,400. Baidu’s RT6 Robotaxi (manufactured by JMC) was introduced way back in May-24 at a sub $30k price tag. While Chinese cars are effectively locked out of the US market (owing to massive ~100% tariffs), they are sold (and taking share) in most other markets (including Europe, Latin America and SE Asia). Further, while Chinese vehicles aren’t sold in the US, Waymo’s Ojai vehicle, which it recently introduced in the US, is built on a Chinese vehicle platform manufactured by Geely. Thus, as a robotaxi operator, Uber might still have a chance at using Chinese-built AVs in the US.
  
  Here is Uber management from a Mar-26 Morgan Stanley conference on Uber’s partnerships with Chinese AV providers in international markets
  
  ”So what we’re seeing in those international markets right now, because we have 3 distinct partners who are already capable of deploying AVs on our network, which is Baidu, WeRide and Pony, we are beginning to see the scenarios where you start seeing a variety of supply coming. So in Abu Dhabi, for instance, we already have WeRide with a driver-out solution. In Dubai, we’ll likely have players like both WeRide and Baidu going on our network and bringing their services to the market. And in that kind of a world, you don’t have a lot of conflict where your partners may be thinking about, ‘Should I go 1P? Should I go 3P?’ They’re trying to get to speed to market and the fastest way to get that kind of an outcome is to work with Uber and also work through all of these solutions we bring, both in terms of our fleet partnerships as well as the Uber Autonomous Solutions kind of stack that you just saw the video right before our chat. We are beginning to see that kind of a theme play out there.

  
  From a cost standpoint, the Chinese partners we’re working with are at a price point for their hardware and software, which is better than anything we are seeing anywhere. So they will certainly have an advantage there. And the fact that they don’t have philosophical debates about whether they want to do 1P versus 3P puts them in a place where they’re maximizing revenues against the lowest cost structure, and that just gives them a leg up in how quickly they can ramp up.”
  
  So while Tesla has a leg-up in the US on hardware costs, it doesn’t seem insurmountable over the fullness of time. Furthermore, in virtually all non-US markets, Tesla is far from the lowest cost producer.
  

• “Price War” Strategy Seems Likely? Tesla will probably be acquired by SpaceX at some point, giving it more cash and a longer runway to be loss-making. It will probably have a real Robotaxi service at some point. And at that point, Tesla will probably try some sort of “scorched earth” pricing (in at least some markets for at least some period of time).

• Data Advantage. Tesla has a ton of real-world driving data owing to their large global fleet of ~9mm vehicles operating all over the world. Surely this is a huge advantage for Tesla? Well, according to Ali Kani, VP of Automotive at NVIDIA, data is not a problem for them. Here is commentary from a Mar-26 Turing Post podcast:
  
  Interviewer: “Do you think you’re close to the amount of data Tesla has for self-driving?”

  AK: “So I think since they’ve been doing this longer than us, they have more real data than us. But we supplement the real data that we’ve gotten with synthetic data. Right now, for where we are, I don’t think we feel that data is our problem. We just generate compute [and it] becomes data, right? So we synthetically generate what we’re missing. But the other thing is we’re now in more and more cars. You can buy this car and we can get data [from it]. Eventually, we will be in every Mercedes-Benz car, every Jaguar Land Rover car. And so, you know, I think we’ll have hundreds of millions of hours of data. And I think that’s very helpful for us on our path to level four. This car, for level two, I think we’re fine. But I think getting incremental data and having the compute to train and test those models together, I think what we feel good about our ability to get this to level four. I think we announced we’re going to start doing trials next year.”
  
  Uber, to its credit, is helping to blow up any remaining chances of a data moat through its AV Labs initiative. Here is Uber CFO Balaji Krishnamurthy in May-26 with something of a status update on the initiative. By the end of 2026, Uber’s AV Labs fleet will be generating 2mm miles worth of data every month (for reference, AV operators have needed at least ~10mm miles of data in total to reach their first public driverless launch):

What is not worrisome about Tesla for Uber:

• Fading AV Tech Leadership. Tesla is falling back into the pack on AV tech. Below is an account of one of Tesla’s first Robotaxi rides in Dallas from Apr-26 (it was also written up by Business Insider):

  The Robotaxi (1) misses its exit, (2) gets on the highway (surely not in its ODD), (3) slows down to 20 MPH on the highway while trying to pull over, (4) rider support attempts to call rider but can’t get through, (5) speeds back up, presumably driven by rider support (unclear), (6) takes the passenger to the wrong destination, (7) gets stuck circling a random hotel until rider support again takes over, (8) has a screen malfunction, telling the rider to “exit safely” as the car is driving 40 MPH, and finally, (9) won’t turn right at a red light, causing rage honking from the car behind it. All of this happened over the course of just one ride! Does this seem like an offering that is ready for primetime?
  
  Reported incident data tells a similar story, which I detail later on. In all, Tesla seems a couple years behind Waymo at this point. They seem behind the Chinese players. They seem in danger of falling behind Nuro. And I think it’s probably only a matter of time before Tesla falls behind NVIDIA.
  

• Camera-Only Sensor Suite. Tesla is the only credible self-driving company of which I’m aware that uses a camera-only sensor suite. Everyone else uses a combination of cameras, radar and LiDAR (which should probably tell you something). Elon’s thinking seems to be that humans only have eyes and can drive, so surely computers equipped with only “eyes” (i.e. cameras) should be able to drive the same or better.
  
  While most agree that a camera-only approach should eventually work, there are two issues with this. One is that we don’t know when it will work. A camera-only approach puts a lot more burden on the self-driving software/algorithms — this may take much longer to sort out. And two, even if the car is eventually able to drive as well as a good human driver, are we not trying for “superhuman” safety? Why would I want to ride in a car that could potentially be much more dangerous than one equipped with LiDAR and radar? Elon is betting that most people won’t care, but he may be miscalculating. I don’t think I’d make Tesla my robotaxi of choice, even if it were a little cheaper, if say there was an order of magnitude greater chance of an accident vs. a Waymo-powered or an NVIDIA-powered robotaxi.
  

• Can Tesla be Great at Operations? It’s hard, if not impossible, for companies to be great at many things at once. Tesla is a great technology and manufacturing company. They have shown less skill, I’d argue, in things like service and operations. For Tesla to operate a robotaxi fleet all over the world while operating a car OEM and building world-class self-driving tech, you’re asking a lot. This is not to mention that Tesla is building robots, installing solar panels on roofs, building its own a fab… One company can only do so much, right? Uber has chosen to be excellent at running its marketplace, aggregating supply/demand, and increasingly operating a fleet of mobility/delivery AVs. This seems tough enough.

• Same 1P Issues as Waymo. If Tesla does try to go 1P, it will have all the same issues that Waymo will have. The same structural ETA issues. The same geographic coverage issues. The same suburbs issue. The same lack of products. The same lack of membership offerings. Etc.

• Who’s Going to Own the Cars? Same issue as above for Waymo. Someone needs to own and finance these cars. Uber’s plan is to eventually bring in 3P owners and financing providers. But for a group like Tesla, it’s going to be harder to line up these parties given the uncertainty around Tesla’s 1P network. If Tesla were forced to own its own fleet, its unclear if they’ll have the resources to pull that off.

If it’s hard to see how things play out with Waymo, it’s even harder with Tesla.

I could see Tesla launching a more robust Robotaxi effort over the next couple years and making life difficult for Uber in many geos. I could also see Tesla’s self-driving efforts lagging for many years before the company eventually pulls the plug on any type of 1P effort. I wouldn’t want to have to bet on the latter scenario, however.

Amazon/Zoox

Perhaps normally, an industry with 3 massive contenders might deter other players from entering. However, that’s not going to be the case with Amazon/Zoox.

What is worrisome about Amazon/Zoox for Uber:

• Also Massive. Amazon, like Google and Tesla, is also enormous. They are the company that invented “your margin is my opportunity.” It’s in their DNA to invest deeply for very extended periods in large markets — that’s what they do.

• Self-Driving Tech is Existential. Amazon has every reason to develop self-driving tech. Even if Amazon completely fails with robotaxis, it still needs it for its retail business. There is no future world where Amazon doesn’t have its own self-driving technology. So even though Zoox is a bit behind today, at some point they will cross the finish line.

• Huge Membership Base. Everyone reading this post likely has a Prime membership. Amazon has ~5x more prime members (~260mm) than Uber has Uber One members (~50mm). Amazon thus has a powerful distribution wedge that Tesla and Waymo do not.

• 1P Desires. Amazon seems to strongly desire a 1P offering. Yes, Zoox has formed partnerships with Uber. In Las Vegas starting this summer and in LA by mid-2027, Zoox vehicles will be available on the Uber app. However, Zoox will still have its own app in these markets. Further, Zoox is not partnered with Uber in SF, Zoox’s most mature market. Zoox is also coming soon to Austin and Miami, and no partnerships with Uber have been announced there.
  
  Here is CEO Aicha Evans from a Oct-24 TechSurge podcast on the importance of owning the customer relationship:
  
  “It takes a lot of capital to deploy this. You, I remember, taught me a very important lesson which is ‘access to the customer.’ Thou who talks to the customer, builds the business, right? And I think what you will see in the future is the conundrum of partnering with Uber or not. At the end of the day, if we’re going to spend billions of dollars building this hard technology with a fairly low barrier of entry to get to the customer ourselves, why should we give that up? Unless it makes sense to partner and I think those questions are still open.”
  
  While Zoox may have to partner with Uber in more markets in order to drive demand, it seems they would do so only begrudgingly. So while the two are partners today, I would not characterize their partnership as “stable.”

• On-Demand Delivery Encroachment. Amazon is more and more a direct competitor of Uber Eats. Amazon is apparently crushing it in groceries.20 Free same-day delivery is increasingly commonplace. Nobody seems better positioned than Amazon to vie for Uber’s turf as the best combined mobility + delivery service (unified by a membership offering).

What is not worrisome about Amazon/Zoox for Uber:

• Trailing AV tech. I think it’s fair to say Zoox is trailing in AV tech today. However, it seems likely Zoox will get to L4 over the next few years.

• Same 1P Issues as Waymo. If Zoox does try to go 1P, it will have many of the same issues that Waymo/Tesla will have. The same structural ETA issues. The same geographic coverage issues. The same suburbs issue. The same lack of products. Etc.

Will AVs Stunt Uber’s Growth over the Next Few Years?

AVs will expand the overall rideshare TAM over time. However, could there be a scenario where AVs take share from human rideshare faster than the overall rideshare market expands? Say if prices don’t fall quickly enough? Such a result could spell rough times for Uber shareholders.

The first set of defense Uber has is that most of its business (I estimate ~90% of EBITDA) is not exposed to AV risk over the next few years.

About 60% of EBITDA today comes from the mobility business (this should fall as the advertising business continues to ramp, but let’s ignore that for now). Of mobility bookings, ~40% is generated within the US21 (but let’s assume 60% of profits come from the US — a geo like India, for example, is still lossmaking). And of US mobility profits, we know ~75% comes from outside the top 20 cities.22

Thus, if over the next few years we assume AVs won’t be able to reach the suburbs, we are talking about only ~9% of profits (60% x 60% x 25%) even being threatened by Waymo/Tesla — a pretty small piece of Uber’s business (and I believe this ~9% will shrink somewhat over the next couple years as the ad biz grows and as Uber further penetrates the suburbs). If we look at this on a revenue or bookings basis, it is even less (closer to ~6%). If AVs do prove able to serve let’s say 50% of the US mobility opportunity (i.e. a greater portion of the suburbs), we’d instead be looking at ~18% of profits being susceptible (vs. 9%).

How is the US mobility segment doing today? Surely growth must be slowing? Actually, no. In fact, the company is very emphatically telling us that the US mobility segment is doing the exact opposite — it’s actually accelerating. From Uber’s Q4’25 prepared remarks in reference to its mobility segment:

“Looking ahead, continued improvements in marketplace health are supporting better per-trip economics. New rider growth reached multi-year highs in 2025 and our new rider cohorts are engaging more than prior cohorts, reinforcing the durability of demand and long-term lifetime value. We expect U.S. trip and gross bookings growth to accelerate further in 2026, on the back of a healthier pricing environment supported by lower insurance costs, strong supply dynamics, and accelerating product innovation. Together, these factors leave us well positioned to deliver another year of both healthy top-line growth and strong margin expansion.”

Uber even underlined the bolded part above just to emphasize its level of conviction in this prediction. In Q1’26, Uber reaffirmed this belief that US mobility would accelerate. So whatever impact AVs may have, it’s not going to show up in 2026, at a minimum.

We also have the curious dynamic that AVs seem to already increase the size of the rideshare TAM within a market, even today when they are not lower cost. Below is commentary from Uber’s Q4’25 prepared remarks, discussing how Uber is experiencing better mobility growth in markets with AVs (even when the AVs are not on the Uber network):

“Reality: AVs are likely to drive incremental growth for the entire category. The history of ridesharing has always been supply-led. Our network benefits from every incremental unit of supply added in a city. As supply increases, customers find more value because rides become more affordable with faster ETAs. This fact alone gives us considerable conviction that AVs (as a new form of supply) will expand—not shrink—our total addressable market. Early data supports this view. In Austin and Atlanta, where hundreds of AVs are operating on the Uber network, our overall (AV and non-AV) trip growth has significantly accelerated. In fact, the Austin and Atlanta AV operating zones are now among our fastest-growing areas in the U.S. Encouragingly, the overall growth in these markets was driven by both an acceleration in new riders trying Uber for the first time and higher frequency among existing riders. Importantly, even as AV penetration has grown in both cities, the number of human drivers and their average earnings per hour are both up YoY because of the hybrid-network approach. At the same time, even in a city like San Francisco, where we don’t yet offer AV trips on Uber, the addition of AV supply to the market has grown the category overall. Uber trips in SF accelerated in 2025, and they are growing faster than the rest of the U.S. We are very bullish about the Bay Area as we make progress toward our own AV deployments there within the next 12 months.”

As mentioned above, even in a market in SF where Uber has no AVs on offer, Uber trips accelerated in 2025. It’s a surprising result, perhaps driven by the fact that if more people are taking rideshare more often, eventually they will need Uber for its lower ETAs and pricing.2324

Further, the only group that’s really attacking Uber at this point is Waymo. And Waymo is still very small. While the company is scaling up quickly, it is doing so off a small base. Waymo is currently doing ~500k rides per week.

Waymo is targeting ~1mm trips per week by the end of 2026, or ~52mm trips on an annualized basis.

By 2030, Bloomberg expects this figure to grow to roughly ~7mm trips per week or roughly ~360mm trips annually.

In 2025, however, Uber did 13.5bn trips between its mobility and delivery businesses. If we assume 50% of these trips were for mobility, and 33% of these trips happened in the US, then Uber did ~2.2bn US mobility trips in 2025. If we assume this figure grows by ~12% annually, then in 2030 Uber might do ~3.8bn US mobility trips. This means that by 2030, Waymo would still only have just ~10% of the US rideshare market. And by that point, Uber’s AV initiatives would be well underway.

From Uber’s Q4’25 prepared remarks:

“As we have said many times before, we have barely scratched the surface of the AV opportunity, with AV trips (on or off Uber’s platform) globally accounting for just 0.1% of global rideshare trips. Given our growth at scale, autonomous vehicles are likely to remain a very small portion of the rideshare category for many years to come. As a point of comparison, our Mobility business is currently adding ~50x the total global AV category volume.”

Lastly, Waymo/Tesla are also trying to scale in an industry (owned fleets) that is not exactly conducive to the kind of explosive growth that we saw in the early days of Uber. Here is Uber management from a Mar-26 Morgan Stanley conference:

“The second thing I’d say here is what tends to be true for winner-take-all or winner-take-most kind of deployments is that they tend to have a very hockey stick inflection curve, and those gating factors we just talked about ensure that this is going to grow at a slower rate than what the fundamental nature of a winner-take-all deployment is. And I think the way I think about this in my mind is the fastest-growing AV deployments right now are, at best, tripling their volumes year-on-year. In the early years of Uber’s deployment, for the first 6 or 7 years, Uber almost 9 to 10x-ed its volumes every year, right? That was the exponential curve, which we were on and where AVs are right now.”

So to recap, AVs currently threaten a relatively small piece of Uber’s business today (especially if we assume the suburbs are mostly off limits). That “threatened” piece of Uber’s business is actually slated to accelerate in 2026. And finally AVs are expanding off a small base, and though rides are growing quickly, there are real constraints around just how fast the shift can go.

If Large, Aggressive Competitors Engage in ‘Growth at Any Cost,’ It Could Suck

In this business, if you have a competitor that does not care about lighting money on fire, it can be especially rough. You can get people to switch from Uber if you make your offering cheap enough. Ultimately getting from A to B, or getting something delivered, is commodity-adjacent. People will switch if they’re getting a good enough deal. And given the large potential prizes in mobility/delivery, and the types of competitors circling the space, we may well get some “growth at any cost” challengers.

“GaaC” challengers has been a persistent dynamic in the industry. Uber itself eventually had to throw in the towel in China, SE Asia, India (in food delivery), and Brazil (also in food delivery), as burn rates were too high for even it to stomach.

Uber and Lyft battled it out for almost a decade in the US. Here is Bill Gurley, in a Feb-26 interview with Stratechery, recounting Uber’s rivalry with Lyft:

BT: “Not just that, but Lyft was about to go out of business, and suddenly they get millions of dollars in funding…You go back a year before, it looked like it was over, and suddenly Lyft has new life because of this.”

BG: “Yes, we could see it in the numbers. The reason, especially in the coastal areas, the brand was taking heavy artillery fire, and people were switching to Lyft. By the way, that took eight years to get some people off of. They would ride Lyft, just because they hated the brand of Uber.”

BT: “It severely impaired Uber’s profitability, I would say, for about eight years.”

During this period at the end of the Kalanick era, Uber cracked the door open for Lyft to raise new funds. Lyft would go on to raise ~$7bn (more than its entire valuation today of ~$6bn) across 7 separate fundraises between 2014 and its IPO in 2019. This gave Lyft the firepower to slug it out with Uber for many years, keeping everyone’s profitability in check.

Meituan, China’s leading food delivery company, is dealing with this problem today. Alibaba recently launched a “GaaC” campaign, and is forecasted to increase its market share from 21% in 2024 to 40% in 2027. Meanwhile Meituan, once thought to have an impregnable moat in China, is forecasted to see its share fall from 73% in 2024 to 55% in 2027. According to this FT article, “Goldman Sachs analysts estimate that Meituan will lose on average about Rmb1 for every instant delivery order this year, helping to push it to a third-quarter loss of Rmb16bn — its largest since going public in 2018.”

It hasn’t been a fun couple years for Meituan shareholders. Here are more quotes from the article:

Even Brazil’s iFood, with some 80-90% market share in food delivery (which you might think would be enough to ward off competitors), is dealing with aggressive new entrants in Meituan and DiDi.

What’s the best defense against a “GaaC” competitor? Unfortunately, the only real option, it seems, is to go scorched earth back, and hope to shut the door on your competitor as quickly as possible. Here is Uber management discussing their response to Chinese rideshare leader DiDi spending aggressively to take share in Latin America. From Q1’24 earnings:

“I think in terms of Latin America and the competitive environment there, first thing I’d say, I’m assuming you’re asking about Mobility, we’re seeing very healthy Mobility volume growth in Latin America, in mid-20s. So we like the market, and we certainly like the volumes that we’re seeing there. I would say that while — I think you’re referring to DiDi, they signaled a bit more capital discipline, we’re not seeing that as of yet. We see DiDi being highly competitive in the marketplace and spending into the marketplace quite aggressively.

Listen, it could be temporary. It might be driven by their desire to show international growth as the China markets have slowed down a bit as the prep for the IPO, but it’s difficult for us to speculate on that. And I’d say, we’ve seen this behavior before, Brian. And we have a very strong record of effectively responding to defend our category position when our competitors spend up and we do the same thing, and typically, we’re much more efficient than our competition in terms of financial efficiency, network efficiency, et cetera. But at this point, we see DiDi leaning in, certainly not leaning out.

And we are leaning in as a response, just like we do with other competitors all around the world. The good news for us is we have a very strong P&L, you see our margins continue to increase, so we have lots of pockets of investments to reach into, but we are going to be aggressive.”

What could deter a “growth at any cost” approach today?

• Uber’s size. Uber’s size is a deterrent. At this point, Uber can clearly spend what is necessary to defend itself successfully in just about any market — so what’s the point in challenging Uber? It has a ~$150bn mkt cap, will do ~$16bn in EBITDA in 2027 (pro forma for DH), and operates across ~100 countries. Picking on Uber today would only really make sense if you are a Mag-7 company.

• Uber One. ~50% of Uber’s bookings now come from Uber One members (a figure that is growing rapidly). That’s a level of customer stickiness that hasn’t existed in the industry historically.

• The industry is more mature. The industry is a lot more consolidated today. Further, most of Uber’s legacy competitors are now public, and scorched earth tactics would likely tank their stocks.

Ultimately, there’s nothing really Uber (or anybody) can do about a competitor going into full “land grab” mode, and it sucks for everyone until somebody throws in the towel.

As mentioned, it seems the only groups for whom it could make sense to challenge Uber are Mag-7 companies. Unfortunately for Uber, three of the Mag-7 (Google, Amazon and Tesla) have robotaxi initiatives.

Uber Partners Can’t Get to the Finish Line on L4 Autonomy

It took Waymo 17+ years and $30bn+ to achieve L4 autonomy. Tesla has been working on (and promising) self-driving for about a decade. Despite a tremendous amount of effort (and a tremendous amount of real-world driving data), Tesla is still at L2+/++.

Why do we think challengers like Nuro (or even NVIDIA) aren’t going to slog away for another few years without fully getting to the finish line?

One strong rebuttal is that there are currently three Chinese players that, while not on the same level as Waymo, have just about solved L4, it seems. Their combined robotaxi fleet in operation is expected to reach “tens of thousands” by the end of 2026. WeRide currently offers autonomous rides in soon-to-be five international cities for Uber. Pony.ai does so in Zagreb and soon Dubai. And Apollo Go (owned by Baidu) is arguably the most advanced of the three.

Another rebuttal is that the path to L4 has gotten much easier since the early days of Waymo. For one, sensor hardware and compute has gotten much better/more powerful and much cheaper, which is a tremendous help. Additionally, the ability to simulate driving scenarios to train a model has become a very powerful thing, somewhat obfuscating the need for a tremendous amount of real-world driving data. But most importantly, the rise of AI/transformers has brought about a second generation of AV companies, which are basically using end-to-end AI to solve for AV. To way oversimplify: a bunch of sensor information comes in (through cameras, LiDAR, radar), there’s a giant neural net in the middle, and driving instructions go out the other end.2526 As one can imagine, training a giant neural net (though still incredibly difficult) is much easier, and way less costly, than employing an army of engineers to basically hard-code driving logic.

Some other reasons for optimism:

• Nuro is close to launching (in ~H2’26) what seems like it should be a pretty robust, unsupervised commercial service in the SF Bay Area. It should be noted, however, that Nuro’s key initial vehicle partner — Nuro — is potentially on the verge of bankruptcy. This would be quite an unfortunate development for Nuro, Lucid and Uber, potentially derailing the whole partnership.

• NVIDIA is NVIDIA. It’s the largest company in the world, and self-driving is an important initiative for them. They have been working on self-driving for ~7 years and are close to launching their own fleet of pilot vehicles (in ~H1’27).

• Some carmakers are going at this alone, like VW (through its MOIA division) and Rivian (which even has its own chips). While we’ll see how good a decision this proves to be, just the fact that they think they can do it, perhaps shows that this is doable within a reasonable set of constraints.

• Waymo has not only achieved L4, but they’ve kind of nailed it. Yes Waymos have issues, but the Waymo Driver is apparently ~10x safer than the average human driver already.

• The reason Tesla can’t get over the finish line may be unique to Tesla: they are the only company attempting a camera-only approach. This increases the degree of difficulty materially, and may be a large part of what’s holding Tesla back.

In all, I think we’ll get at least a couple more players over the finish line reasonably soon. Here is Dara from a May-26 Decoder interview, expressing strong confidence in this belief:

“All the evidence that we see is, yes, Waymo is past the finish line. They’re the leader, they are in many ways inspiration for many, many companies in this industry. They’re a great partner of ours in Atlanta, in Austin.

There are many other companies that are getting to the finish line. WeRide for example, or a Pony.ai, or Baidu, these are Chinese companies, are already at the finish line. And we are in market, for example, with WeRide in the Middle East, and there are players like a Nuro or a Waabi, or an Avride, or a Wayve, all of whom are accelerating to the finish line. And if anything, the speed of getting to the finish line is accelerating.

One, model capabilities are much, much better now, used to be kind of deterministic, you know, kind of code that you had to slog through. Now, obviously it’s learning AI models. ‘Sim’ capability is much better, so that data will go much further in terms of model training.”

Here is Xinzhou Wu, Head of Automotive at NVIDIA, on a Jul-26 Decoder podcast going on the record that L4 will be an ADAS-like commodity feature in less than 5 years:

Interviewer: “The question I have is the mainstream experience feels like you just buy a car and, just like level two ADAS is kind of a commodity in cars now, level four will be a mainstream commodity in cars. You push the button it starts driving itself. How far away do you think we are from that?”

XW: “Well first of all that’s exactly my mission, trying to help the industry to get there. I would say if I need to give a time I would say five years, less than five years.”

Over the long-term, the most likely case may be that the number of AV providers will be whittled down to just a handful, however. The most pertinent reason being that it’s just very costly to train these models. Once a few winners are established, generating real revenue, and extending leadership in their tech, I suspect others will start to fall away. Perhaps the most likely case is that Waymo, NVIDIA, 2-3 Chinese players, Tesla, probably Zoox and perhaps another 1-2 players find some degree of success in AVs.

No More Supply “Exclusivity?”

What if fleet owners make their cars available on any 1P network (Uber, Lyft, Waymo, Tesla, Zoox, etc.)? This is probably much easier to do in an AV world?

I suspect the answer is “yes” and that this probably will happen to some degree. It already happens today — many drivers simultaneously drive for both Uber and Lyft.

Here are reasons this may not happen:

• Uber’s Autonomous Solutions. This offering does a fair amount of things that seem pretty sticky: handling insurance, handling in-car experience including in-car customer support, “control tower” software for fleet owners, handling complex rides (like shared rides, “reserve” and “wait & save”).

• Depots / fleet partnerships. It seems likely that Uber will own many of its own depots/charging infrastructure in certain markets. It may have exclusive partnerships with certain fleet operators in certain markets. If AV owners want access to these things, they’d have to make their cars available on Uber, perhaps exclusively.

• Financing complications. Some AVs will operate like today’s Ubers — sometimes referred to as the “agency model.” There will be an owner, the owner pays for everything (except insurance when driving for Uber) and Uber will take a percentage of the ride fare and pay out the rest to the AV owner. In an agency model, an owner could make their cars available on many platforms at once, in theory. But some AVs could be financed through a “leasing model.” In this model, Uber pays out a fixed fee per mile (with a minimum mile guarantee), and keeps the rest. This provides more certainty to vehicle owners, but could also lock them into the Uber network.

• Other strategies. There might be other strategies Uber could employ, like offering “priority” to vehicles exclusive to Uber’s platform. Something like this might drastically change the calculus for fleet owners.

• Some AVs will likely just be owned by Uber

Given all this, it may not be feasible for AVs to operate on more than 1-2 networks.

Further, as long as Uber is one of the networks that everyone makes their AVs available on, then Uber should theoretically retain its crown as the most liquid marketplace and be able to somewhat sidestep this risk.

Are We Sure Uber is Capital-Lite?

For a capital-lite business, Uber sure is deploying a lot of capital. In fact, according to the FT, the company has committed in the neighborhood of ~$10bn to AVs between its fleet commitments and investments in AV tech providers:

From the article:

“These deals put Uber on track to invest more than $2.5bn in equity stakes and spend over $7.5bn on robotaxi fleets in the next few years, according to FT calculations based on analyst estimates and people familiar with Uber’s deals. The agreements are contingent on its partners hitting certain deployment milestones.”

Quoted in this article is Walter Piecyk, who hosts the AUTNMY podcast (a good listen!), and is a close follower of the space. He believes Uber is transitioning into a more asset-heavy business:

“Walter Piecyk, analyst at LightShed Partners, a venture capital investor, believes Uber’s recent robotaxi spending spree is ‘step one in a complete narrative change’ around its business. ‘I think they are easing people into the concept of Uber owning fleets of vehicles,’ he said.”

While the company is using its balance sheet to seed the platform with AVs, the company has been clear that its preference is to remain asset-light. Here is Andrew Macdonald on a Feb-26 The Compound podcast:

“So Uber’s famously an asset light marketplace. It’s been one of the appeals for investors about our business model and our intent long-term is to maintain that. 10 years from now I don’t want to be sitting here as the largest owner of autonomous vehicles in the world, even if I am confident that we’re going to be the largest network for getting autonomous rides…

I want to have them on our network. I don’t want to have them on our balance sheet. But there’s a difference between what we think the end-state model looks like and what we think we need to do in the next couple of years to make sure that we lead in that end-state model. And I think the reality is there’s a lot of value in having a company like Uber step up and be able to underwrite some of the asset purchases, or in the case of working with OEMs, being able to help them stand up a supply chain on their side and start mass producing these things knowing that they’ve got guaranteed locked in demand from Uber. So we will play that role in the short term.”

Getting more asset-heavy during this transition period seems necessary. The economics of robotaxis are not proven yet (and frankly probably aren’t good at this point), so there is no one willing to buy these cars. And if there is no one willing to buy them, there of course is no one willing to build them. But Uber needs carmakers to build AVs. Otherwise, Waymo, Tesla, etc. will pull ahead even further. So Uber’s strategy has been to use its financial might to buy cars and drive the broader AV industry forward. I have no qualms with this strategy — it seems like the right thing to do.

The question is, will they be able to pull back on buying these cars at some point (with others coming in to take their place)? It’s Uber management’s belief that, as the economics are proven out, investors/fleet operators will begin buying up these cars.

In their mind, the category will “financialize,” similar to how the world of hotels has financialized over the years. In hotels, brands like Marriott and Hyatt no longer actually own the hotels. They own the hotel brand and receive rev share in exchange for said branding and other services. You have independent groups that staff and manage the hotels (i.e. fleet operators, in this scenario). And then you have the groups that actually own the hotel (i.e. the land and buildings). These groups might be anything from a large REIT (that owns hundreds) down to an individual (that may own just one).

In Uber’s view, AV fleets will be similar. Large financial groups like a Blackstone may own big fleets operating in big cities like NYC. Single individuals (like current Uber drivers) might also own perhaps just a handful. And there will be many in between.

From a Mar-26 Semafor podcast:

“So the ecosystem that we see is that in the big cities, for example, there will be very large fleets that are going to be owned by big financial institutions, the Blackstone’s of the world etc. They they will own and finance the fleets. We will build that whole operating infrastructure, the fleet management tools, the recharging etc. And so in large markets, you will have institutional supply.

And then on the on the outskirts of those markets, in a Westchester County, Tarrytown, New York or that area, you’re going to have smaller entrepreneurs. There may be two people with a garage that have bought 20 cars that are running a smaller fleet management operation, and it’s going to be quite entrepreneurial. Hopefully, a lot of those people will be our drivers. And then I do think there will be a minority of individuals who kind of send their cars out. It will certainly be something that you can do.

But I imagine it to be a little bit like Airbnb, where you’ve got these “superhosts.” And then you’ve got — a lot of Airbnb’s volume sure comes from individuals, but there are these “superhosts” as well. You will have “superhosts” in the big cities — there’ll be institutions — and then on the outskirts in the suburbs, you’ll have kind of smaller fleet managers who are entrepreneurs who are building business for themselves.”

I think it probably is true that if there is a healthy, predictable return to be made, investors will come in. However, none of this has been proven out yet, and there are real risks for investors in AV supply today — for example, how long will these cars last, what will residual values be, what happens when cheaper next-gen cars undercut prices, can we trust that Uber won’t be disrupted by Waymo/Tesla?

I also believe management has started to convey an openness to, let’s say, becoming less asset-light. Here is COO Andrew MacDonald from a May-26 Rapid Response interview:

“We have 25 autonomous vehicle partnerships today globally and [are] starting to work with our partners to scale, helping them build out the physical world infrastructure they need to scale whether that’s depot or charging capacity or remote vehicle management or customer support…

I think many models will emerge. I think you’ll have fleets that will own these assets. I think you’ll have financial investors that will own these assets. I think you’ll have individuals that will own these assets. And so whether or not we need to own vehicles directly, own fleets directly, own the technology directly, I think is something that will get answered over time. Our position today is that that’s not our role in the ecosystem. But like any management team, we should always be open to changing our mind. If the facts change or our strategy is proven to be wrong, then then we’ll have to change direction.”

I think the chances are we will see Uber continue down the asset-heavy path for awhile longer, longer than probably Uber initially thought. Over time, as the industry matures, I think it’s also reasonably likely that most vehicles are owned by third parties.

Agent Disintermediation

In the future, some percentage of rides might come through an agent. Maybe you request a ride from your personal assistant agent. The agent knows your preferences, etc. It goes out and books you a ride.

The problem for Uber is that before booking your ride, your agent might compare the cost of a ride on Uber with the cost of a ride on Lyft, Waymo, Tesla, etc., and book you the cheapest one. This would have a further commoditizing effect on Uber — you might not even know you are riding in one.

I could also see this being a problem with Uber Eats. An agent could take your order and check pricing on Uber Eats, DoorDash, or directly with the restaurant. Oftentimes, direct to the restaurant may be the cheapest route. While Uber Eats might ultimately still make the delivery (through Uber Direct in this case), Uber might extract a lower take rate.

A mitigating factor across both delivery and mobility is Uber One, which has the effect of locking users into the Uber ecosystem. However, it’s unclear if the lock-in from Uber One would be powerful enough to counteract this disintermediation.

Other mitigating factors, with respect to delivery, are that (1) it’s not that hard to open the app and put in an order (and there is a smaller chance the order is placed wrong by doing so), (2) I’d say most users like to browse the app to see what they feel like eating and (3) probably only by opening the app will you be able to see what deals are on offer.

Lastly, Uber could also just elect to not allow agents to access its platforms. Amazon has done this in shopping, preventing third party shopping agents from “crawling” its site. I think Uber might ultimately have trouble with such a strategy, however. Blocking personal agents from Google or Meta, for example, seems like it would do more harm than good.

In all, while I don’t quite see the risk yet, I understand why some are focused on it and it’s certainly something to monitor.

Is Delivery Going to Hit a Wall?

Here is Jim Chanos making the case that we can’t expect food delivery to outpace food restaurant SSS forever.

And indeed if we look at Uber Eats and DoorDash’s combined delivery business, we do seem to be getting “up there” in terms of their share of the market. One estimate, detailed later on, puts their combined penetration at around ~10% of total restaurant spend — how much higher can we reasonably go?

At the same time, even Chanos qualifies his tweet above by saying it can’t be expected for the next “decade.” And he is smart to qualify, because Uber Eats and DoorDash continue to grow their delivery businesses quite nicely.

Look at these bookings growth rates from Uber Eats…

and DoorDash…

Uber Eats and DoorDash have grown by 15% and 22% (organic) CAGRs since 2021, respectively. Further, growth in 2025 actually accelerated for both companies. Based purely on momentum, healthy growth seems poised to continue for awhile longer.

For growth to sustain over the medium to long term, however, one probably needs to believe one or more of the following things: (1) that DoorDash and Uber Eats can successfully penetrate the grocery market (a market much larger than food delivery), (2) that DoorDash and Uber Eats can have some level of success with retail and/or their white label delivery offerings (fulfilling orders placed directly with merchants), (3) that autonomous delivery will happen which will lower costs and unlock a greater piece of all of the markets above, and/or (4) restaurants will reduce the cost of making food through automation (ala Travis Kalanick’s Cloud Kitchens), allowing restaurants to take some incremental share from grocery.

I would also mention that folks do not seem to be paying an unsustainable amount on delivery platforms like Uber Eats and DoorDash, like some articles would lead you to believe. In fact, total spend on restaurants (as a share of total consumption) has actually never been lower, according to this chart (see red line below) from Mike Konczal.

As noted in Konczal’s article, while delivery spend is way up, most are simply substituting going to a restaurant with delivery. They are not expanding total spend on restaurants. In fact, that spend is continuing to fall, while food at home (i.e. grocery) spend has ticked up.

I think delivery will likely experience healthy growth longer than some anticipate. It is so much more convenient to have paper towels delivered, or diapers in a pinch, groceries, etc. As more and more people deliver, density improves and costs come down.27 As autonomous delivery likely proliferates over the next decade, costs will fall even further.

Additionally, Amazon is relentlessly investing in faster and faster delivery (which drives conversion). As free same-day delivery proliferates, what are other retailers to do? They must offer the same, and Uber/DoorDash are well positioned to be the partners of choice for these retailers.

Here is Michael Morton, a Sr Research Analyst with MoffettNathanson, discussing the impact of same-day delivery in a Jun-26 interview with Stratechery:

BT: “I think the point you made before about speed, there’s a mental shift that happens when you actually believe you’ll get something in a few hours, where it becomes absurd to think you’ll go to a store. Your mind starts calculating, “What if I forget to go to the store?”, “What if something comes up?”, whereas if I order it right now, it’s going to be here in a few hours. That window just closes in terms of even considering going to bricks-and-mortar, and in my experience it drives a really marked shift in your default purchase pattern.”

MM: “You wouldn’t believe the conversion rates — when we tell people we can get them something at the end of the day or tomorrow morning when they wake up, they go through the roof.”

While it’s of course important to be realistic about how much more penetrated delivery can get, I do think there’s a good amount of headroom left.

Other

• Uber Has a Captive Insurer. It’s hard to be good at insurance. It’s especially hard if you are not even an insurer. Uber reportedly self-funds nearly 95% of its total insurance exposure. Thus, Uber is a non-insurer that must run its own pretty giant book of insurance (the company had $12.5bn of insurance reserves as of Q4’25). This is the same book that has confounded actual insurers in the past (which ultimately led to Uber having to self-insure). However, despite this, the insurance risk seems <gulp> manageable at this point. The company reported in its latest 10-K that changes to insurance reserve estimates as a result of prior period claims (a key metric in judging the quality of insurance underwriting) amounted to just $158mm, -$78mm and -$21mm in 2023, 2024 and 2025, respectively. I’m no insurance expert, but that seems like a pretty solid (recent) track record vs. a total quantum of insurance reserves in the multi-billions. I think we can be reasonably confident that after a period of poor underwriting (by Uber’s insurance partners), Uber has gotten its hands around how to price its insurance (for its non-AV business at least). Uber has also made it a point over the past few years to seek regulatory relief to help rein in insurance costs — an initiative that seems to be working.

• Macro Risk. How will Uber respond in a severe downturn? Surely there will be less food delivery and less mobility trips, all things equal? The answer to this is yes — however, all things will also not be equal. There is a curious thing that happens with Uber in a downturn, which is that more “supply” (or labor) comes online. When more supply comes online, prices fall and demand picks up. Uber has made this case (as well as proven this in actual downturns) many times.282930 So while a downturn would not be good for Uber, it may not be as bad as some expect.

• Assault Lawsuits. Uber is facing more than 3,300 assault lawsuits, which have been consolidated in US federal court. In one of the first of these lawsuits to go to trial (in Feb-26), the plaintiff was awarded $8.5mm in damages. In this case, the driver had no criminal history, and had made over 10,000 trips on Uber with a “nearly perfect rating from riders.” The plaintiff was seeking $140mm in damages. The jury awarded “compensatory damages” but declined to award “punitive damages.” From the article: “In a statement, an Uber spokesperson noted that the jury rejected Dean’s other claims, that the company was negligent or that its safety systems were defective, adding that the company plans to appeal.” In another one of these lawsuits that went to trial, according to article, the jury sided with Uber: “The jury found that while the company had been negligent with its safety measures, that negligence was not a substantial factor in causing the woman’s harm.” More recently in Apr-26, a jury in North Carolina found that Uber was liable for a driver who grabbed the inner thigh of a passenger as she was leaving the front seat of his car, awarding the plaintiff $5,000 in damages. In this case the driver denied touching the plaintiff, and the plaintiff never reported the incident to law enforcement. Uber only learned of the incident 3 years later when the lawsuit was filed. From the article: “‘The jury’s award here should further bring these cases back to reality, as it represents a tiny fraction of previous demands,’ the Uber statement said, adding that the company has strong grounds for appeal because it believes the jury was incorrectly instructed on the question of liability.” These cases are considered “bellwether” cases that will impact what will likely be a large negotiated settlement between Uber and the collective group of plaintiffs (each plaintiff can individually opt out of the settlement, ultimately, if they choose). If we assume the average settlement is $150K, across 3,300 plaintiffs the total settlement would be $165m. Uber recently enabled a “woman-driver preferred” option as part of steps taken to address rider safety.

• Bad AV Accident(s). If we have an especially bad accident (or series of accidents) from AV partners operating on Uber, Uber could develop a reputation for unsafe AVs. Because Uber works with more partners (that are less proven than Waymo), Uber seems more at risk. While Uber is perhaps better positioned in that they would have other AV providers to work with if one provider’s tech was faulty (unlike Waymo), its unclear whether the average consumer would understand the nuance (i.e. “don’t worry — the faulty AVs are gone, the remaining AVs are safe”).

• Waymo/Amazon buys Lyft. If Waymo were to buy Lyft, you’d suddenly have a deep-pocketed, formidable #2 player in US mobility. Even worse might be if Amazon acquired Lyft. Then you’d have 3 deep-pocketed, formidable players. Reasons to not be as concerned: (1) if Waymo/Amazon bought Lyft, it would seemingly detract from their current advantage (that all cars are AVs), (2) human-drivers are likely not seen as the future at either Waymo or Zoox, and having to operate a hybrid offering would ratchet up the complexity of these businesses (both factors making a deal less likely), (3) Lyft only operates in the US and only in mobility, which limits its usefulness and (4) in the SF market share data, Lyft has been taking more of the market share damage vs. Uber, so Lyft might start to see real impairment to its business as Waymo and Uber ramp AVs over the coming years.

Financial Performance

Ride/Delivery Economics

Uber takes a cut of every “trip” (i.e. ride or delivery) that takes place on its marketplace. In the S-1, they provide example breakdowns of the economics of a trip.

The table below from Uber’s S-1 illustrates two different “ride” scenarios — without and with “excess driver incentives:”

In this example, the total cost of the ride is $10.00, which is recorded as “gross bookings.” The driver earns $8.00 in the scenario without excess driver incentives, and $11.00 in the scenario with excess driver incentives, taking all of the cost of the ride (and then some). Uber records the first scenario as $2.00 of revenue to Uber (which represents a “take rate” of 20%). In the second scenario, no revenue is recorded (a “take rate” of 0%) and in addition to that, there is $1.00 of CoR (i.e. Uber loses money in this scenario).

Incentives are provided to drivers in order to balance supply (# of drivers/couriers) with demand (# of riders/consumers) on the platform at any given time. For example during peak rush hour, or during large scale events, Uber will increase its payouts to drivers in order to activate more drivers in its app (note that Uber will also implement “surge pricing” in order to recoup some/all of this excess payout; “surge pricing” also tamps down on rider demand to help balance supply/demand).

The table below from Uber’s S-1 illustrates the components of a delivery order:

In this example, the total order value is $22.00 including food and delivery fees. This is recorded as “gross bookings.” Of the $18.00 that the restaurant is “owed,” they receive $13.50. Of the $4.00 that the courier is “owed,” they receive all $4.00 (and an additional $2.00 of “excess driver incentives”). Uber receives $2.50 in this scenario ($4.50 less the $2.00 in “excess driver incentives,” which is again recorded as CoR).

Gross Bookings for Uber does not include tips (for both mobility and delivery).31 This is actually not the case for DoorDash, which includes tips in its GOV (Gross Order Value).32 Both Uber and DoorDash “pass through” 100% of tips to drivers/couriers.

Note that a couple of significant accounting changes took place following Uber’s IPO in 2019:

• In Q4’20, Uber began recognizing all “excess driver incentives” as revenue reduction, removing them from CoR.

• Beginning in 2020, for certain markets where Uber was deemed the “merchant of record” and not simply an “agent” (primarily the UK, I believe), Uber began recognizing gross bookings as revenue and associated driver/courier costs as CoR. Over the 2021-2022 period, this materially boosted revenue growth rates and the overall “take rate,” though it compressed gross margins (no impact on bookings or overall profitability).33

• Further complicating things, beginning in Jan-26 following a UK tax law ruling, Uber decided to reclassify all of the UK (ex London) from the “merchant” model back to the “agent” model. This will have the opposite impact of before: revenue growth will slow, the overall “take rate” will decrease, but gross margins will increase (again, no impact on bookings or profitability)34

Uber Financial Model

The flow through from gross bookings to revenue is a little more complicated than the examples provided above.

The main differences are that (1) “market-wide promotions” including the discounts available under the Uber One membership program are also recorded as a reduction of revenue, as are (2) certain “refunds and credits” due to end-user dissatisfaction.

“Cost of revenue” consists of primarily driver/courier costs for mobility/delivery transactions in markets where Uber is the “merchant of record,” insurance costs and credit card processing fees. CoR also includes data center and networking expenses, and mobile device and service costs.

Note that insurance costs (Uber provides “commercial” automotive liability insurance on behalf of its mobility/delivery drivers) are essentially a 100% “pass-through” item. This means that the cost of insurance is “priced-in” to the cost of a trip (thereby increasing gross bookings) but then deducted in the CoR line. Thus, rising insurance costs (they ratcheted up materially over the 2021-2024 timeframe) had the effect of optically increasing the take rate (when take rates are calculated as simply revenue/gross bookings).

“Operations & support” primarily consists of expenses related to employees that support operations in cities, including general managers, driver operations, platform user support representatives and community managers.

“Sales & marketing” primarily consist of advertising costs, product marketing costs, discounts, loyalty programs, promotions, refunds, and credits provided to end-users who are not customers, compensation costs to sales and marketing employees, and the allocation of certain corporate costs.

“Research & development” primarily consists of compensation costs for employees in engineering, design and product development. In the earlier years, Uber’s ATG effort (i.e. it’s self-driving initiative) was recorded primarily in R&D (this business was divested in Dec-20). Substantially all R&D is expensed as occurred.

Gross Bookings

Gross bookings have ~5.5x’d since 2017. This represents a 24% CAGR over 8 years.

As we can see below, a big driver of this has been delivery bookings, which increased by over 30x over this period (vs. mobility, which increased by roughly 3x).

Prior to the DH acquisition, gross bookings was split 50%/47%/3% between mobility, delivery and freight, respectively. Including the impact of Delivery Hero, total bookings were $235bn in 2025, with 41%/56%/2% coming from mobility, delivery and freight, respectively.

Both sides of the business have performed well over the last three years, with CAGRs of 23% and 18% for mobility and delivery, respectively. Much of Uber’s growth can be boiled down to (1) strong user growth and (2) more spend per user. Encouragingly, both factors continue to exhibit healthy trends, as detailed below.353637

Strong user growth has resulted from a broad set of factors, including: (1) expansion into new countries (like Japan, South Korea, Germany, Spain, Turkey and Argentina), (2) a much stronger focus on the suburbs (an effort boosted by new products like “reserve” and “wait & save”), (3) expansion into younger and older demographics, (4) affordability initiatives (including Uber One and new products like “Moto,” a two-wheeler product for emerging markets) and (5) new products like Uber for Business (U4B generates ~$5bn in annualized bookings and is growing 45% YoY as of Q1’26).

More spend per user has also resulted from a broad set of factors, including: (1) driving multi-product usage (Uber, Uber Eats, groceries, retail), (2) Uber One, which drives higher trip frequency and better retention, and (3) a higher quality product (for Uber Eats, that means shorter delivery times, more merchants on the platform, etc.).

MAPCs (Monthly Active Platform Consumers) / Uber One Members

Growth in MAPCs and growth in Uber One members has been very healthy.

MAPCs have grown at a 16% CAGR over the last three years, and growth actually accelerated in 2025 to 18%. Meanwhile the number of Uber One members has grown at a 3-year CAGR of 57% (Uber One was only recently introduced in 2021). Uber One members hit 46mm in 2025, and impressively maintained a growth rate north of 50% in 2025.

Delivery Hero brings another 49mm MAPCs. If we assume there is 20% overlap with existing Uber mobility users, and that 1/3rd of the remaining MAPCs convert to Uber One members, then in theory, DH would generate another ~13mm Uber One members.

Uber One members now represent just under 50% of total bookings on the Uber platform. This is up from just 25% in 2022.

The number of monthly “trips” (deliveries + rides) per MAPC (a measure of trip “frequency”) shows a healthy looking trend as well.

There is headroom to increase MAPCs and trip frequency much further.

Firstly, only ~5% of the adult population across Uber’s current operating footprint are Uber MAPCs.38 For Uber’s top markets, the number is closer to 15%.39

On top of this, in terms of trip frequency, 50% of MAPCs only use Uber once or twice per month, per the company’s 2024 investor day. COO Andrew Macdonald on The Compound claimed that in Uber’s best markets, monthly trips are around 9 (vs. ~6 across all markets), and that for the top decile of customers in Uber’s top markets, monthly trips are 20+.40 Meituan, the top delivery player in China, apparently generates around ~9 trips per month.

Lastly, the percentage of users who use both Uber and Uber Eats together is only around 20%. All of these metrics could seemingly go materially higher.

Revenue

Revenue has grown at a slightly faster overall clip than gross bookings, ~6.5x’ing overall since 2017 (representing a CAGR of 27%).

While, pre DH, mobility/delivery was split roughly 50/50 in terms of bookings, on a revenue basis the split was closer to 67/33 in favor of mobility. I believe the difference mostly arises from: (1) mobility having a slightly higher “true” take rate vs. delivery, (2) higher insurance costs for mobility, and (3) mobility having a higher proportion of revenue where Uber is deemed the “principal,” requiring Uber to recognize bookings as revenue.

While mobility growth has outpaced delivery growth over the last three years, delivery growth of 25% in 2025 significantly outpaced mobility growth of 18%.

Below is a breakdown of revenue by geo. We can see the US grew the slowest of all Uber’s geographies (by a wide margin) in 2025 at 11%. EMEA (especially non-UK EMEA) was particularly strong, growing 31%.

The US is likely growing the slowest simply because it is the most mature of the geos. Many of these other regions have countries where Uber was relatively slow to penetrate, oftentimes for regulatory reasons. Many of these countries have enjoyed outsized growth, as shown in the slide below from Uber’s 2024 investor day:

I also think geos where Uber enjoys high market share in both mobility and delivery (EMEA, Canada) are likely getting a boost.

If we include the impact of Delivery Hero, US revenue as a percentage of total revenue declines to 37% (from 46%).

We can see that “take rates” (calculated as revenue/gross bookings) dipped in 2020 before recovering in subsequent years. For mobility, take rates jumped to a materially higher level beginning in 2022.

The dip in take rates in 2020-2021 was likely caused by increased driver incentives to boost driver supply coming out of COVID (which would have reduced revenue).41 As driver incentives abated, take rates would have recovered. However, on the mobility side, take rates rose to higher levels than the pre-COVID period, likely as a result of the accounting policy changes in the UK related to Uber being deemed the “merchant of record.” Higher mobility take rates would have also been driven by increases in insurance costs, which surged during the 2021-2024 period.

Note that the calculated take rates are muddied by the factors discussed above. Management oftentimes refers to ~20% as a more accurate representation of the “true” take rate across both business lines.

For example, here is Dara on the HD in HD podcast (Nov-25):

“The company does about, we’ll [say a] run rate of $200 billion. Our take rate there, our ‘true’ take rate — because the accounting is a little funky — is probably 20%. So that’s about you know call it $40 billion that we uh bring into revenue and profitability off of that is probably $8.5 to $9 billion.”

Gross Margin

Gross margins stepped down during the 2020-2022 period and have settled into the ~40% range.

I believe the step up in GMs in 2020 was likely due to the reclassification of “excess driver incentives” from CoR to a contra-revenue item. The step down from 2020-2022 was driven by the impact of the UK merchant-of-record changes. Finally, over the last four years, it’s likely that surging insurance costs weighed on GMs (higher insurance costs increases both revenue and CoR by the same dollar amount).

Opex

Uber has demonstrated pretty remarkable operating leverage on its opex lines.

As we can see below, across each functional area, opex has decreased materially as a percentage of revenue.

The categories demonstrating the most leverage have been S&M and G&A which have decreased from 32%→9% and 29%→6%, respectively, since 2017.

Below are the opex buckets on a dollar-basis. Starting in 2023, we can see a major deceleration of opex spend. Total opex has grown by just a 3% CAGR since 2022!

This operating leverage has resulted, in part, from (1) the end of ZIRP-era incentives (equivalent to raising prices), (2) a highly scalable model with relatively low variable costs, (3) savings generated by automation and driving marketplace efficiencies (e.g. better matching, bundling), and (4) the growth of Uber’s highly profitable advertising business.

Here is Dara from an Oct-25 Decoder interview on Uber’s ZIRP-era spend:

“Travis would hire a GM, parachute them into a market. And you’re right, we did go out and buy supply and really we are as a company a supply-led company. Which is first, you build liquid supply, then you invest in demand. You have to, in the early days, you essentially have to price under-market and you lose a bunch of money in a market. As liquidity increases and the matches increase and the efficiency of the market increases, you can start pulling more profits from a marketplace. That was essentially the formula.

And it might have seemed crazy at the time, but I think that Travis and the founding team got it right, which is creating liquidity in the marketplace is exactly what allowed you to get to profitability. And whoever created the liquidity of supply and demand fastest was the one who ultimately won. So in the early days it was all GMs, spend a ton of money on liquidity.”

As ZIRP spend petered out, Uber was able to price more rationally. They could essentially charge customers a little more (or use less incentives) and pay drivers a bit less (or use less incentives). Impressively, the company has been able to sustain very healthy growth through this process (including strong user growth).

Management would also tell you they’ve been able to generate operational efficiencies through the use of technology/automation. Here is Dara from a Nov-25 HD in HD podcast:

“I do think that ourselves and a bunch of other companies went through a phase where capital was a weapon. And basically it was about how much money can you raise in order to spend to capture markets. And having an enormous amount of capital to spend actually hurts operating discipline because it’s painful to optimize these businesses.

If you build an incredible algorithm that is able to predict demand versus, let’s say, can [just] match [existing] demand in a market…That’s one little product that the team has has built. It’s actually a pretty cool product. That kind of [thing] can bring, let’s say, 1% [greater] efficiency into a market. And then you have 15 of those projects and lo and behold, you’re growing the company at attractive rates.

In the early days, someone would throw a hundred million dollars into the market and all of that work would be blown away. So, to some extent, the excess capital actually worked against excellent operations. And so for me, the art was pulling the capital back over a period and then driving excellent operations and efficiency across the company. And it’s something that we’ve been able to do now which is we’re able to keep innovating…and at the same time have a bottom line that grows significantly in excess of our top line…

I would say the pattern for Uber over the past 6, 7 years is using technology to allow us to essentially automate a bunch of, call it, overhead functions. And improve customer service, reduce error rates, match more effectively, price more effectively. If you’re able to use tech to do so, you can grow revenue without growing your overhead.”

Here is Dara from a Jun-23 Acquired podcast discussing how the company uses ML to to enhance Uber’s pricing/matching engine:

“I think we’re in the scale business…One of the secret sauces that we have is we have a very large and capable marketplace team. These are ML engineers who are building out the systems that match price, all of this connectivity. And when you’re working over an ecosystem of two billion transactions a quarter, the data sets that we have, the experimentation that we can do in terms of what’s the most optimal match, how do you price, etc. It’s just a bigger database than anyone else. So every year matching and pricing probably improves by five percent. So you improve your the marketplace throughput by about five percent, everything else being the same, and that’s like free growth. And when it’s on top of call it $120, $130 billion dollar run rate, it gets big and, again it’s compounding. Every year this machinery gets better.”

How does this ML engine help Uber in practice?

One example is an ability to bundle more deliveries. As the marketplace gets more liquid and as algorithms improve, Uber is better able to match couriers on two (or more) deliveries at a time instead of one (currently about half of deliveries are “batched” per Jun-26 Invest Like the Best podcast). When two orders are batched, Uber can afford to pay the courier more, but not quite as much as what Uber would have to pay for two orders delivered separately. The total value of the order does not change, however, so Uber can pocket the difference. This is a very important generator of profits for a food delivery business.

Another example is the ability to improve “bidding.” When a delivery order comes in, or a ride is requested, Uber will send out bids to drivers/couriers until one of them accepts. Uber needs to decide who is the best person to send a bid to first, then second, third, etc. Uber also needs to decide how much to offer for each bid. Here is Dara describing this process on a Oct-25 Decoder podcast.42 A more liquid marketplace and better matching/pricing algorithms lets Uber squeeze a bit more profitability out of every transaction.

EBITDA/Operating Margin

The result of strong revenue growth and flattish opex spend has been an inflection in margins. Uber has rather steadily (COVID interruption notwithstanding) increased its EBITDA margin from -33% in 2017 to 17% in 2025.

The Delivery Hero business, once synergies are fully realized, stands to generate a ~19%+ EBITDA margin by my math. Thus, the business should be slightly accretive to margins.

Looking at EBITDA by business line, we can see that the majority of profitability was generated by mobility, pre DH. Mobility generated an EBITDA margin of 27% in 2025, while delivery generated a 21% margin.

Rising delivery margins, however, have been driving much of the inflection in overall profitability as we can see above.

If we add Delivery Hero’s ~$2.2bn est. EBITDA post synergies, we get to ~$11.0bn in combined EBITDA. Of that, ~$5.8bn would come from delivery and ~$7.9bn from mobility (less $2.7bn in corporate G&A and platform R&D).

At the 2024 investor day, the company provided us with an interesting look at profitability by business line by market (for their top 10 markets).

Almost every geo was profitable at the time and almost every geo also improved margins over the preceding 2 years (note this chart shows EBITDA margin as a percentage of gross bookings, not as a percentage of revenue). However, there were still considerable differences in profitability by geo. For example on the mobility side, Uber had both a handful of markets in the ~10% range as well as some that were breakeven (India? Brazil?). Similarly, on the delivery side, Uber had a couple markets in the ~7% range, but most clustered in the 0-2% range.

The profitability of a market depends, in part, on a few things: (1) the maturity of a market (i.e. how much money competitors are “investing” to compete and gain share), (2) Uber’s competitive position within a market, (3) product mix (grocery, for example, is currently a money-losing offering; as is Route Share and Moto on the mobility side), and (4) for delivery, the level of ad monetization.

Here is commentary from a Mar-24 Morgan Stanley conference:

“So it’s a combination of, I’d say, the maturity of the market, competitive position in the market and then especially product mix. So there are a couple of markets on Delivery, where the grocery business is actually quite large compared to the online food delivery business. We’re still investing. We’re still in investment mode on the grocery business. On average, margins are negative in grocery. So in the markets where grocery has -- in some of these markets where grocery has very high penetration, the overall market can be negative. But again, as long as grocery margins move in the right direction, we think will mix into positive for those markets as well.

And then there are a couple of markets where, from a competitive standpoint, we’re not the #1 player yet, let’s say, in Delivery. And there are 3 players who were vying for the #1 position. And the fortunate position that we’re in is, because we’re #1 in a bunch of markets, we have some excess margins that we can reinvest in these markets to grow at outsized rates and hopefully get to the top category position.

So that, typically, is either product mix or its competitive position that we’re aiming for that account for the margins. But we’re quite — this is — it’s a pretty mature model at this point. And based on various take rates and based on product mix, we’re quite confident that we can get every single country that we operate in into EBITDA — solid EBITDA positive margins.”

Here is additional commentary from a 2017 NYT Dealbook conference:

Interviewer: “Let’s talk about the operational business itself. And part of an IPO I imagine — I don’t know if you get there by then — but how do you make this company profitable? There has been an argument made over the years given all the capital that’s come to you and so many of your competitors that we are all having our rides home and elsewhere subsidized by the venture capital community and that there may very well never be a day where the math actually adds up.”

DK: “That day is today because the math is actually working in certain areas and in certain geographies. So if you look at the portfolio of the business, there are certain markets such as the Asia Pacific markets — if you look at Southeast Asia or India where we are in heavy, heavy investment mode and appropriately so. And we lose money. The bigger the company gets, the more money we lose because we’re actually subsidizing those rides. But we have been able to demonstrate a track record — and you’ve got to look in the numbers and look in the geographies that where over a period of time we can pull back on the subsidies — and over a period of time get into a position of profitability on a geographic business. It doesn’t happen everywhere, but I think that we can certainly demonstrate it.

What’s the key and what’s the timeframe for that to happen? It depends on the market and it depends on the competitive dynamics in the marketplace. To some extent the larger the market, the longer you want to subsidize that market, because the ultimate target as far as the rides market or the delivery market is so big that you want to actually juice that business. So it depends. It also depends on the competitive nature of the business. The US is very, very competitive right now between us and Lyft, so I don’t see the US as being a particularly profitable market for the next six months. We’ll see what happens after six months. Depends on where the competition goes. Right now we have a situation where a competitor, Lyft, is spending very, very aggressively in order to gain share.”

Uber is still in “investment mode” across a number of its geos and products. Presumably, at least according to the chart above, if/when these markets/products “mature,” Uber still has a good amount of headroom to grow margins from this alone.

Here is the full Net Income to adjusted EBITDA (to non-GAAP operating income) bridge:

Beginning in Q1’26, Uber will no longer report adjusted EBITDA. Instead it will report the non-GAAP operating income metric above. This metric actually accounts for (i.e. subtracts) both depreciation and SBC (huzzah!).

Free Cash Flow

Uber generated $9.8bn of FCF in 2025, or 19% of revenue. This is up sharply from Uber’s FCF nadir of -$4.9bn in 2019.

Free cash flow, however, is not really the best metric with which to evaluate Uber’s profitability. This is due to Uber’s captive insurance program.

Insurers generally take in money upfront in the form of premiums before paying out claims at a later date. This phenomenon will distort an insurer’s FCF. For example, an insurer that aggressively underwrites a wave of bad policies can look very profitable for a stretch on a FCF basis. However, FCF will ultimately crater as claims on these policies eventually explode. As such, insurance companies are virtually never valued on a FCF basis.

Because Uber underwrites its own insurance, and because the business has been growing strongly, its FCF is being inflated. I believe the quantum of this inflation is roughly comparable to Uber’s growth in accrued insurance reserves. Over 2023-2025, growth in its accrued insurance reserves totaled $2.2bn, $2.8bn and $2.7bn, respectively.

Management will reference a “$10bn in FCF” metric often, but this is a little disingenuous. It doesn’t account for $1.8bn in SBC (that can be forgiven, I suppose). But it also includes growth in accrued insurance reserves of ~$2.7bn, which I believe should be backed out.

Mobility

Key Priorities

Build AV Supply

The introduction of AVs create risk for Uber. In time, self-driving cars will undercut the cost of traditional ridesharing. If Uber is not leading this transition (or a fast follower), they will be at risk.

The quicker Uber’s AV partners can get to L4, and the quicker Uber can add cheap AVs to its platform, the better. Uber is having to use its balance sheet to drive AVs forward at this stage in their development. As discussed above, this seems like the right strategic decision. This does mean, however, that Uber’s business will be getting more capital intensive for at least some period of time.

Uber’s goal is to be the largest purveyor of driverless rides by 2029.

Build Everything Else AVs Will Need

Just getting access to AVs may not be even half the battle, however. As discussed previously, there’s a lot more that is needed to operate a successful AV program (namely AV infrastructure, fleet operations, and a slew of other items).

Uber’s answer to the “slew of other items” is its Autonomous Solutions offering, introduced in Feb-26. Uber breaks down is Autonomous Solutions offering into three buckets: (1) Infrastructure Solutions, (2) User Experience Solutions and (3) Fleet Management Solutions. Within each bucket, Uber offers a handful of capabilities.

Here are the descriptions of each capability from a recent PR:

Infrastructure Solutions

• AV2.0 Training Data. “Uber’s data-collection fleet includes thousands of specially-equipped vehicles across dozens of cities, which has captured millions of diverse miles of easily searchable multi-sensor data across the US and Europe. Working closely with engineers from Uber AV Labs, and building on partnerships like our Data Factory with NVIDIA, we’re helping our AV partners train their models and scale toward Level 4 autonomy much faster.”

• Data-Enriched Mapping. “Informed by tens of billions of trips worldwide and our own in-house mapping solutions, Uber’s dynamic geospatial data help AV partners refine pickups, routing, and ETAs with real-world precision, whether that’s avoiding specific intersections during rush hour or mitigating fleet-wide risks from weather and sudden road closures. Custom APIs help ensure AVs are always in the right places at the right time, even as Operational Design Domains evolve, continually driving both better utilization and a superior rider experience.”

• Complex Venue and Event Management. “With millions of trips completed at airports, stadiums, and event venues, Uber has developed unique data on the complexity of these high-traffic environments and how to ensure a good experience for riders. We are also leveraging existing relationships to work directly with high-traffic venues, like Q2 Stadium in Austin, to optimize venues appropriately for both human- and AV-driven rideshare.”

• Fleet Financing. “Our scale and predictable demand support new models for AV asset financing, helping partners bring more vehicles and new autonomous form factors to the road faster.”

• Regulatory Compliance Services. “Uber’s global policy and regulatory support teams have deep experience and trusted relationships to help bring new technology to markets all over the world, which will be crucial for the commercialization of autonomous technology.”

User Experience Solutions

• In-Car Experience. “Drawing from our best-in-class product design, Uber has developed an in-car AV-first software interface that puts riders in control, with seamless access to sound, temperature, and rider assistance. Designed to work across different hardware configurations and deployments, this unified experience creates a consistent user experience for riders. Launching later this year, the Nuro-Lucid-Uber robotaxi will be the first AV to bring this experience to its in-car tablets, seamlessly integrating Nuro’s real-time driving visualization.”

• Complex Use Cases / Shared & Reserved Rides. “Preparing autonomy for real-world demand will require a diversified product offering. Building on our experience with UberX Share and Uber Reserve, we’re helping autonomous partners develop new products that provide the reliability riders expect, offering the demand density, flexible supply, and regulatory support necessary to make these products possible. We already offer AVs through Uber Reserve in Phoenix, and were the first to announce a shared AV product with Volkswagen, launching later this year in Los Angeles.”

• End-to-End Support. “Uber’s customer support network is built into every stage of the journey. From pre-trip to post-trip, riders can access real-time help with the tap of a button, powered by the same infrastructure that supports more than a billion trips each month worldwide.”

Fleet Management Solutions

• Remote Assistance. “In a perfect world, AVs would never need help. In the real world, Uber provides real-time, on-road support for riders when AVs encounter issues, monitoring situations and providing live communication to the rider and the ADS itself. Additionally, we are designing a new remote assistance platform, built on top of our end-to-end support capabilities, and powered through a custom agent console that provides the operator with everything they need to understand the AV’s status and take action.”

• Field Support Operations. “Live today in multiple markets, our field support teams manage in-field issues, like lost items, towing, manual assistance, cleaning, or alternate rides to keep every trip moving–all visible through AV Mission Control. Where appropriate, we’ll also leverage our hybrid network, whether that’s returning a lost item through a courier, or seamlessly dispatching a human driver during inclement weather.”

• AV Mission Control. “Uber designed and built a comprehensive fleet intelligence and management solution that gives operators a real-time view of every vehicle, turning dozens of per-AV status indicators into system-level insight that keeps fleets efficient, connected, and ready to move. The system’s supply-state machine constantly ingests raw telemetry and signals from the AV fleet, determining a vehicle’s precise status, then uses an intelligent orchestration layer to help tee up possible actions or human interventions. This process is governed by a command authority system that strictly arbitrates who can issue commands, while providing a complete supply-history ledger for auditability.”

• AV Insurance. “Uber’s industry-first insurance program is built for the realities of autonomous operations. All aspects of an AV deployment require bespoke coverage, whether the vehicle is on-trip, off-trip, or charging at a depot. Uber’s Autonomous Vehicle Insurance Policy brings together protection for manufacturers, ADS providers, owners, fleet managers, and other supporting participants into one simple policy that is backed by several of the largest names in the insurance industry.”

Some of these capabilities seem like simply a re-packaging of current capabilities, like “complex venue and event management.” However, many of these are highly important (e.g. field support, data-enriched mapping, AV mission control, AV insurance, remote assistance, in-car experience), and capabilities that a 1P operator would need to develop if they were to try to sidestep Uber or another rideshare operator. Uber is able to build these things once, and then all AV players can benefit. There is no need for each to “reinvent the wheel.”

Note that Autonomous Solutions omits two critical elements to running an AV fleet: (1) fleet operations and (2) physical infrastructure (charging and depots).

For fleet operations, Uber will rely on a network of boots-on-the-ground “fleet operator” partners across the globe. Luckily for Uber, many of these partners already exist, just in a slightly different form. Many Uber drivers today are actually employed by fleet managers (Uber works with 50k+ fleet managers who represent ~20% of mobility bookings), who provide the cars and do all the behind-the-scenes work (repairs, cleaning, etc.) to maximize vehicle uptime.

Here is some commentary from the company’s Q4’24 supplemental which highlights the company’s fleet operations relationships and know-how:

“Fourth, successful commercialization requires scaled, efficient on-the-ground operations, something that sits squarely within Uber’s domain of expertise. It is important to note that an average-utilized AV can run as much as 100K miles a year, compared to a typical consumer vehicle at 10-15K miles a year. This means that AVs need to be charged multiple times a day and serviced monthly. AVs will also require consistent cleaning and available parking, all of which Uber is uniquely equipped to manage. Uber also has 15 years of experience to support a host of other key operational issues like fare disputes, lost item returns, stranded vehicle rescue, and insurance claim resolution.

Uber can deliver the lowest operational costs for our AV partners because we are leaps and bounds ahead on every aspect of the go-to-market capabilities that are critical to commercialization. For example, we partner with more than 50,000 fleet operators that together manage more than a half a million vehicles on our network globally, accounting for nearly 20% of our Mobility Gross Bookings. We also have a highly optimized cost structure that has been honed over the 12 billion annualized trips now happening on Uber. Our AV partners get to plug into that cost structure instantly—with advanced capabilities in customer acquisition, customer support, and the underlying technology and payments stack. Put simply, we are the player with the scale and expertise to run AV operations at the highest efficiency, period.”

Here is Uber COO Andrew Macdonald from a Feb-26 The Compound podcast:

“I also think the physical world deployment is more challenging than people understand. The long tail to launching a new city might not actually be the the L4 software. It may not actually even be the hardware. It may be the amount of power you need to stand up a depot that can house 400 cars at the right location in a new city. And so Uber, taking our existing fleet network — because we have dozens of [them] that operate dozens of large scale fleets that operate on our core business today — and deploying those folks to own and manage assets on behalf of our AV deployments, I think is also really powerful and not well understood. And then there’s a whole host of other things. You think about customer support, payments, identity, risk, all these core capabilities that we can extend to our partners to make their deployments successful — I don’t think is well understood.”

These fleet partner relationships won’t necessarily be exclusive to Uber. However, Uber having a large rolodex of them and the know-how of partnering with them at scale is likely a nice advantage in these initial stages of AV fleet rollouts. I imagine it’s very attractive to say, Lucid/Nuro, that Uber can help you deploy across the globe with various partners all at once, while simultaneously using Uber’s size/leverage to perhaps negotiate more attractive terms than you would have enjoyed otherwise. Uber is also already fluent at helping these operators finance their vehicles, should they opt to own captive fleets (which they already do today, just the non-AV kind).

On infrastructure (charging and depots), it seems increasingly the case that Uber is going to step in to provide it where necessary. This is a capability that sophisticated fleet operators can and do bring. But it seems Uber is again willing to use its balance sheet to push things forward and fill in the gaps. From a Mar-26 Morgan Stanley conference:

“From our perspective, when we think through this deployment curve over the next few years, once we start getting conviction that software is going to hit the mark and it’s going to get to that L4 deployment with superhuman safety, our efforts start shifting to a couple of other priorities.

The first one is securing OEM supply, right? We need to ensure that any of our partners who can deliver that kind of software have enough scale volumes coming to them, and then that volume can get deployed on Uber’s network.

And then the other thing we have to do is to ensure that we are bringing AV infrastructure on the ground. And for instance, for our 2027 deployments, we’re already scouting sites across our footprint. We already have about 50 different sites that we are beginning to get into leasing conversations, and those deployments will start once we have the software there.”

The fact that Uber is in talks with 50 different sites (50?!) for 2027 deployments seems to indicate that Uber is not far behind Waymo (perhaps ahead even) when it comes to a global AV infrastructure rollout.

Perhaps underrated as well is that Uber already operates all over the world. This means Uber already has “boots on the ground” that are familiar with these geos that can help liaise with fleet operators, help from a regulatory standpoint, help secure depots/charging infrastructure, etc. And nobody knows better than Uber (given its decade+ worth of data) where to put depots in all these cities across the world to maximize utilization.

Penetrate the Suburbs

Uber’s mobility business has been relatively slow to target/penetrate the suburbs. It makes sense — in the suburbs, more people have their own cars, parking is much easier to come by, there’s less nightlife per capita, and it’s less dense so it’s harder to avoid “deadheading” as a driver.

Uber has over time, however, developed products that have helped crack open the suburban market.43 One example is Uber Reserve, which allows riders to reserve a time at which an Uber is guaranteed to be at a pick-up location. This helps alleviate the issue of long ETAs that plagues less dense markets, and better unlocks use-cases like the airport trip (though it costs riders more). Another product, Wait & Save, is basically the opposite of Reserve. It allows riders to wait longer for a ride but get a better rate. This helps more cost-conscious riders somewhat mitigate the higher cost of suburban rides. Another product is Uber Teens, designed for teenagers aged 13 to 17. Uber Teens allows teens to request their own rides (and order food via Uber Eats), but with a suite of safety features and parental oversight.

Here is Uber CFO Balaji Krishnamurthy commenting on Uber’s momentum in mobility in the suburbs from a Mar-26 Morgan Stanley conference:

Analyst: “Let’s come back to the core in the U.S. because there’s — the overall business is being run really well. And one of the shining points has been the dense — the less dense or the more sparsely populated market. I think you’ve talked about them growing 1.5x faster than the denser market. Maybe walk us through what has driven that faster growth. And how do you think about sort of investing to keep that more sparse market growth moving quicker through the overall P&L?”

Balaji Krishnamurthy (CFO): “Yes. So what I would say there is that it has been a function of our product innovation as well over the years, right? Historically, the reason we’ve not been able to serve sparser markets is that reliability in those markets is much worse. And over the years, we were able to address the reliability challenge on 2 dimensions. First, we introduced Reserve which, as a consumer, allows you to trade for reliability in exchange for price.

Reserve is more premium, you pay a price and get the reliability that brings you. And then, at the same time, we were also able to launch and expand a product like Wait & Save, which gives you the opposite choice. You get reliability in exchange for time, right? You’re making a trade-off on the time dimension versus price. So as we’ve done that -- and we have gotten better about things like taxi integrations because, historically, again, we didn’t have the 3P integration available.

In many markets, taxi is the best way to go into expansion. All of that has allowed us to go into these sparser markets and serve consumer needs in a way we couldn’t earlier. So what we’re beginning to see now is, today, when you think about the disclosure we gave on earnings where trips happening within our top 20 markets in the U.S. represent about 30% of our gross bookings but only 25% of U.S. mobility EBITDA, that just is a function of our non-top 20 markets growing faster, getting to a profitability position that’s better than our top 20 markets and continuing to grow from there, right?

It’s not at mature state margins either. And the further out you go on that density quartile, the better that story becomes for us, and we feel that the runway here is very, very long. And this is not just a U.S. story. We see similar sort of themes in Europe and markets like LatAm as well, where we are expanding with our Moto product, which is today 15% of first trips in Brazil and 90% of these Moto consumers migrate from Moto to UberX and other Uber products, which are quite profitable for us as well. So lots of runway with the product and geographical expansion we are driving here.”

Another interesting aspect of Uber’s suburban expansion is that these markets, counterintuitively, generate higher margins for Uber. From the Q1’26 prepared remarks:

“We also made solid progress expanding into suburban markets, with YoY trip growth in lower density geographies growing faster than the overall business with higher margins, contributing to our record segment operating income margin of 7.7%.”

The suburbs represent a very large, relatively under-penetrated market for Uber. The success Uber is having penetrating them, and doing so in a nicely profitable way, is a great development. Further, if Uber can win more customers in mobility in the suburbs, this should help them better penetrate the delivery opportunity in the suburbs as well (as the Uber One program would have greater appeal).

Drive Affordability

Key to driving sustainable growth for Uber is driving down prices. Uber has come at this from a few different angles in recent years.

One has been the Uber One program. Uber One materially reduces price on the delivery side, but also does so on the mobility side. As mentioned, Uber recently doubled down on the Uber One program by extending it across families and offering a discount for college-aged kids.

Another has been driving down the cost of commercial insurance. Commercial insurance costs exploded over the 2021-2024 period. Uber has helped cull prices by incentivizing safer driving and through investing in regulatory relief. Here is management from a Dec-25 UBS conference, discussing their regulatory relief efforts, specifically around reducing the size of their mandatory policy coverages:

“On the policy side, we have been working with a number of state situations to improve the equality of -- and the fairness really of the circumstances in various states on how Uber has to -- what levels of insurance Uber has to provide. California is probably one that many of you have heard about more recently. We used to have to carry $1 million of liability for uninsured and underinsured, whereas the car next to us would only have to carry a liability of $300,000. That delta ended up building a cottage industry of folks who were looking for Uber as a source of torque and legal claims. I mean, it really — we became their TAM. And by readjusting that down, we know that the amount of fraudulent use of the legal system in a number of states, California, we’ve done some reforms in Georgia, Nevada, et cetera, are also starting to give us benefit. And the last is, overall, the cost of insurance industry-wide in the U.S. has also started to moderate. So that’s now down to sort of mid-single digits from when I started at Uber, it was — I think it was in the low 20s.”

According to Uber, because it was required to maintain coverage of $1mm for uninsured and underinsured motorists in California, it became a “cottage industry” to extract awards from both Uber and Lyft (taxi companies were only required to carry a $300,000 policy).44 Uber and Lyft successfully reduced their mandatory coverage in California from $1,000,000 per incident to $300,000 per incident (and $60,000 per person) through SB 371, signed into CA law in Oct-25 (effective Jan-26).

Lastly, Uber has introduced new products to help lower prices, including:

• Route Share. Offers pickups every 20 minutes along busy corridors during weekday commute hours. Fares are up to 50% cheaper than the cost of an UberX

• Ride Passes. For $2.99 per month, a rider can lock in a price for a specified route during a specified 1-hr window (avoids high fares due to weather, traffic, etc.). This is a product designed for routine commutes. Riders can also prepay for 5, 10, 15 or 20 rides to unlock even greater discounts

• Moto. Moto, or two-wheelers, is an increasing component of Uber’s offering in certain developing countries like Brazil and India. Moto now represents 15% of all first-time trips in Brazil.45

Here is Dara discussing the Route Share product at its introduction at the 2025 Go-Get event:

“Building out a true low-cost product, shared product, getting 2 or 3 people into a car is a very significant challenge in -- operationally, algorithmically. We’re making progress there. So I’m happy about the progress there, but this is -- I think we will be possibly the only company in the world that’s going to solve this at scale. To be clear, we’re losing money in share, but the loss rates are coming down. The efficiencies continue to increase, and I’m quite confident of the activity of that team.”

The granddaddy of them all, however, will be AVs. AVs will drive a sustained, decade-long+ reduction in pricing.

Strengthen the Product Suite

Uber’s expanded product suite is clearly something that’s moving the needle. The aforementioned “Wait & Save,” “Reserve” and “Uber Teens” products are examples.

One especially key “product” for Uber, which Uber detailed in its Q1’26 prepared remarks, is “Uber for Business” (or U4B). The product-market fit here seems quite strong. Businesses of all flavors have all kinds of mobility and delivery needs. On the delivery side, providing lunch for employees would be a common use-case. On the mobility side, of course, there is business travel. And nobody can provide travel options in more places (and with both value and premium offerings). U4B is a $5bn business today in terms of annualized bookings, serving 300k+ organizations. It is “high margin” per the Q1’26 prepared remarks. And it is growing 45% YoY. Uber is targeting $10bn in gross bookings (and 1mm business customers) by 2028. Like other loyalty programs, U4B allows employees to keep any Uber One credits generated from work trips for personal use.

TAM

Below is the US TAM sensitivity table from earlier showing the US rideshare market opportunity as a product of price per mile x rideshare market share (and assuming 3tn total miles traveled).

This compares to Uber’s current US mobility gross bookings of ~$40bn. Just eyeballing it, a market in the $400-600bn range (or a ~10x’ing of the market) over ~15 years seems possible.

What % Will AVs Actually Take?

Whether AVs take 10%, 20%, or 30%+ of the market, and over what time period, is obviously of huge importance.

A big deciding factor is likely the price per mile at which AVs ultimately asymptote. For example, if AVs land at half the cost per mile of personal car ownership, you could make a case for a much higher ultimate market share (perhaps even over 50%). If the price lands in a range that is around parity with personally owned cars, it becomes harder to handicap.

One set of projections I’ve seen is from Goldman Sachs from Jul-25, which pegged the timing of the crossover in the ~2036-2038 range, so call it a decade from now. In this example, it seems GS has the price of AVs dropping materially below personally-owned, but this particular post is light on details, so it’s hard to know how they arrived at these figures.

Below I’ve done my own back-of-the-envelope analysis. I analyze the cost per mile for robotaxis at a range of price points for the vehicle (from $150k down to $20k). I compare this to the cost per mile across a range of personally-owned vehicle types (new & used 4Runners, Corollas and Priuses). I also include a theoretical personally-owned AV Prius (i.e. one that can drive itself; I assume the hardware for this is in the ~$10k range).

When I do the math, I land in a zone where robotaxis and personally owned are around parity, even at low AV vehicle price points.

Note that this analysis is laden with assumptions. The goal is to get a ballpark sense of the AV price at which AVs and personally-owned vehicles “crossover,” and how low the price of AVs could potentially get. This analysis assumes supporting infrastructure is well utilized (i.e. we’re hitting economies of scale on things like charging infrastructure). Note as well that my numbers don’t necessarily line up with the Goldman numbers above. Goldman’s current AV prices bake in high insurance rates, costs for “remote operators,” etc. I’m effectively fast-forwarding to a world where these costs are already at maturity (e.g. I don’t have any costs allocated for “remote operators;” I assume an insurance rate lower than today’s personal car insurance rates, etc.).

Based on my rough figures below, as the cost of a robotaxi dips into ~$30k range, the cost per mile will be just about equivalent to that of a personally owned AV Prius (~68c per mile vs. ~63c per mile). However, the cost per mile of a $30k robotaxi will still be above that of a personally-owned non-AV Corolla or Prius (~68c per mile vs. ~55c per mile).

This analysis is based on a host of assumptions/estimates. Some that are worth calling out in particular:

• Life of the car. Taxis drive a lot more miles, in total, than the average personally owned car. EV taxis even more so. This is a big part of what makes the robotaxi math work. In this analysis I assume a new personally owned vehicle will last 200k miles, while an AV taxi will last 300-425k miles. Note that in years, this means the taxi is only lasting ~5 years (I assume 80k miles driven per year), while the personally-owned vehicle lasts ~16 years.

• Insurance costs. For the personally-owned insurance, I take this AAA estimate for annual costs ($1,715) divided by avg miles driven per year (of 12.5k) to get to 14c per mile. For the AV car, I assume a better rate per mile. I assume a better rate per mile because Waymos are already ~10x safer than the average human driver. The rate I use is not 10x better, however, as I assume the better safety is offset by more expensive repairs and by the fact that commercial insurance rates run higher than personal insurance. For the $50k AV, I assume a rate of 10c per mile.

• Fuel/electricity costs. For the Corolla, for instance, I assume $4.00 per gallon for gas and 35 MPG (which equates to 11c per mile). For the 4Runner I assume $4.00 per gallon and 23 MPG (or 17c per mile). For the AV I assume 8c per mile for electricity. While home powered electricity is supposedly only 4-6c per mile, fast charging is considerably more (~9-15c per mile). AV taxis will sometimes need to be rapidly charged multiple times per day. They also require more power than the average EV due to the additional compute/sensor hardware. That said, fleet AVs should enjoy electricity rates that are closer to wholesale, helping to offset the need for faster charging and higher electricity demand.

• Maintenance. For the personally-owned vehicles I use a 10c per mile estimate from AAA. I assume a slightly lower cost for the AVs due to EVs generally being much cheaper to maintain. Fleet AVs should also benefit from “wholesale” repair costs vs. personally-owned which are subject to a retail mark-ups. I assume the cheaper cost is somewhat offset by the addition of expensive sensor/compute components. In all I assume a cost of 8c per mile for the $50K AV.

• Parking. This is an area where dramatic swings can happen. If you live in the average suburban home, your parking is “free.” However, if you live in a luxury condo in a dense urban core, you might pay $350+ per month for parking. If you assume 12.5k miles driven per year, that equates to 35c+ per mile. For AVs I assume they are mostly parked in relatively low-cost depots. I assume an all-in rate for parking of ~$5 per day (~$150 per month), which translates to 2c per mile ($5 per day * 365 days / 80k miles per yr).

• Utilization Mark-Up. AVs are not always occupied, of course. In fact, it’s estimated that around ~40% of miles are “empty miles.” Thus, we need a mark-up on occupied miles to recoup to cost of these empty miles. In my analysis I assume the robotaxis travel 80k miles per year and that 50k of these miles are “revenue generating” (or 62.5% of total miles).

• 20% Rideshare Mark-Up. This represents Uber’s 20% take rate that they will layer on top of the rest of a robotaxi’s cost structure.

Dara actually makes the case in a Nov-25 HD in HD podcast that once AVs get into the $50-60k range, the AV becomes a “very, very compelling economic proposition.” He states that it will likely take “a couple generations” of cars to get there.

Interviewer: “Do you think that the AV thing is going to happen kind of like little by little then all at once or it will be incremental over some period of time?”

DK: “I think the way I’d put it is little by little by little and then all at once. In that if the vehicles are cheap enough and safe enough, then the economics of AV are spectacular. And not only will they be a huge opportunity for ridesharing as it exists, they will extend the TAM hugely of the overall mobility pie. But it’ll take hardware platforms that are both safe and cheap and we’re probably two generations away from that hardware platform right now.”

Interviewer: “What’s the bottleneck right now?”

DK: “The bottleneck right now is a lot of the sensors, compute and then the software are essentially intermingled. You don’t have industry set of APIs that are formulating. As the industry matures, standards are going to come into play. So for example, we announced an effort with NVIDIA where, you know, they are building the next generation AV stack with Hyperion. And they are building their own L4 software that they’re going to essentially offer to all vehicle operators. And we will bring the network, we’ll do data collection and then we’ll run fleet management as well. And then the OEMs can focus on manufacturing cars. And if you have a future where 10 years from now, every single new car sold comes at L4 ready and is $50, $60,000. That is a very, very compelling economic proposition. But you need a couple of generations of cars and compute’s probably there, but a couple of generations of cars to get there.”

The big hurdle, of course, in getting to a $50K AV is in the hardware costs (compute, sensors). Encouragingly, the cost of AV hardware has been falling rapidly.

From Q2’25 earnings:

“The real question for us is the cost of the vehicles, the cost of hardware coming down. And again, we’re seeing very encouraging trends there. LiDAR, which used to cost $20,000, $30,000 a pop, now solid-state LiDAR coming out of some of the Chinese manufacturers is costing $300, $400 per LiDAR as well. So the affordability of hardware is moving absolutely in the right direction.”

Here is Alex Kendall (Founder, CEO of Wayve) from a Nov-25 Training Data podcast:

“The industry has really, outside of Tesla, coalesced around a common architecture of a surround camera, surround radar, and a front-facing LiDAR stack. Now, this costs under $2,000. It’s automotive grade components, not the retrofit robotaxi components you see today. But having frontier GPU compute, automotive grade GPU on the car and that kind of sensor architecture is a really great platform to to build L3/L4 autonomy, eyes off or driverless. It gives you the necessary redundancy. It lets you deal with edge cases that — you know, cameras alone, I agree they can get you to human level but we want to go beyond human level — and so I think this kind of architecture is affordable, scalable, it’s got the supply chain for mass manufacturing and it can…drive superhuman levels of performance. So that’s what we’re seeing many manufacturers bring out on their vehicles and where we’re integrating our AI.”

Another hurdle is eliminating the “retrofit” costs of adding AV hardware to a vehicle after it has already been produced. The Jaguar I-Pace Waymos, for example, were retrofitted with all the necessary hardware for L4 autonomy. Increasingly (including for the Nuro/Lucid Gravity robotaxis), these components will be integrated into the vehicle during normal-course production (or “in-line” production). This will introduce additional cost savings.

If AVs Achieve Price Parity Only, What Market Share Can AVs Take?

In this case, whether you are a car owner or not may depend a good deal on where you live. If you don’t have kids and live in a dense urban core where you are paying 35c per mile for parking and it’s painful to drive around, the decision to go rideshare-only is more straightforward. If you have kids and live in the suburbs, you probably still own at least one car (for the car seat, the convenience of being able to keep stuff in the car, etc.).

Here is Gemini’s break down of where people live in the US based on population density:

Perhaps we assume 50% of the “urban core” miles are robotaxi, 25% of the “dense transit-adjacent,” 10% of the “auto-centric suburbs” and 0% of rural. This would equate to 15% of total miles being robotaxi.

If 15% of miles are robotaxi, and we assume a price per mile of $1.25, that equates to a ~$550bn market size (in US mobility alone).

How Much do AVs Cost Today?

Uber ballparked the cost of an all-in AV at around $200k in 2024.46

In Feb-26, Waymo struck a deal with Hyundai that was rumored to be $2.5bn for 50k IONIQ 5 vehicles. That comes out to a price tag of $50k per vehicle. The IONIQ MSRPs in the range of $35k. So the assumption is that the additional $15k per vehicle was for all the necessary autonomy hardware. If this were the case, then we are actually almost at Dara’s $50K bogey today.

When Uber recently expanded its Nuro/Lucid partnership to include an additional 15k minimum order of Lucid’s new midsize vehicle, the PR quoted a price of “less than $50k” as a starting price for the vehicle. The way the PR is phrased, the $50k seems to exclude AV hardware costs. But if those are in the $15k range, then Uber would be looking at a ~$65k AV.

Further, as previously mentioned, Chinese-built AVs may already be well below $50k. BYD’s latest Seagull EV, equipped with LiDAR and a “mid-tier” smart driving system, retails for just ~$13,400. Baidu’s RT6 Robotaxi (manufactured by JMC) has supposedly carried a sub $30k price tag since May-24.

What ultimately gets a US-built AV to a sub $30k level might be innovation on form factor. For example, Tesla’s Cybercab, which it claims will retail for ~$30k, is stripped down and purpose-built for autonomy. It is a two-seater with no steering wheel, no mirrors, no pedals, a smaller battery, etc.

If I had to guess, if we exclude Chinese manufacturers, I’d say we are ~4-6 years away from AVs (with super human safety) coming off the line at a ~$30-40k price point. This would be in line with (perhaps slightly below) Dara’s estimate of “a couple generations away.”

Competitive Landscape

Waymo

Waymo was founded in 2009 as a “moonshot” experiment within Google’s X lab. The company was spun out in 2016, but is still majority owned by Google.

In Jan-26, the company closed a $16bn round of financing at a $126bn valuation. Investors in the round included Dragoneer, Sequoia Capital, DST Global, Andreessen Horowitz and Mubadala (Abu Dhabi’s sovereign fund). Waymo was reportedly doing north of $350mm in annualized revenue at the time of the raise (in Waymo’s case, revenue and bookings are going to be equivalent).

Waymo launched its commercial service in Phoenix in 2020. In Aug-23, it launched in San Francisco, it’s second official commercial market. The company now operates in ~15 markets: San Francisco, Los Angeles, San Diego, Las Vegas, Phoenix, Denver, Austin, Dallas, San Antonio, Houston, Nashville, Atlanta, Tampa, Orlando, and Miami. The company reportedly had a total ~3,000 vehicles in its fleet as of Mar-26.

Waymo’s vehicles are classified as level four autonomous. They don’t require a driver or active supervision. By all accounts, in terms of safety, they compare very favorably to human drivers. From the WSJ:

“Swiss Re, one of the world’s premier reinsurance companies, analyzed 25 million miles of Waymo driving. It concluded these vehicles are safer than human drivers, with 92% fewer bodily injury claims and 88% fewer property damage claims. A peer-reviewed study examined 56.7 million rider-only miles and found an 85% reduction in suspected serious injuries compared with human drivers and a 96% reduction in intersection crashes.”

These stats suggest Waymos are roughly 10x safer than the average human driver. Waymo’s Co-CEO, Tekedra Mawakana, had the following to say at the Nov-25 TechCrunch Disrupt conference, suggesting Waymos were ~5x and ~12x safer when it came to accidents with other vehicles/property and with pedestrians, respectively:

“We’ve driven over 100 million RO [Remote Operations] miles, but we had a study at 96 million RO miles. And it’s actually been determined that the Waymo driver is five times safer than a human driver and 12 times safer than a human driver as it relates to pedestrians.”

Waymo is expanding quickly, growing its rides per week by 10x over the past 2 years. The pace of city expansions, in fact, has increased dramatically. Early markets took years to advance from initial mapping to on-road autonomous testing to a pilot with riders to a commercial launch. More recent city launches like Orlando have condensed this timeline to <1 year.

The company is targeting 1mm rides per week and to have ride-hailing operations in over 20 cities (including two international cities: Tokyo and London) by the end of 2026.

In Jun-26, Waymo introduced “Waymo Premier,” a subscription product priced at $29.99 per month. Waymo Premier allows riders to (1) access priority pickups, (2) earn 10% Waymo Cash back on all trips, (3) access new cities early, and (4) cancel up to five trips monthly for free.

As referenced previously, the Waymo service, at this point, is generally more expensive and ETAs are longer than traditional rideshare. The Waymo fleet is fixed and thus can’t “flex up” like Uber’s can to meet demand during peak hours. As a result, Waymo must choose whether to carry enough cars to meet peak-ish demand (and thus be forced to idle more cars during troughs), or whether to carry less cars to drive better utilization (knowing this will result in higher ETAs during peak hours). It appears Waymo has chosen the latter approach for now. It will likely continue to be challenging for Waymo to offer low ETAs during peak hours.

Waymo’s fleet is currently almost entirely Jaguar I-Pace’s. The company is in the process of introducing the Ojai from Chinese carmaker Zeekr (a subsidiary of Geely). The Ojai is a small minivan/SUV, and will include Waymo’s sixth-generation AV technology. The sixth-generation sensor array consists of 13 cameras, 4 LiDAR sensors and 6 radar sensors. This is a material reduction to the fifth-generation suite which consisted of 29 cameras, 5 LiDAR sensors and 6 radar sensors. According to this FT article, analysts pegged the cost of a Jaguar I-Pace at ~$150k each. Waymo announced in May-25 that it had received the last I-Pace that it would add to its fleet. In all, Waymo ended up taking delivery on ~3.5k I-Pace’s (vs. initial plans for as many as 20k). Waymo is also planning to outfit South Korea’s Hyundai IONIQ 5 SUV with its sixth-generation AV technology.

In mid-2025, Waymo opened a large, 239k sq ft production facility in Mesa, AZ in partnership with Magna International, an automotive supplier. The facility will be capable of upfitting “tens of thousands” of vehicles per year, and will reportedly handle both the Ojai and the upcoming IONIQ 5. In the case of the Ojai, most of the production is done in China (i.e. the platform, structural elements, powertrain). The Mesa facility takes this “glider” base vehicle and adds the autonomous outfitting and performs final assembly. While the Mesa plant does not represent pure “in-line” production (which would be the cheapest form of AV production), it constitutes a step in that direction.

While Waymo has shown nothing other than an accelerating roadmap in its 1P robotaxi service, when asked about their future business model, they do maintain that first and foremost, they are building a “generalizable driver.”

Here is their Co-CEO Tekedra Mawakana from the Nov-25 TechCrunch Disrupt conference:

“Our business model is the same. We’re building a generalizable driver. That driver, through our ride-hailing service, is our first application. Then local delivery. Then long haul trucking. And then eventually we will license this technology to automotive companies for personally owned cars. That’s always been our road map, and it’s still our road map.”

She has reiterated this idea in multiple interviews. And Waymo actually has a partnership with DoorDash, delivering food in Phoenix.

Here is Sundar Pichai on a Jun-25 Lex Fridman podcast also emphasizing Waymo’s focus on the Driver tech (and its “generalizability”):

“We don’t compete with Tesla directly, we are not making cars, etc. We are building L4, L5 autonomy. We’re building a Waymo Driver which is general purpose and can be used in many settings. They’re obviously working on making Tesla self-driving too. I’ve just assumed it’s de facto that Elon will succeed in whatever he does. So like, you know, that is not something I question. So but I think we are so far from — these spaces are such vast spaces — like I think about transportation, the opportunity space, the Waymo Driver is a general purpose technology. We can apply in many situations. So you have a vast, green space. In all future scenarios, I see Tesla doing well and Waymo doing well.”

Here is Alex Immerman (a16z lead investor on Waymo) on a Mar-26 Sourcery podcast, with some slight hedging re whether Waymo will, in the end, choose to operate a fully blown rideshare service:

“I think where margins land over time is an interesting question. On one hand, Waymo could go the route that they are today and own and operate cities like in San Francisco and Los Angeles. They own the experience with Waymo One. They are managing the operations and the depots. As an alternative, they could outsource some of that as they are in Phoenix, Atlanta and Austin, where third parties like an Uber or a Moove are managing the operations. When you outsource that, and those are real costs, that could lead to a higher margin product. This is never going to be like high margin software, but it’ll be better than operating the fleet themselves.”

David Risher, the CEO of Lyft, has expressed some doubt around whether Waymo will ultimately want to build out a global robotaxi service. In his mind, as the leader in the space, Waymo is still in an experimentation phase. From a Nov-25 Decoder interview:

Interviewer: “Let me ask you about Waymo, and then I want to kind of zoom out again. You’re Waymo’s partner in Nashville. Waymo’s partnered with Uber in other markets. Waymo’s running their own service in some markets. It does feel — and you know, they’re obviously backed by Alphabet, they’ve been backed by Alphabet for a long time. They’re going to spend a lot of money to win. It just seems very obvious. If you asked Sundar about it, he is like, we’re just going to keep spending money because now it’s very clear that we’re close to winning. They’re going to spend a lot of money until they win. They’re looking for partners, right? They’re looking to see what winning looks like. You’re obviously — you’re in a kind of a weird competition, right? You’re in a weird kind of bake-off. You’re the partner in one market, they’re the partner in another market. They’ve got their own service. How do you perceive that competition? Have they told you what winning looks like?”

DR: “You know, I think they’re figuring it out. I really do. And I take them at their word that they... Look, I think a word that they use pretty often is ‘optionality,’ right? They have something pretty cool, right? They’ve got technology that works about as well as anyone’s in the world. You know, again, I think really only Baidu would be the real competition there. I think everybody else is some, you know, degree off of where they are. And it works and it works well and people like it. So that’s a good place to be. Thus, if you’re in that position and you think, ‘Well, gosh, I’m not really sure how this is all gonna play out,’ a totally reasonable strategy is, well, let’s try a little bit of everything, right? Let’s try doing it ourselves end-to-end, let’s try partnering in a certain way with one company, let’s try partnering in a different way with another company, and we’ll kind of see.

When I fast forward, I think a very likely outcome is they will realize, ‘Gosh,’ as I just said, ‘Running a rideshare business is quite expensive and it’s very physical.’ It involves, like, again, think of the scale, 800 million rides we do every single year, 50 million riders. You know, 1 1/2 million drivers — I know those aren’t part of the picture, but someone’s got to own those cars. There’s gonna be someone else in this picture who’s owning these things and wants these things utilized. You know, if you’re Google, do you want customer service? Do you want all these different things that you have to do to operate that service? Maybe you don’t mind it in a couple of markets — maybe it’s kind of cool, you can have direct access to your customers and you can do sort brand building and maybe get some data from it, whatever. All good. But do you really wanna do it in 280 cities around the United States or all around the world? I don’t know. So I think a very likely outcome is they will be, you know, frenemies, right? Or what’s that word, co-opetition or whatever. Like they’ll compete in some markets and in others they’ll partner. And our job is to be the best partner they can possibly have so that they, over time, give us more of their business and stay focused in their own way in some small number of markets.”

If Waymo were to only license a generalizable driver, could they still grow into their $126bn valuation? Perhaps they are locked into “going for it” as a byproduct of their valuation?

Based on some back-of-the-envelope math, it seems reasonable to think Waymo could grow into its valuation with just a generalizable driver. There are about 100mm passenger cars + light commercial vehicles sold annually around the world. If you assume AV tech can be licensed to all of them (a stretch over the near term, of course) for $2,500 per car, that would equate to a $250bn market. If Waymo could take 10% of that, you’d be looking at $25bn in revenue. That would generally support its current valuation (though it’s also a long ways off).

However, it’s also a good bet that smaller, on-demand delivery vehicles will also proliferate over this timeframe. That might unlock another tranche of vehicles into which Waymo could sell AV tech. Long-haul trucking is another clear opportunity. Robots, or even drones, could be other opportunities for Waymo (though different from driving, they are areas where Waymo’s “world models” could be leveraged).

Alex Immerman suggested on the Sourcery podcast that a16z, at least, views the future of AV software as most likely a “productive oligopoly”:

Interviewer: “From a deal perspective, I want to unpack the process. How do you underwrite a company like this? Do you look at Tesla?”

AI: “So this is one where you have to think about a variety of scenarios off into the future. So to make it simple, I would frame it as there are three like base scenarios. The first is Waymo stays the only game in town. I certainly, like, hope that is the case. They have dominant market share of a market that grows quite significantly.

The second case is the market still grows significantly because it’s such a great value proposition — self-driving. However, there are multiple players. That might be a Tesla, it could be a Zoox, it could be a Wayve. There are Chinese players as well. But because of the technical challenges, I think it’s a productive oligopoly.

And then the third scenario is where things don’t work out as well for Waymo and, you know, maybe there’s impairment. Maybe one of those other providers has an advantage over them in the fullness of time.

I think the most likely scenario of those is that it is a productive oligopoly. That there are many players but we expect Waymo to be a meaningful share of that overall market.”

Reading between the lines, this would seem to suggest that a16z also believes that just Waymo’s generalizable driver tech could eventually support its current valuation.

Tesla

Tesla is a number of things these days. It’s of course an automaker that builds EVs. Its bestsellers are the Model 3 (sedan) and the Model Y (SUV), which together dominate EV sales in the US.

Tesla also sells a Class 8 electric truck called “The Semi.” The company runs Tesla Energy, which builds the Powerwall and Megapack batteries, and installs solar roofs. It is building the Optimus humanoid robot. They operate huge AI training clusters (for their self-driving efforts and training the Optimus). Tesla is even building its own semiconductor “fab,” or fabrication plant, in partnership with SpaceX and xAI that it has dubbed “Terafab.” The Terafab will cost an estimated $20-25bn and build Tesla’s fifth generation AI chip, with volume production targeted for 2027, in theory.

In 2025, the company sold ~1.6mm vehicles and generated $95bn in revenue. It ended 2025 with $44bn of cash on the balance sheet and $8bn of debt.

For our purposes, we will focus on Tesla’s efforts in self-driving and its Robotaxi initiative. Tesla has been working on self-driving tech since 2013. In 2014, Tesla began installing a single camera and radar system (developed by Mobileye) on its Model S. In 2015, Tesla launched “Autopilot,” which allowed cars to steer themselves on highways, a first of its kind “L2” system. In 2016, Tesla moved to Hardware 2 (HW2) which included the use of NVIDIA chips. In Oct-16, management famously promised that every car produced from that day forward had all the necessary hardware for full autonomous driving. In 2019, Tesla moved to a custom chip designed specifically for neural networks. In 2023, Tesla moved to “Version 12” of its self-driving software (which it used to refer to as FSD, for “Full Self Driving” — now it’s “Full Self Driving (Supervised)”). This was a step-change event, as it replaced hundreds of thousands of lines of human-written code with an end-to-end neural network.

In Oct-24, as Tesla started to approach L4 autonomy, it unveiled its “Cybercab.” The Cybercab is a purpose-built “robotaxi.” It has no steering wheel, pedals, or side mirrors. It also has no NACS charging port. Instead it’s charged “inductively” by parking over a charging pad. It has a relatively small 35 kWh battery that is optimized for city driving, and carries just ~200 miles of range. The Cybercab’s sensor suite is estimated to cost roughly $400. While Tesla will own its initial fleet of Cybercabs itself, they claim that third parties will be able to buy them starting in late 2026 for “under $30k.”

In Apr-26, the Cybercab officially went into mass production at Tesla’s new “Giga Texas” facility. Giga Texas is unique in that it employs an “unboxed” process. Instead of using a traditional linear assembly line, the car is built in separate “sub-modules” (the floor, the sides, the roof) which are then “snapped together” at the end. This has supposedly reduced the factory footprint by 40%. Giga Texas also has no paint shop. The Cybercab uses colored polyurethane body panels, where the color is “molded in.” The paint shop is typically one of the (if not the) most expensive and time-consuming parts of the car manufacturing process. Musk has said that Tesla intends to ultimately manufacture two million Cybercabs annually.

Tesla’s Robotaxi operation went live in Austin and the SF Bay Area in Jun-25 and Jul-25, respectively. It added Dallas and Houston in Apr-26, and Miami in Jul-26.

According to Morgan Stanley research, Tesla’s Cybercab should already be in the ~$0.74 cost per mile range vs. Waymo in the ~$1.36 range, and rideshare in the ~$1.71 range. Personal car ownership, according to Morgan Stanley, is in the $0.71 range.

While the Robotaxi operation is “live,” it lags majorly behind Waymo in a number of different ways.

For starters, the vast majority of the cars in the fleet are supervised — there is a safety driver sitting in the driver’s seat ready to take over if the car encounters issues. Waymo has no safety drivers. Neither does Zoox, Avride, WeRide, etc.

Secondly, the Robotaxi fleet seems to be having a lot of incidents, based on the anecdotal evidence as well as reported crash figures. Anecdotally, you have stories like this one:

I covered this previously — it’s a lot of issues for one trip. Here is a ride from Houston where the Robotaxi doesn’t make its exit (unclear why) and has to circle back 3 times (while performing 2 illegal U-turns). Someone actually compiled a whole list of incidents at one point.

In terms of the accident data, it’s hard to calculate exactly how frequently Tesla’s Robotaxis are crashing. From Jul-25 (when Tesla’s Robotaxi fleet began operating in Austin) through mid Mar-26, Tesla reported 18 crashes to the NHTSA, all in Austin. You can see the data here, under “ADS Incident Report Data.” The problem is we don’t know the number of miles driven by the corresponding Robotaxi fleet. This would give us the incident rate (or accident per miles driven).

We do have this chart from the company’s Q1’26 earnings deck, showing us Tesla’s cumulative paid Robotaxi miles:

However, these figures likely include miles driven in the SF Bay Area (where Tesla’s been operating a commercial service since Sep-25). As mentioned above, there are zero accidents reported in the SF Bay Area (or anywhere other than Austin), so I’d guess the SF Bay Area accidents are not required to be reported (perhaps because in the SF Bay Area, the fleet operates under a limo/private charter permit and always carries a safety driver — thus it’s more of a chauffeured service).

Furthermore, the Robotaxi fleet in the SF Bay Area is actually much larger than the fleet in Austin according to Robotaxi Tracker. It seems the active fleet in the Bay Area has been something like ~4x larger than Austin’s on average since it launched. If we conservatively assume the miles traveled in Austin is 25% of the total cumulative miles (implying only 3x the number of cars in the Bay Area), that would mean ~375k miles were traveled in Austin. That would imply an incident rate of 18 / 375k miles or 1 incident per ~21k miles. If we assume that there were simply no crashes in the Bay Area, or that these cumulative miles are only Austin miles, then the incident rate would be 18 / 1.5mm miles or 1 incident every 83k miles.

In comparison, according to the NHTSA, the average human driver has 1 (police-reported) crash every 500k miles. Because this figure is based on police-reported crashes, the actual incident rate will be higher. Tesla’s own Vehicle Safety Report claims the average driver has a minor collision every 222k miles and a major collision every 660k miles:

Thus, it seems at best the Tesla Robotaxi is ~3x worse than the average human driver (1 incident every ~83k miles vs. 1 incident every ~222k miles). More likely, however, is that Tesla’s Robotaxi is more like ~10x worse (1 incident every ~21k miles vs. 1 incident every ~222k miles).

Making this even worse is the fact that the vast majority of Robotaxi trips to date have been supervised (i.e. there have been safety drivers behind the wheel). Thus, we can only assume that incident rates would have been even worse — perhaps much worse — if these cars had actually been fully autonomous.

This says nothing of:

1. Robotaxi reportedly features a “hidden backbone of teleoperators who can pilot the vehicles remotely when needed.” “Remoting in” to drive a vehicle is considered highly unsafe. Waymo and others are very clear that while customer support can be reached to help with certain things, they never drive the car remotely. Driving the car remotely is highly dangerous due to latency issues and potential gaps in cellular service.

2. Existing Teslas, which Musk claims can be added to the Robotaxi fleet with the touch of a button, may not have the necessary redundancies to be AVs at all. To be an L4 AV, a vehicle needs overlapping sensors, more than one power source, redundant braking and steering, and two compute stacks. Thus, if any of these “points of failure” fail, the car can still drive (and safely pull over). According to this article, Tesla’s Models S, X, Y and 3 may not have this type of redundancy: “Because Tesla is Tesla, it doesn’t share a lot of information about what’s underneath the skin of its vehicles. Independent reporting and teardowns have turned up some information about redundant systems. The full steer-by-wire system used on the Cybertruck features redundant power and controls, with two motors incorporated into the steering rack, likely fed by separate power sources. It’s a good bet that a similar system is used on the controls-free Cybercab that Tesla plans to deploy as part of its Robotaxi service. Note that redundancy like this is not present on the Models S, X, Y or 3 that came before the Cybertruck. And without such backups, many of the Full Self-Driving promises simply could not have been fulfilled safely.”

3. In order to sell the Cybercab, Tesla needs approval from the National Highway Traffic Safety Administration (NHSTA), as the Cybercab lacks a steering wheel, pedals and side mirrors (i.e. the Cybercab is not compliant with NHSTA regulations). NHTSA grants exemptions to automakers to sell vehicles not fully in compliance, but caps these sales at 2,500 per year. According to the WSJ, NHSTA indicated that as of Mar-26, Tesla hasn’t yet applied for an exemption on the Cybercab.

4. Elon’s long list of false promises re self-driving

Tesla’s “Achilles heel” when it comes to self-driving may be its insistence on a camera-only sensor suite. The Robotaxi and the latest Model 3s / Model Ys come equipped with 8 cameras, but no radar or LiDAR. Elon’s view seems to be that because humans drive with their eyes only, a machine with 8 cameras and a neural net should be able to drive just as well, if not much better. The problem with this, as discussed earlier, is that (1) we don’t know how long it will take to get this to work and (2) this doesn’t seem like a recipe for “superhuman” safety and could result in a substantially less safe product than Waymo and others.

Here is Dara from a Mar-26 HD in HD podcast:

DK: “I think Tesla is — they have gotten the hardware bill of materials to an industry-leading level in the US. The Chinese OEMs can match it and then some. But in the US, in terms of bill of materials, Tesla’s a leader. What they haven’t yet proven is that a camera only system can be safe enough as it relates to AV. And you know time will tell on that one.”

Interviewer: “Why do you think that that’s the hill Elon dies on?”

DK: “I think that he thinks from a first principals standpoint and fact is human drivers have two eyes and they can drive. So why can’t a car which is a computer with wheels and that has eight eyes, do the same? Now, eventually that sounds right. But the question is over the next 5 years, over the next 10 years as AV is developing, are they going to get there quick enough? I think the other thing that probably has taken some folks by surprise is that the cost of LiDAR and radar — the cost for LiDAR has come from like $40,000 down to a couple hundred. So the hardware curves have come down probably faster than the industry expected.”

Interviewer: “So why not just add a LiDAR there?…”

DK: “Yeah, I can’t speak for him. I mean, the one thing I would say is Tesla is economically motivated to make L3 plus or L4 work for the fleet that they already have on the ground. So, that could be a factor.”

As mentioned above, in Oct-16 Tesla began promising that all of its cars manufactured with HW2 had all the hardware necessary to achieve L4 autonomy. Tesla would eventually have to walk that back. Then, more recently, in Q1’26, it had to walk back that cars manufactured with HW3 would be able to achieve L4 autonomy. On the earnings call, management floated the idea of setting up “micro factories” around the world that could swap HW3 with HW4.

“Unfortunately, Hardware 3 -- I wish it were otherwise, but Hardware 3 simply does not have the capability to achieve unsupervised FSD. We did think at one point it would have that, but relative to Hardware 4, it has only 1/8 of the memory bandwidth of Hardware 4. And memory bandwidth is one of the key elements needed for unsupervised FSD. It’s just generally a thing that’s needed for AI. If you’re doing an autoregressive transformer, memory bandwidth is the choke point.

So for customers that have bought FSD, what we’re offering is essentially a trade-in -- like a discounted trade-in for cars that have AI4 hardware, and we’ll also be offering the ability to upgrade the car, to replace the computer. And you also need to replace the cameras, unfortunately, to go to Hardware 4.

So to do this efficiently, we’re going to have to set up like kind of micro factories or small factories in major metropolitan areas in order to do it efficiently. Because if it’s done just at the service center, it is extremely slow to do so and inefficient. So we basically need like many production lines to make the change. And I do think, over time, it’s going to make sense for us to convert all Hardware 3 cars to Hardware 4 because that’s what enables them to enter the Robotaxi fleet and have unsupervised FSD.”

If Tesla ultimately admits that they cannot achieve L4 without radar/LiDAR, then Tesla would have to somehow remedy this, which could be quite painful. Thus Tesla has boxed itself in, to a degree, to making L4 work with a camera-only sensor suite.

This clip from Rivian’s autonomy day does a nice job of explaining the power of a “multi-modal” sensor suite that includes cameras, radar and LiDAR. While the cameras are undoubtedly the workhorse sensors for autonomy, they struggle in “non-ideal” lighting conditions such as low light, excessive light and fog. In such cases, radar and LiDAR can help fill in the gaps. Radar and LiDAR also provide richer, more accurate data around the depth and velocity of objects that surround the car.

The video below of a Tesla hitting a deer while in FSD mode shows how challenging it may be to achieve self-driving with cameras only (this is from Oct-24 granted, so it’s “dated”). It’s dark, the deer is only visible for a few seconds before impact (with the car on the highway, and accelerating, it would’ve been impossible to avoid impact at all, even for a human driver), it’s mostly the deer’s butt showing (making it harder to recognize) and it somehow blends in with a patch of unpaved road on the highway.

The Tesla software simply doesn’t recognize the deer at all (the car reportedly didn’t slow down at all before impact). However, in theory, every other AV player (using radar/LiDAR) would have clearly seen (before the headlights even shined on the deer) that a meaningful object was in the road and avoided an accident.

This example demonstrates how Tesla may be fundamentally constrained from achieving a superhuman level of driving by not adopting radar/LiDAR. We also have the fact that literally nobody else is building a cameras-only AV at this point, which suggests that everyone else thinks this is the worse approach.

Elon (to his credit?) on Tesla’s Q1’26 earnings call, basically all but admits that the company’s self-driving tech is still a ways off:

Questioner: “The next question is, is v14.3 still the last piece of the puzzle to enable large-scale unsupervised FSD and Robotaxi? Or do we have to wait until V15?”

Elon Musk: “Well, I think 14.3 is last piece of the puzzle for unsupervised FSD. Now the question is like degrees of safety. Like how — safety and convenience, I suppose. We have a lot of known improvements — like major architectural improvements that we know would improve the probability of safety significantly. So I think it’s not going to make sense for us to deploy unsupervised FSD or Robotaxi at large scale when we know that there are major architectural improvements to the software that can improve safety.

So I think we’re going to want to finish writing that software, validate it and release it before going to large-scale unsupervised FSD. Depending on what large scale means. I mean we are, of course, as I mentioned earlier, doing unsupervised FSD in 3 cities, and we’ll expand to, like I said, probably a dozen states or more later this year. So it kind of depends on what your definition of large scale is. But I do think it wouldn’t be right for us to go to like very large scale unsupervised FSD when we know that there are software improvements in the pipeline that would improve safety.”

All together, Tesla seems not ready to deploy a large-scale, unsurprised fleet of robotaxis for some time. I’d say they are officially behind Waymo, WeRide, Pony.ai and Apollo Go. I think it’s likely Tesla also falls behind at least NVIDIA as well over the coming years.

Here is Uber’s official position on Tesla’s L4 timeline from CFO Balaji Krishnamurthy from a Feb-26 interview with The Information:

“I would just come back to where they are today which is based on [Tesla’s] publicly reported data, their incident rate is effectively where Waymos’ incident rates were in Q1 of 2023. So on that basis, it looks like they’re about 3 years behind. And you know we’re not saying that Tesla is not going to get there. It’s entirely likely and possible that they would be there. But in that time there are going to be other players who will also deliver commercially deployable autonomous solutions and we have the broadest set of partnerships, right? We have 20 partners. We are looking forward to bringing them to the 15 markets we’ve talked about this year and that number is only going to go up in the coming years. And on our platform the utilization and revenue generation capacity is going to be higher than any anyone else’s. So we’ll compete on that.”

From Uber’s standpoint, Tesla looks to be ~3 years behind Waymo. This seems about right.

Autonomous Transition

The Peak-to-Trough Issue

Rideshare demand is, of course, variable. It varies a lot over the course of a day — peaking during afternoon rush hour and troughing in the middle of the night. It varies over the course of a week — Friday generally sees peak demand, and Sunday trough demand. From Uber’s Q4’24 AV Spotlight:

Demand also varies over the course of a year — it generally peaks in the spring and fall, while summer tends to be relatively slow.

Thus, as an AV-only provider, you have an issue. Do you carry enough cars to fulfill demand at peaks (but then leave many cars idling during troughs) or do you limit your fleet size to keep cars generally better utilized (but deliver poor ETAs during peak periods)?

Here is commentary from Uber’s Q4’24 supplemental:

“Finally, assuming that all four of the aforementioned commercialization issues have been addressed, AV operators will face another daunting challenge: the highly variable nature of ridesharing demand, which has large peaks and troughs throughout the day, week, month, and year. Any standalone player with a fixed depreciating asset will need to build against that reality: choosing between running a highly underutilized network (if supply is built for peak demand) or a highly unreliable network at peak periods (if supply is built for anything less than peak).

For instance, in a typical large city, a fixed fleet designed to meet the weekly peak will have up to 95% of vehicles idle during the multiple weekly troughs. Conversely, a fleet sized below peak cannot deliver the reliable 4-minute ETAs that consumers expect. This dilemma is compounded by the fact that vehicle supply will need to vary dramatically throughout the year: in many North American cities, demand peaks around March and trends down through the summer, only rebounding in the fall. Weekly peaks may be 50%+ higher than the lowest week of the year. On top of that, supply will need to constantly grow to meet overall industry expansion (Uber is still growing near 20% YoY), the trajectory of which is non-linear. This could mean that a fixed network that is well supplied in March may find itself idling a third or more of its vehicles for several months, until Halloween. Because demand patterns look similar (although not identical) between cities, underutilized vehicles cannot be easily repositioned and will likely sit unused for months, generating additional cost and complexity—against zero revenue.

Given the scale of the Uber platform, and human drivers’ ability to dynamically fulfill demand spikes—and take a break during demand troughs—partnering with Uber allows AV players to move much faster than they could on their own. This fact gives us confidence that the Uber network, with a hybrid of AV and human drivers, will deliver the highest asset utilization and revenue generation opportunity for our partners.”

In the geos in which it operates, Waymo has chosen to keep its fleet relatively constrained so far. This helps drive a better utilization on its vehicles (take it from Bill Gurley):

But it also, as Alex Immerman (an investor in Waymo) points out, allows Waymo to take advantage of some physical world first-mover advantages.

The optimal level of AVs to carry, I believe, is around ~1.0x average demand. If you go below that, say optimizing for 0.4x demand, your fleet will be better utilized (during troughs, the majority of your fleet would still be utilized), but (1) you’d leave a lot of money on the table and (2) customers would switch to providers with better ETAs. If you go above that, say optimizing for 1.4x demand, the majority of your fleet would sit idle during troughs, which would spoil your ROICs.

How Close are Uber’s Partners to L4?

Technically already crossed the L4 threshold:

• Apollo Go. Apollo Go is owned by Baidu, the Chinese internet/search giant. Apollo Go is the largest operator currently in China with more than 1,000 driverless vehicles in operation. Their weekly ride count surpassed 250k in Feb-26. The company is in the process of expanding into the Middle East, Southeast Asia and Europe (it already operates a driver-out commercial service with Uber in Dubai and Abu Dhabi). It hopes to have 20,000 robotaxis operating worldwide by 2027, including more than 1,000 vehicles in Dubai by early 2028. Baidu’s RT6, built by Jiangling (a Chinese carmaker), reportedly costs just $30k. In H2’26, Apollo Go is slated to begin testing a robotaxi pilot in London (with the RT6) in partnership with both Uber and Lyft.

• WeRide. WeRide is a public company listed on Nasdaq and based in Guangzhou, China. As of Jan-26, the company had more than 1,000 vehicles operating worldwide. It hopes to grow that figure to 2-3,000 by the end of 2026, and tens of thousands by 2030. It aims to have between 500-1,000 robotaxis operating in the Middle East alone by the end of 2026. WeRide already operates a commercial robotaxi service on behalf of Uber in Dubai, Abu Dhabi and Riyadh. In Jun-26, WeRide and Uber announced that they would launch a commercial robotaxi operation in Madrid, their first joint partnership in Europe. The service is slated to begin operations before the end of 2026. The two companies have outlined a partnership to bring joint operations to 15 cities by 2030 (with Madrid marking the fourth such city).

• Pony.ai. Pony, like WeRide, is also a Nasdaq-listed public company based in Guangzhou. Pony had ~1,000 vehicles in operation at the end of 2025. They are targeting 3,000+ by year-end 2026. The company has formed a partnership with Stellantis to deploy AVs in Europe, and launched Europe’s first commercial AV service with Uber in Zagreb, Croatia in Apr-26 (using cars from Chinese automaker Arcfox). The ODD was initially ~35 square miles across the Zagreb city center, and including the Zagreb airport. In Apr-26, Pony announced that it had switched to driverless in Dubai (commercial service targeted for H2’26). They also announced that their next-generation robotaxi in China will cost ~$33k, including the cost of AV hardware. Further, the company has a partnership with Toyota. The two companies are targeting the deployment of 1,000 Toyota bZ4X Gen-7 robotaxis by the end of 2026 in China.

Probably get there soon:

• Nuro. Nuro was founded in 2016 by two ex-Waymo engineers — Jiajun Zhu (“JZ”) and Dave Ferguson. The company is based in Mountain View, CA, and has raised a total of ~$2.2bn since its founding. Its original strategy was to build vertically integrated delivery bots, and it signed partnerships with Kroger, Domino’s, CVS, 7-Eleven and Uber. In 2024, however, in the face of a dearth of funding options and a very capital intensive business plan, they pivoted to a strategy of licensing the Nuro Driver to car OEMs/the robotaxi market. Nuro most recently raised at an $6bn valuation in Aug-25. Recent investors have included Uber, Baillie Gifford, NVIDIA, T. Rowe Price, Fidelity, Tiger Global and Greylock. In Jul-25, Nuro announced a joint partnership with Lucid and Uber to deploy 20,000+ Lucid vehicles with the Nuro Driver on Uber across “dozens” of markets. In Apr-26, Uber increased its purchase commitment to at least 35,000 Lucid vehicles (both commitments are based on Nuro/Lucid hitting certain milestones). Nuro has one of the strongest safety records in AV tech. The company has logged over 1.7mm autonomous miles on public roads with zero at-fault incidents or injuries. As mentioned previously, the Uber/Lucid/Nuro commercial service is slated to launch in H2’26 across the SF Bay Area. It stands to be Uber’s first robust, driverless commercial offering outside of Waymo in the US. It should be noted that Toyota’s VC arm, Woven Capital, is also an investor in Nuro. In Mar-26, it was reported that Nuro was testing its autonomous software in Tokyo with the Toyota Prius. A partnership with Toyota, the largest carmaker in the world by far, would clearly be a big win for Nuro (Waymo, it should be noted, has also been testing with Toyota). According to Gemini, Nuro has been primarily testing in Mountain View, Palo Alto, the broader SF Bay Area and Houston. According to Gemini’s review of Nuro’s twitter account, the company has been on “road trips” to Los Angeles, Austin and Phoenix. This may suggest that these cities are in line for 2027 deployments.

• Wayve. Wayve was founded (and spun out of the University of Cambridge) in 2017 by Alex Kendall and Amar Shah. The company is based in London and has raised a total of $2.5bn over its lifetime. Wayve is considered one of the pioneers of “AV 2.0.” The company employs a giant neural net that learns to drive purely from raw data inputs (i.e. there’s no rules-based logic programmed into their driver), similar to how a LLM learns language. Further, the technology does not rely on extremely detailed, high-definition maps like Waymo. It employs a “mapless” approach that can adapt to a new city or vehicle in weeks, supposedly. In Feb-26, Wayve raised $1.2bn at a $8.6bn valuation from investors including Mercedes-Benz, Stellantis and Nissan. Investors from prior rounds include Uber, NVIDIA, Microsoft, SoftBank and Balderton (a London-based VC firm). As part of the round, Uber outlined plans to invest as much as $300m into a Wayve-powered robotaxi fleet to be deployed in 10 cities across multiple countries, including London and Tokyo (cities which are slated go live in H2’26). Wayve also has plans to integrate its ADAS (or L2++) tech into Nissan vehicles (slated for production in 2027) and Stellantis vehicles (slated for 2028).

• Waabi. Waabi is first and foremost a self-driving trucking software provider (though it also plans to service the self-driving car market). It was founded in 2021 and is based in Toronto. In Jan-26, Waabi raised a $750m Series C at a $3bn valuation co-led by Khosla Ventures, G2 Venture Partners and Uber (which invested $250m). The round has an option to be upsized by an additional $250m from Uber based on milestones that would support the deployment of 25k Waabi-powered robotaxis on the Uber platform. There was no timeline provided for the potential robotaxi rollout. The company’s CEO, Raquel Urtasun, was previously the Chief Scientist at Uber’s autonomous division, Uber ATG. Waabi already had a partnership with Uber Freight on the autonomous trucking side — Waabi has been running commercial trucking pilots on the Uber Freight network since 2023. One of Waabi’s key differentiators is “Waabi World,” an “ultra-realistic” genAI simulator. Using Waabi World, the company is supposedly able to significantly reduce its need for real-world testing while also enhancing safety performance. Waabi World supposedly has a 99.7% success rate when it comes to mimicking real-world physics.

Big Boys:

• NVIDIA. The $5tn mkt cap chip giant, NVIDIA, is also a “full-stack” AV tech provider. NVIDIA offers a full “soup-to-nuts” solution — the hardware, the system architecture, and the AV software (including data, training and simulation). Mercedes, for example, is a full-stack customer. Customers can also work with NVIDIA in just certain parts of the stack. Despite NVIDIA being a competitor of Tesla’s in AV software, Tesla is actually NVIDIA’s largest customer in automotive due to their partnership in training.47 Nuro, Waabi and Wayve are “frenemies,” building on top of NVIDIA’s Hyperion platform and using pieces of NVIDIA’s broader AV offering. At CES in Jan-26, NVIDIA launched Alpamayo, a “family of open AI models, simulation tools and datasets designed to accelerate the next era of safe, reasoning‑based autonomous vehicle (AV) development.” Alpamayo seems to be “quasi” open source. The software is free to use, and in the future should get you all the way to L4 autonomy. You can even use it with your hardware of choice (i.e. you don’t need to use NVIDIA chips in-car). But to train the model and run simulations for your specific vehicle (apparently all car OEMs will need to do this to get production-ready), you’ll need to work with NVIDIA chips.484950 At NVIDIA’s Mar-26 GTC conference, Uber and NVIDIA announced “an expanded partnership to launch a fleet of autonomous vehicles entirely powered by the full-stack NVIDIA DRIVE AV software across 28 cities and four continents by 2028.” The rollout will begin with LA and the SF Bay Area in H1’27. Car OEMs are TBD, but could include any OEMs building on NVIDIA’s Hyperion platform (which includes NVIDIA compute and a pre-validated/open reference sensor suite/architecture). These OEMs include Mercedes, Lucid, Stellantis, Nissan, Hyundai, BYD and Geely.5152 Jensen mentioned at the Jan-26 CES that NVIDIA had been working on self-driving for at least 6-7 years (including 5 years in partnership with Mercedes).

• Zoox. Zoox was founded in 2014, and is headquartered in Foster City, CA. Amazon acquired Zoox in Jun-20 for ~$1.2bn. Zoox’s AVs are designed in-house and purpose built to be “robotaxis.” The Zoox vehicles are toaster-shaped, aloe-green, feature carriage-style seating, and drive “bidirectionally” (i.e. there is technically no “front” or “back” of the vehicle). In late 2025, Zoox launched limited robotaxi services in Las Vegas and SF. The Zoox service in SF came with a very limited ODD (parts of SoMA, the Mission, the Design District). In Mar-26, Zoox quadrupled its SF service area to include the Marina, North Beach, Chinatown, Pacific Heights and along the Embarcadero. In 2025, Zoox opened a 220k facility in Hayward, CA capable of producing 5k robotaxis annually, with the aim to eventually ramp to 10k. Zoox has so far partnered with Uber in Las Vegas (set to launch summer 2026) and Los Angeles (by mid-2027). Zoox intends to keep the SF market exclusively 1P to Zoox. Further, the company intends to make Zoox bookable through their 1P app in all markets, including Las Vegas and Los Angeles. Zoox recently launched limited offerings in Austin and Miami (without Uber), which makes for four markets served in total. Zoox is currently mapping and building a presence in at least eight additional US cities, including Dallas and Phoenix.

OEM-owned efforts:

• VW. In Apr-25, VW and Uber announced plans to deploy “thousands of all-electric, fully autonomous ID. Buzz AD vehicles within multiple US markets over the next decade, starting in Los Angeles.” These vehicles are essentially minivans, and my understanding is that VW plans to operate them as ride-pooling shuttles. Testing began in 2025, with a commercial launch expected in 2026. As of Apr-26, VW was still on track to launch the LA pilot by the end of 2026 (with safety drivers). It’s unclear what the initial ODD will be, but VW plans to expand to “multiple US markets.” The autonomous tech is powered by Mobileye, a long-time ADAS provider based in Israel and founded in 1999. “Pre-series” production has started on the autonomous ID. Buzz AD at the company’s Hanover plant (targeting 500 vehicles by the end of 2026). Large-scale “series” production is slated for 2027. My understanding is the AV hardware components will be integrated “in-line” during production. The company has over 100 test vehicles on the road across the US, Germany and Norway. VW aims to have 100k AVs on the road by 2033.

• Rivian. Rivian is an EV carmaker, founded in 2009, that is building its own autonomous software. The company is even building its own chips. Uber and Rivian stuck an agreement in Mar-26, where Uber would invest up to $1.25bn in Rivian, “subject to the achievement of certain autonomous milestones by specific dates.” Uber’s initial investment is $300m. Uber is expected to purchase 10,000 Rivian R2 robotaxis with the option to purchase up to 40,000 more beginning in 2030. The intention is to expand to 25 cities by 2031, starting with San Francisco and Miami in 2028. The R2 SUV is Rivian’s next-generation model that will debut with a ~$58k price tag (vs. $75-100k+ for its flagship R1), with more affordable versions planned, including a $45k model debuting late 2027. Rivian’s software is powerful enough that Volkswagen elected to strike a $5.8bn JV with the company to access Rivian’s EV architecture and software. VW is launching its first vehicle built on the Rivian platform — the ultra-cheap ID EVERY1 (a small, electric, four-door hatchback with a $21.5k price tag) — in 2027.

• Motional. Motional is a Boston-based self-driving venture owned by Hyundai. Originally a JV between Hyundai and Aptiv, Aptiv pulled out of the JV in 2024. In response, Hyundai put an additional $1bn into Motional while the company was forced to lay off ~40% of its workforce. Motional and Uber launched a robotaxi service (using factory-built Hyundai IONIQ 5s) in Las Vegas in Mar-26. The initial service features a human safety operator, but it is expected to go driverless by the end of 2026. Hyundai vice-chair Jaehoon Chang recently told his staff: “our survival and future depend on the shift to a software-driven mobility company.”

Wait, there’s more?

• Momenta. Momenta is another Chinese AV software provider that flies more under the radar than the “big 3” of Apollo Go, WeRide and Pony.ai. However, the company has drawn investment from Tencent, Temasek, SAIC motor, Toyota and Mercedes. The company raised funds in Sep-25 at around a $6bn valuation, reportedly. Momenta and Uber struck a partnership in May-25 to jointly target markets outside the US and China, and will begin testing a robotaxi service in Munich in 2026. Momenta is a partner of both Mercedes and BMW for their Advanced Driver Assistance System (ADAS) which powers partial autonomy offerings like hands-free driving on highways.

• Avride. Avride was formed in 2017 by parent company Nebius Group, formerly Yandex (it sold off the Russia-based Yandex in 2024). The company is based in Austin. Avride develops self-driving technology for passenger vehicles as well as for sidewalk delivery robots. Uber uses Avride both for sidewalk robots and for a robotaxi service in Dallas, launched in Dec-25. The Avride robotaxi operation is currently limited to a very small ODD within Dallas. Avride’s delivery robots operate across Philadelphia, Jersey City, Austin, Dallas, Ohio State University, Tokyo, Seoul and Dubai.

• May Mobility. May Mobility is a start-up based in Ann Arbor. The company has formed partnerships with Uber and Lyft in the US as well as Grab in SE Asia. The company is partnered with Uber in Arlington/Dallas. May works with modified Toyota Sienna minivans, and operates in only a small portion of the city. The company is partnered with Lyft in Atlanta.

What Happens to Human Drivers as AV Prices Fall?

As the price of an AV drops below the price of a human driver, what will happen to human drivers? Will Uber have to cut driver rates/pay? And if so, will human drivers stop driving?

Uber has signaled that as AVs come onto the platform, they will take on the position of “baseload” supply. As a result, however, there won’t be enough AVs on Uber’s platform to meet peak demand — far from it. Thus, human drivers will need to come onto the platform to handle these peaks.

Peak demand will still require “surge” pricing. The surge pricing of the future, however, will not look like surge pricing today — it will probably look more like the standard pricing of today (as prices fall). But because of this, Uber drivers should still be able to earn reasonable pay (though less than before).

Some drivers will leave the platform as a result. If a driver only drove during peak hours before, as they didn’t want to drive for any less than the surge pricing rate, they will churn. However, there will be drivers that stay. For them, they will take the non-surge rate, even if they can only get it during peak hours.

Dara was asked explicitly about this issue on a Oct-25 Decoder podcast.

Interviewer: “I took an Uber to work today. I figured I should use your app before I talk to you. I asked the driver and he said I want my rates to go up…I’m curious how you’re going to make the rates go up as you have big fleet operators who now have giant fixed costs who are happy to have 24-7 utilization and maybe take a lower rate for greater use while the drivers still just want high rates. How do you how do you square that circle? Because it feels like the driver rates are going to get pushed down as autonomy comes into the mix.”

DK: “So I think that our job is going to be to manage the balance between demand and supply. I think autonomous is actually going to drive more demand. So that, you know, even though as a percentage of our overall inventory autonomous is going to be an ever increasing percentage, the overall demand — the number of drivers — that we need on our network over the next 5 to 10 years is going to increase because demand increases more, so to speak.

20 years from now you might be right and from that standpoint we just have to manage our inventory which is there’s a turnover of drivers all the time and we want to make sure that we give drivers the right expectations of earnings etc. And in a city where demand isn’t growing, maybe because autonomous supply is coming in, we just have to slow down recruiting of drivers and not make them [false] promises as it relates to earnings. So I think communication, being straight with people — that’s ultimate answer. But I don’t take it for granted. This is going to be a transition that we have to manage very, very carefully and make sure that our driver base is taken care of.”

I think Dara is potentially right that demand will actually go up for some period for human drivers as overall demand for rideshare increases. However, as AV prices really begin to fall (as well as take up a larger percentage of rides), Uber will need to gradually reduce pay and hours for their drivers.

Here is Uber’s Chief Product Officer Sachin Kansal from a Nov-25 WSJ podcast, highlighting Uber’s unique ability to activate drivers during times of peak demand:

“For anyone to have enough cars on a Friday evening at 6 PM, when you both have this combination of commute as well as party traffic that’s starting to happen. You just can’t have enough cars, or if you try to deploy enough cars to be able to meet that demand, you’re gonna have a lot of idle time on your hands because these are assets that you’re deploying in the city.

Now we have the benefit of having a flexible supply pool in our core business, and we know when to deploy that supply pool, depending on the peaks and troughs of the demand in that particular city…So the flexible supply pool is our driver base. And we have now more than a decade of experience on how we work with our drivers to be able to bring them on the platform at the right time. If you look at the driver app, there is so much guidance that we provide. We communicate with them, the app tells them when it is busy, when they should come online, when there are incentives, what the earnings [will be]. So that’s a very well honed system for us on the driver’s side.”

What happens when prices are so low for AVs (say <$1 per mile vs. a human driver at $2+) that nobody would want to be matched with a human, even during peak hours?

In this case, Uber could be forced into a situation where it subsidizes human driver fares/pay in order to keep them on the platform. This would allow Uber to keep pricing lower for riders, but it would negatively impact Uber’s take rate and margins.

What Would a Nationwide Robotaxi Fleet Cost?

Based on my back-of the-envelope math, the cost of a nationwide fleet actually seems somewhat achievable. This is probably not a great thing for Uber, as it shows that Google/Amazon/Tesla are capable of building fleets that could have sizable impacts on the market.

If Waymo’s 3,000 vehicle fleet today is doing 500k rides per week, that’s 26mm rides per year or 8,667 rides per vehicle per year. If Uber is doing 2.2bn rides a year in its US mobility business, that’s in theory achievable with a fleet of ~250k cars (using the same 8,667 number). However, we know that a lot of these rides are happening simultaneously during peak hours. Thus, the actually fleet size will need to be something like ~3x larger, or ~750k vehicles.

If ultimately the price of an AV settles at ~$50k, then the cost of an entire US Uber-sized robotaxi fleet would be ~$38bn, in theory (or ~$7.5bn annually, assuming a 5-year vehicle life). However, if robotaxis were to grow to say 5% of all miles traveled in the US, that would mean a ~10x’ing of today’s fleet. Thus, instead of $7.5bn, you’d be looking at something more like $75bn annually, just for the US. If robotaxis grew to 15% of all miles traveled, you’d be looking at ~$225bn annually.

This figure doesn’t take into account the build-out of depots and charging infrastructure that would be needed to support such a fleet. It also doesn’t take into account operating costs, which might be ~3x the fleet’s depreciation cost. But let’s assume these costs are absorbed by the fleet operators and/or offset by revenues.

If we assume 75% of the cost of these vehicles can be offloaded to financing providers (i.e. we assume Waymo, Uber, etc. are only putting 25% down), then we are actually back down to manageable figures. Instead of $75bn annually, for instance, the capital required would be more like ~$20bn annually (to support a fleet that could handle ~5% of all miles traveled). Google or Amazon alone (with operating free cash flows in the ~$150-200bn range) could manage this type of spend.

Delivery

Key Priorities

Supply, Supply, Supply

The first rule of building a vibrant marketplace is building supply. Supply, according to Uber management, “broadens the appeal of the marketplace, attracts new customers, and drives engagement with existing customers.”

Uber Eats is still relatively early across its top markets in terms of bringing merchants onto the platform. The chart below from Uber’s 2024 investor day shows the percentage of total potential restaurant and grocery locations that were actually on the Uber Eats platform. As we can see, there is still a lot of room for Uber to grow its merchant base.

From the 2024 investor day:

“Another key opportunity for us is bringing more merchants to the marketplace. We think of new merchants like a streaming platform thinks about new content. It fundamentally broadens the appeal of our marketplace, attracting new consumers while driving more engagement with existing ones. We’re making great progress, but we’re still quite early across our top markets with roughly 20% of available restaurants and grocery stores on our platform. We’re building tech to drive down on-boarding time, improve self-sign-up and self-service and tools to make our commercial teams more effective and more productive.”

When I compare the restaurants on Uber Eats in my area (SF Bay Area) to those on DoorDash, it’s hard to tell much of a difference. All the restaurants I regularly order from (from the national chains to the local single-location restaurants) are on both.

Penetrating the Suburbs

The suburbs are a big focus for Uber in delivery, as they are in mobility.53 In delivery, like in mobility, Uber originally focused its efforts in city centers. This made sense to most people at the time. The city centers are the most dense in terms of supply and demand. So it should follow that these areas would be the most profitable (high volume, high amount of batching). It is also where Uber was strongest in mobility.

Here is Dara on a Nov-25 HD in HD podcast:

“I’d say one mistake was that we focused too much on the big cities. Uber was in the big cities. Uber was a big city business. That’s where mobility was. Started in Paris and then New York and Sao Paulo. Those are our biggest cities, London. And it turns out there’s a huge market outside of the large cities, especially with delivery.

Suburbs, [more] sparse areas — it’s actually one of our top growth areas now. You know, seven, eight years in, we’re still growing at like one and a half to three times the speed that we’re growing in the big cities. And we could have focused on that earlier…We overfitted to big cities and we completely ignored the suburbs because we just made assumptions about the suburbs that turned out to be wrong.”

It turned out that the suburbs are actually a great (superior?) market for delivery: (1) it’s a larger overall market than the city centers (the suburbs are ~60% of the delivery market, according to Uber estimates), (2) order sizes are larger (read: families) which are more profitable orders to serve, all things equal (higher delivery fee), (3) it’s much easier to drive around in the suburbs, find parking, etc. (helps with ETAs, driver retention), and (4) the most obvious downsides of serving the suburbs (less batching, more downtime) turns out are somewhat mitigated by the fact that restaurants still tend to be clustered together in “downtown” areas (which helps with batching, positioning drivers).

Key to penetrating the suburbs is (1) supply, (2) low ETAs, and (3) attractive pricing — three dimensions where Uber is making progress. Here is CFO Balaji Krishnamurthy at a Mar-26 Morgan Stanley conference describing Uber’s strategy in the suburbs. He alludes to the fact that Uber has yet to really go “all-in” on the suburbs.

Analyst: “If we think about sort of pairing the sparse market mobility strategy with the cross-pollination strategy into Eats, how far away are you being able to say, ‘Look, we’re making progress in the sparse markets on rides. Now we’re going to really sort of press more into the delivery and the food and everything else?’ Is that a ‘27 dynamic? Is it ‘26? When can that investment happen?”

Balaji Krishnamurthy: “Yes. So it’s definitely something we think about. But before we start pressing on those kind of investments, we have to ensure that the quality of our product is excellent on both sides, right? And for delivery, the way to measure that is do we have the right kind of selection in every market. So if you’re in a sparse market, we cannot just show you McDonald’s, Subway, Starbucks and hope where consumers to be sticking to us because we do need to bring you the local champions in that market, and we have to get you the best selection from that town or city. So there’s still selection work that we’re doing, right?

Then, you have to ensure that the way your courier motion is in that market is also different, right? In dense markets, you can get to couriers on a very hands-off basis. They can come in and out of the system whenever they like because there’s enough density, whereas in sparser markets, you need to allow scheduling, you need to be able to tell couriers when to log on and when they can expect to find business, right? So there’s work that’s happening on that side of the marketplace as well. Suffice to say, it’s not lost on us.

It’s definitely a part of our plan, but we need to ensure that the product quality on both sides is good before we start making that investment. Otherwise, we’re not getting the returns we expect there.”

While Uber has been late to the suburbs, it’s now a clear focus for the company. And Uber seems to be having good success. Per Dara above, Uber is growing 1.5-3x faster in the suburbs than in the cities.54

Grocery, Retail, Direct

Uber is extending delivery beyond restaurants. The next categories up are grocery, retail and “direct” (or offering “white label” delivery for merchants who take orders through proprietary channels). Uber now even allows users to return retail items through Uber Eats (this is, to date, limited to items that were also purchased through Uber Eats).

The grocery and retail businesses are doing $12-13bn in gross bookings today (ex DH). In terms of grocery/retail “supply” on the Uber app, Uber management would tell you its around the level where food delivery was in 2019-2020. At that point in time, the food delivery business was at a very similar scale to where grocery/retail is today.55

Grocery is particularly attractive in that (1) it’s a huge market (2x+ bigger than restaurants), (2) it’s highly repetitive in nature, (3) the basket sizes are relatively large (so delivery is more profitable, all things equal), (4) you can batch more easily because food temperature isn’t as important and volume per store is higher and (5) it’s a relatively straightforward cross-sell opportunity to tack on to food orders (ala DoorDash’s “DoubleDash”).

Grocery is also, however, a total knife fight. It’s low margin to begin with. And Uber is going up against very large, very formidable, and very verticalized players in Walmart, Costco and Amazon, who also offer same-day delivery. Amazon, as mentioned earlier, seems to be blowing the doors off their grocery business. There is also fierce competition from DoorDash and Instacart in the US (globally Uber considers itself in “pole position” when it comes to groceries56).

Uber has some tailwinds behind its grocery business: (1) Uber is in the “top-up” category, a category that is growing quickly (it’s increasingly easy/cheap to buy groceries more incrementally vs. large periodic orders) and (2) it benefits from a very logical cross-sell motion to its huge base of Uber Eats users.

Here is Dara from a Sep-25 Goldman conference discussing Uber’s “top-up” positioning:

“So there’s a very, very long road map as it relates to grocery and delivery. Today, the grocery habits that we see are more of a top-up habit, which is, call it, $30, $40, $50 basket sizes versus the $100-plus weekly shop. So we’re much more in the top-up category. But what we’re seeing is 2 things. One is top-ups as a percentage of the overall grocery category are going up because it’s just becoming easier.

It used to be — I used to be an appointment shopper, my wife and I would go to the grocery store once a week, once every 2 weeks and load up, where you can just kind of order — just like with Amazon, it’s like I’ll order a single thing and press delivery there. With Eats, if you’re a member, get 2 or 3 things. It’s no big deal. So we’re seeing kind of top-ups grow as a percentage of the overall category. And we think eventually, as you get more consumers embedded into our ecosystem, we can become that weekly shop, and we can increase the basket size substantially.

But right now, the focus is just drive selection, just drive frequency within that top-up area and then eventually, you can become kind of the prime area for shopping.”

Still only ~18%57 of Uber Eats users have tried Grocery and Retail, so there is significant room to grow these categories. While grocery is not currently a profitable business for Uber, as orders get larger, and as more orders are batched, profitability should improve. Further, grocery is also the category most easily monetized through advertising (even more so than food). Uber is only in the initial stages of its ad monetization journey in grocery.58

Drive Affordability

Driving affordability is a big priority for Uber Eats.

The shear amount of merchant-funded offers I see in Uber Eats is noticeable. Uber seems very intent on showing a consistency large number of offers. From a Mar-25 Morgan Stanley conference:

“When you look at Uber Eats, for example, we are — one of our biggest programs is what we call merchant-funded offers. So actually, merchants providing buy one, get one free, if you spend $30 on a basket, gets a $10 item for free, et cetera. These are promotions that these merchants run. Those promotions give merchants more access to our audience, and merchant-funded offers are now running at over $1 billion of, call it, savings for our eaters per year as well.”

$1bn of savings on ~$75bn of GB (if we take Uber Eats’ GB from the time of the quote) may not seem like a ton (~1%). But if we assume the average savings on an offer is 10%, that would imply ~13% of orders came with a merchant-funded offer attached.

The Uber One membership also adds considerably to affordability ($0 delivery fee, up to 10% off orders over a minimum subtotal). Within Uber Eats, Uber One members represent 60%+ of bookings (significantly greater than their percentage of overall bookings).59

As density further improves, greater batching should also drive affordability. Around 50% of orders are batched today, per Dara’s Jun-26 Invest Like the Best interview.

Lastly, and perhaps most importantly, AVs will very likely ratchet up affordability in the not too distant future (no tips, lower cost per delivery over time).

Grow the Ads Business

One particularly profitable line of business for Uber, housed almost entirely within Uber Eats, is its advertising business. Uber generated “well over” $2bn in ARR from advertising in Dec-25, up over 50% YoY. Delivery advertising as a percentage of delivery GB crossed 2% in 2025 and seems poised to go significantly higher.

See Investment Merits for more discussion.

TAM

According to DoorDash, as of 2022, total restaurant spend across the markets it serves (which overlap strongly with Uber Eats’ markets) is over $1 trillion (and grocery & convenience spend is over $2.5 trillion).60

Uber hasn’t provided much in the way of an official delivery TAM over the past few years. Management did have this to say on Q3’25 earnings:

“And listen, food is a huge category. It’s no surprise. This is a $2 trillion TAM in food and kind of a $10 trillion TAM in grocery.”

Since this statement, Uber has expanded into an additional 20+ new markets through its acquisition of Delivery Hero. Thus, its TAM estimate will have gone up materially.

If we combine the 2025 bookings for Uber Eats, Delivery Hero (the portion Uber is acquiring) and DoorDash we get to $235bn of total bookings. Assuming 85% of this is food delivery, this gets us to ~$200bn of combined food delivery bookings. Assuming a TAM of $1.5tn, $200bn of bookings would equate to ~13% of the market. If we assume a TAM of $2.0tn it would equate to ~10% of the market, and if we assume a TAM of $2.5tn, it would equate to ~8% of the market.

In terms of food delivery penetration of overall restaurant spend, Uber has estimated that it’s in the mid-teens (vs. the mid-20s for overall ecommerce).61

Competitive Landscape

Uber Eats’ competitive set varies by geo.

In the US, by far Uber’s toughest competitor is DoorDash, which owns ~60% of the market, more than double that of Uber Eats. Internationally, Uber is a more dominant platform, especially post DH. Internationally, Uber competes mostly with DoorDash (which owns Wolt and Deliveroo) or regional players, many of whom are owned by Prosus (Just Eat, Lieferando, iFood, Skip, Thuisbezorgd, etc.).

Below is a highly unofficial table of market share by country for food delivery.

As a side note, the TAM estimates by country (in the table above) were provided by Gemini. It estimates the restaurant TAM across Uber’s markets to be ~$2.8tn — well above DoorDash’s TAM estimate (of $1bn) and Uber’s TAM estimate (of $2bn). Gemini’s estimated restaurant + grocery/convenience TAM across Uber’s markets is ~$8.5bn. The US represents roughly half this TAM according to Gemini.

We can see from above that while DoorDash has strong share in the US, Uber has similarly strong share in many international markets.

It seems Uber will be tough to dislodge in international markets. Uber offers both mobility and delivery, an advantage that none of Uber’s competitors benefit from. And Uber’s higher market share should provide for lower ETAs and prices. It’s a formidable combination and it doesn’t seem to allow for any type of wedge with which a competitor could take share.

Here is CFO Balaji Krishnamurthy detailing the mobility + delivery advantage on Uber’s Q4’25 earnings call:

“Our international footprint on Eats is not quite what it is with our mobility business. Every time we launch Eats along with mobility, we just have a structural advantage over the other players. And we’re able to grow category position. We were the #3 in the U.K. We’re #1, all organic. We launched from scratch in Germany and now are neck and neck with a top player in Germany in a lot of individual cities. Japan was another organic launch, and we’re by far the #1 player in Japan.”

In this Nov-25 HD in HD podcast, Dara interestingly expresses his regret in exiting the India delivery market. He knows, given Uber’s strength in mobility in India, that Uber would’ve ultimately had wedge to take share in what is a huge and growing delivery market.

“We made some moves that I regret. Like we got out of food in India. It’s a huge market. It’s a challenging market. But I would have liked to be multi-platform the way we are in mobility and delivery in the US and the UK. It gives us — the platform — gives us huge advantages in terms of Uber just being a part of your everyday life. And while those were good financial decisions at the time, boy, I’d love to be in food in some of the countries that we pulled out of.”

Here is management from a Mar-26 Morgan Stanley conference on their success in taking share internationally:

“So U.S., our category position is relatively stable. It’s a great category where both Uber Eats and DoorDash are benefiting from the expansion of the category. But outside the U.S., we have continued to gain share even though our position is 2 to 3x as large as the next player in the market.

And I’m talking about large markets like Canada, France, Australia, Japan, Taiwan, Turkey, right, all of which we are now at a strong #1 position and the U.K. as well, we are in a good #1 position now. So from a category position standpoint, there’s a lot of momentum that we’re getting there.”

DoorDash

DoorDash is a strong competitor in the US and globally.

DoorDash generated ~$100bn in GOV in 2025 (Uber Eats generated ~$132bn post DH). DoorDash generated ~$14bn of revenue and ~$2.8bn of adj EBITDA (Uber Eats generated ~$29bn of revenue and $5.8bn of adj EBITDA post DH, however Uber Eats’ EBITDA figure excludes its share of Uber’s “corporate G&A” and “platform R&D”).

DoorDash ended 2025 with 35m DashPass members, up 59% YoY. Uber ended 2025 with 46m Uber One members (ex DH), up ~53% YoY. DoorDash, it should be noted, is closely aligned with JPMorgan Chase. Premium Chase cardholders (Sapphire Reserve & Preferred), of which there are ~5-7mm, receive complementary DashPass subscriptions. This partnership has been in place since 2020, and is currently locked in through Dec-27. Uber has its own partnership with Amex. Premium Amex cardholders (Platinum & Gold), of which there are ~6-7mm, receive free Uber Cash varying from $120 to $320 annually.

As mentioned above, DoorDash focused its efforts in the suburbs early on, while Uber focused on the urban core. This proved to be a smart decision on DoorDash’s part and explains much of the move in DoorDash’s US market share from ~20% to ~60% over the last ~8 years (see chart below).

Below Dara comments on competition with DoorDash in a Stratechery interview from Feb-25. Uber is betting that over time, its combination of delivery + mobility (and the superiority of the Uber One membership), combined with a greater focus on the suburbs (increased selection, better reliability), will give Uber an opportunity to take back share:

“But coming back to the competition against DoorDash, I do think that one thing that they got right was really focusing on the suburbs. If you look in the US, in terms of the suburbs, it’s actually 60 plus percent of the marketplace. Uber has always been a really urban market. We grew up in cities. That’s where we already had the liquidity of the marketplace, and Uber Eats was kind of — they drive people, drive stuff as well. It was an offshoot, so I think DoorDash got the suburban family first, we got the urban 20-something year old first. We’re now kind of coming into each other’s markets.

DoorDash’s market, the suburbs turned out to be larger than the urban markets, but for us, suburbs and less dense markets, actually, not only are there a huge opportunity for Uber Eats, not just in the US, but all over the world. We’re going into secondary cities, tertiary cities, but it’s actually a really big opportunity with Uber. As we started expanding Eats into the suburbs, we’re seeing that there’s also Uber demand there. So actually, moving into these less dense cohorts and neighborhoods is a huge opportunity for Uber Eats, but also Uber as well, which is I think we’ve got another three to five years of penetration ahead of us.

You looked in the last quarter, Eats actually accelerated its gross bookings level, so it’s going to be a dog fight with DoorDash, but they’re a strong competitor. We’re a really strong competitor, and I’m very, very confident that the network effect that we have, Rides, Eats, along with the Uber One Membership program that’s now 30 million strong, ultimately is going to come out on top of the full competitive field in food and grocery…

I think it’s very difficult for companies to do more than one thing or build more than one brand. Not only are we multi-service, but we also expanded internationally much faster than, let’s say, the DoorDash’s of the world and many other competitors of the world, and on a short-term basis, it’s much harder. You’ve got to build more services, you have to generalize for the international markets. It takes longer to build things, but over the long term, we think it’s a much stronger strategic position to be in.

Again, DoorDash, they do only one thing, we think our superpower is in the platform. Their superpower, you could say, is in their focus. Every piece of evidence that we’re seeing is that the platform is winning against the whole ecosystem. Obviously, the markets that we’re chasing are much, much broader than just the DoorDash’s out there.”

Here is Dara again making the case that Uber has a path to take share on a Jun-23 Acquired podcast. He is pretty explicit that Uber will be willing to be loss-making in its efforts to gain share in the US:

“DoorDash is a tough competitor. DoorDash is larger than we are in the U.S. We are focused on keeping share in the U.S, gaining a bunch of share outside the U.S., and then over a period of time using the structural advantage of, one, profit pools outside of the US and then use the structural advantage that we talked about in terms of customer acquisition, over a period of time to hopefully gain category position against DoorDash. But they’re a tough competitor. We respect them. We don’t like them, but we respect them.”

Internationally, DoorDash has made aggressive moves to expand inorganically. In Nov-21, Uber acquired Wolt for a massive ~3.2x multiple of GMV. And in May-25, DoorDash acquired Deliveroo for a much smaller ~0.3x multiple of GMV. Wolt was founded in Finland in 2014. The company has leading market share in the Nordics, and has expanded into Eastern Europe over time. In it’s two largest markets, however (Germany and Poland — which Wolt entered in 2020 and 2018, respectfully), Wolt remains stuck in a distant trailing position. It is in the #3 position in Germany behind Lieferando and Uber (Uber entered after Wolt in 2021), and in the #3/4 position in Poland behind Pyszne.pl and Glovo (and perhaps tied with Uber).

DoorDash seems to lack a clear wedge through which to expand internationally other than trying to slowly grind it out through incentives and perhaps an incrementally better product (greater selection, higher reliability, product innovations like “DoubleDash”).

I think its fair to say that Uber has fared much better internationally. Here is management from Q1’25 earnings:

“To your question in terms of competition, especially in Europe, we’re really, really happy about our results in Europe. We recently, we believe, got to the #1 category position in the U.K. with Eats, entirely organically. We didn’t have to buy our way into glory, so to speak. France remains a top market for us. And we think that Germany, for example, is a market that holds a significant amount of promise. I think we launched in Germany 3 to 4 years ago, and our category position continues to increase in Germany as we invest in that market, both on the mobility and delivery side.

So we’re seeing very encouraging trends in Europe. And frankly, it’s not a surprise to see some of our competitors look to expand there inorganically. We like organic expansion more. We’ve been investing for years in these marketplaces, and I think it shows in our results.”

Shots fired!

DoorDash has also recently lost out to Uber on a couple of key partnerships.

After years of rumors that DoorDash would launch its own POS software (putting it into competition with Toast), in Nov-25 Toast (the leading POS software provider in the US with $100bn+ of GPV, or ~15-18% market share) announced that it would make Uber Eats the default delivery partner for “native” restaurant orders (or orders coming directly to a restaurant’s website), replacing DoorDash. It also announced that restaurant owners could launch Uber Eats promotions, update menus and changes photos right from the Toast dashboard.

In Mar-25, OpenTable, the leading aggregator of restaurant reservations with ~50% market share in the US, and Uber announced a partnership where Uber would integrate OpenTable’s reservation inventory into Uber’s new “Dine Out” tab. OpenTable will also be powering reservations for Uber in Canada, Mexico, the UK, Ireland and Australia. The integration highlights discounts that restaurants are offering and provides Uber One members with exclusive access to certain reservations. Additionally, within the OpenTable app, after making a reservation, users will be prompted to pre-book an Uber “reserve” ride (and in doing so will receive a discounted fare).

Instacart

Instacart is a ~$10bn mkt cap grocery delivery marketplace based almost entirely in the US.

Instacart generated $37bn in GOV in 2025, which was up 11% YoY. It generated revenue of $3.7bn and EBITDA of $1.1bn. The company served 26mm+ customers in 2025, and ~10mm in just the month of Dec-25. It is partnered with 2,200+ retailers across ~100k total locations. It has ~600k “shoppers” on the platform fulfilling orders. In Q4’25, GTV was up 14% YoY, representing Instacart’s strongest quarterly growth in 3 years.

Instacart’s key grocery partners include Costco, Kroger, Aldi, Publix, Albertsons (including Safeway and Vons), Publix, and Meijer.

Instacart generates much bigger grocery “baskets” than the likes of Uber Eats and DoorDash. Since 2020, Instacart’s average order has “consistently been $110 or more.” Instacart also has a robust advertising business, which generated ~$1.1bn of revenue in 2025, or ~2.9% of GTV. If you apply a ~60-70% margin to that, advertising dollars would have represented well north of half of the company’s overall EBITDA in 2025.

The company has ~$1.4bn of cash and no debt. It bought back $1.4bn in shares in 2025, including $1.1bn in Q4’25 alone (10% of its market cap!).

Instacart seems like a prime M&A target for Uber and Uber is perhaps the most logical buyer for Instacart.

Adding significant heft to Uber’s grocery business through an Instacart acquisition would be a win for Uber. Uber trails DoorDash in food delivery by a wide margin in the US, and Instacart would help close the gap meaningfully. A deal would seem highly synergistic, to boot: (1) tech platform, ad platform, SG&A cost synergies, (2) a meaningful uplift in batching by combining volumes, (3) synergies in combining their pools of couriers (more liquid supply, better pay for couriers given more volume/less downtime), and (4) better ad monetization of Uber’s grocery business. DoorDash has actually gotten more “verticalized” in its grocery offering with DashMart, putting it somewhat into competition with it’s grocery merchants. Uber has shied away from going “vertical,” so Instacart would be a better fit for Uber in that respect. Additionally, Uber and Instacart actually already have a close partnership: Uber powers the restaurant delivery feature in the Instacart app. Uber would also likely have an easier time (vs. DoorDash) of overcoming antitrust concerns.

Steve Jang, an early Uber investor, basically makes the case for an Uber/Instacart acquisition in a Nov-24 interview with Yahoo! Finance:

“Another area of acquisition — you know, the FTC aside and those approvals necessary for it. I really feel that the grocery category is adjacent, it’s relevant, it plugs right into the network. I think something in that area would be very, very exciting and makes a lot of sense as a plug-and-play acquisition.”

Delivery Hero Acquisition

Delivery Hero is a conglomerate of food delivery brands that operates across ~70 countries. In 2025, the company generated €49bn of GMV (up 9% YoY), ~€15bn of revenue and ~€900m of EBITDA. ~90% of its GMV comes from markets where it holds the #1 position, according to Delivery Hero management.

Its key brands include: Glovo (Europe), talabat (MENA ex Saudia Arabia), Hungerstation (Saudi Arabia), PedidosYa (LatAm ex Brazil and Mexico), foodpanda (SE Asia), and Baemin (South Korea).

Delivery Hero was founded in Berlin in 2011 and was originally a platform where restaurants with their own delivery drivers could take orders (i.e. simple lead gen). It expanded globally through acquisitions that included PedidosYa in 2014, talabat, Hungerstation and foodpanda in 2016, Baemin in 2019 and Glovo in 2022. In 2018, Delivery Hero actually sold off its entire German business (Lieferheld, Pizza.de and foodora’s Germany unit), citing intense competition.

In Jul-26, Uber announced that it had agreed to the acquisition of Delivery Hero for €41.50 per share. This equates to an equity value of $14.8bn, or $13.7bn taking into account Uber’s “prior stake purchases” (Uber built a position in Delivery Hero shares at lower prices over the last ~1.5 years). The acquisition is expected to close in H2’27.

Uber management believes it is paying ~8x EV / 2027E adj. EBITDA for Delivery Hero. This figure is based on Delivery Hero’s consensus 2027 EBITDA estimate and includes $1.2bn of expected run-rate synergies.

Delivery Hero was acquired after a period of turmoil for the company. In the midst of its expansion, the Delivery Hero over-levered itself to the tune of €4.1bn in net debt at the end of its 2023 — a year in which company generated just ~€250m in adjusted EBITDA and negative free cash flow.

The situation attracted activist investors Sachem Head in 2024 (which took a Board seat) and Aspex Management in 2026 (which amassed a ~9% stake). In Dec-25 the company formally announced a strategic review, and in Mar-26 confirmed that it had engaged JPMorgan as part of the process.

In May-24, the company agreed to sell its foodpanda Taiwan business to Uber for $950m, but the deal was ultimately blocked on antitrust concerns. Uber had to pay a termination fee of $242m to Delivery Hero in Apr-25 as a result. Delivery Hero eventually agreed to sell this business to Grab Holdings for $600m in Mar-26. Additionally, in Dec-24, Delivery Hero IPO’d its talabat business, selling a 20% stake for proceeds of €1.8bn.

As a result of these actions, Delivery Hero was able to meaningfully reduced its debt burden. In the meantime. the company also significantly enhanced its profitability, generating €903m of EBITDA in 2025.

DH’s strategic review process also drew the attention of Uber. As part of the blocked foodpanda Taiwan acquisition, Uber separately bought $300m in newly issued Delivery Hero shares (good for around ~2.5% of the company). Then, starting Apr-26, Uber began rapidly expanding its stake in Delivery Hero. Through a series of transactions, Uber ultimately amassed a ~37% stake in the business prior to the transaction announcement.

Uber, however, was not able to ultimately acquire all of Delivery Hero’s segments due to presumably regulatory/anti-trust hurdles, primarily in Europe. We can see below that Uber didn’t acquire most DH segments in which it already had a presence:

Delivery Hero’s crown jewels seem to be (1) its Baemin business in South Korea (~60% market share), (2) its talabat and Hungerstation businesses in MENA, (3) its Glovo business in Spain (~55% share) and (4) some of its PedidosYa units in LatAm. Its foodpanda business seems relatively challenged (i.e. trailing Grab) across most of the SE Asia markets in which is operates.

I think DH is ultimately a smart acquisition for Uber:

• The price is reasonable at 8x EBITDA. This bakes in a large amount of synergies ($1.2bn). But Uber management is adamant that, if anything, they are sandbagging this synergy estimate.62

• Provides Uber with significantly more scale. The acquisition will add $42bn of gross bookings, taking Uber’s total GB from $193bn to $235bn (a 22% increase). EBITDA will increase by ~$2.2bn, from $8.7bn to ~$11.0bn (a 26% increase).

• Adds leading markets. Of the 50 markets added, 38 of them are in the #1 position, according to Uber management. Baemin, talabat and Hungerstation in particular seem like crown jewel assets. Undoubtedly Uber would have liked to have added Glovo’s Spain unit, but Uber already operates in the country in the #2 position.

• Shifts more profits to delivery. Prior to DH, Uber generated ~31% of its profits from delivery. Post DH, Uber will generate ~42% of its profits from delivery. The delivery business is less exposed to AV risk.

• Shifts more profits overseas. Prior to DH, Uber generated ~54% of total revenue from international markets. Post DH, Uber will generate ~63% of revenue from international markets. International markets are less exposed to AV risk.

• Almost doubles Uber’s number of multi-product markets. As discussed, Uber’s multi-product offering is a key differentiator for the company. Thus, the ~doubling of these markets is a great development. The Delivery Hero segments being acquired will be better businesses on a go-forward basis, as they will be combined with a mobility offering.

  

  The acquisition also gives Uber some “delivery-only” markets for the first time (e.g. Philippines, Singapore, Cote d’Ivoire, Morocco, Venezuela). Perhaps this gives Uber the opportunity to add mobility in these countries.

The downside to this acquisition, of course, is that it adds leverage. The deal will be financed with a €14bn bridge facility (to be refinanced) and cash on hand. However, if all goes according to plan, Uber will only be ~1x levered (on a net basis) by 2027. That’s pretty manageable, even in a fast-changing industry like Uber’s. My hope would be that Uber pays down its debt rapidly. The company has the ability to do so, but Uber has also committed to continuing with its buyback activity and also buying a significant number of AVs (discussed later).

Autonomous Transition

We’re headed for a world, as a result of AVs, where the cost of delivery (delivery fees + tips) could fall by 50%+. There should be a lot more deliveries in this world.

The growing consensus seems to be that in the future, some combination of (1) sidewalk robots, (2) mid-range delivery bots (that can drive on residential streets and in bike lanes), and (3) drones will handle our on-demand delivery needs.

Here is Dara on a Nov-25 HD in HD podcast:

“We think delivery is going to be a combination of sidewalk robots that we have in place. We’ve got 10 partnerships now all around the world — Serve, Cartken, Avride, and a bunch of others. And these are like, I don’t know if you’ve seen them, these robots that walk along the sidewalks for short deliveries. They’re not around in New York yet. So that’s one. And then there’ll be drones. So there’ll be drone delivery that gets you deliveries longer distances in the suburbs. And there are a number of players there. We recently made an investment in Flytrex, who we’re going to partner up with. And I think we’ll have some other partners. And then there might be an in-between, like bike lane robots or maybe cars although I think that’s probably too expensive a bill of materials to carry around food.”

Here is Dara again on the On with Kara Swisher podcast from Dec-25. He discusses the three AV form factors that will be a part of the delivery solution. He also discusses the “last mile” challenge related to AV delivery.

“So, number one, I would say Uber Eats is an absolute star in our portfolio. Like, it was it was 10% of our business when I took over. It’s now almost 50%. The team is doing a great great job there.

There isn’t going to be — I think AV for mobility is clear because the first mile and last mile taken care of. Like you walk to the car, you walk away from the car. With delivery, it’s actually an amalgamation of solutions. So, we have sidewalk robots for short deliveries within a mile, mile and a half. But for a three-mile delivery, it just doesn’t work. For those three mile deliveries in urban destinations, we’re probably going to have to use either AVs that go in bike lanes or go on the road as well so that they can get there faster. I’d say that is behind [versus] sidewalk robots [which] are coming first. And then in suburbs etc., it’s going to be drones. And we are far from commercializing that technology. We’re in the experimentation phase, but it’ll be the three of those.

And then the question is who’s going to take care of the first mile and last mile? Restaurants still, they’re busy. They don’t like coming out of the restaurant, putting the food in the robot, etc. And people are lazy, you know, they don’t want to come down to — if they’re in a high-rise. I remember when I was delivering food — like it’s amazing, people would deliver from a Chipotle three blocks away because they just want the food at their doorstep. So the first and last mile is still to be solved, but I do think AV is also going to be a huge application as it relates to Eats. And I think our Eats business is ultimately potentially going to be bigger than mobility.”

The last mile challenge might initially be solved through simply lower pricing. If you want a cheaper delivery through a sidewalk robot, you might have to walk out to the curb if in the suburbs or downstairs to the lobby if in a high-rise. For restaurants, they will have to decide if they are willing to walk orders out to delivery bots sitting outside (my sense is most will be).

As mentioned, Uber is already partnered with a handful of providers of both sidewalk robots and drones. Interestingly, they are finding that the economics of using sidewalk robots are actually already better, in some instances. Here is CFO Balaji Krishnamurthy from a Mar-26 Morgan Stanley conference discussing Uber’s progress with delivery AVs:

“I think on that front, we’re in a pretty good spot. We already have over 1,000 delivery bots on our network in more than 10 cities. We’re working with about 7 partners in that ecosystem. It spans both sidewalk robots and drones.

And what we are seeing right now is sidewalk robots have potential, but they come with their own friction, right? You are -- if you’re a merchant or a consumer, you’re now used to seeing a courier show up to your counter, pick up the order and drop it off at your doorstep, whereas sidewalk robots have friction on both ends, and you have to go out and meet the bot. Drones, on the other hand, can be faster. They can cover a larger radius, and they can potentially even drop off the package in your backyard. So there’s a broader use case there.

I think we are going to keep exploring and see where this goes. The good news is that economics of these deployments is already quite attractive for us. It’s interesting for us to be able to deploy these while not having to make a deep investment, and we’re learning as we go. Long term, no question in our mind that drones and bots will play a much bigger role in delivery, but there’s still some things to be ironed out.”

How might delivery economics look in an AV future? Below is some back-of-the-envelope analysis, showing what the cost of delivery could be for both sidewalk robots and drones.

For sidewalk robots, I assume the cost of the robot is $4k and that it lasts 3 years. I assume $3.50 per day for cellular costs, $0.25 per day for electricity and $2,500 annually for storage, repairs and tele-operations. If we also assume the robot does 7 deliveries per day, this works out to $2.04 in costs per delivery.

Next, I show how the economics of an $18 food order could be divvied up — contrasting a human courier delivery with an AV delivery. With the AV, I eliminate the $2.50 tip and take delivery fees down from $4 to $3.50. As a result, the consumer is paying $21.50 instead of $24.50 (12% savings). Uber still earns its $4.50, and the restaurant actually earns a bit more. The “courier” earns just $2.50 instead of $6.50. Alternatively, you could keep the delivery fees at $4.00 and give more to the restaurant, etc.

Any way you slice it, its a better outcome for the consumer, the restaurant and Uber (even if Uber earns the same amount per transaction — or even less — there will be more of them). In the example above, the consumers is saving ~46% on delivery fees (paying $3.50 with AVs vs. $6.50 without). With larger basket sizes, the savings would be even greater.

Here is the same analysis but for a drone. Again these numbers are just estimates and theoretical. Here I assume the drone costs $15k and that it lasts 4 years. We get to a pretty similar result in terms of cost per delivery.

Note that both examples above don’t assume any batching, which would further reduce the cost per delivery.

Here is Uber’s COO Andrew Macdonald on a Feb-26 The Compound podcast, again, detailing how the economics for AV delivery are actually already cheaper than the human courier equivalent for certain deliveries:

“On the delivery side, what I’d say is interesting and not well understood is the unit economics already work. Our cost per trip for some of our Serve deliveries [using AVs from Serve Robotics], for example, in markets like Los Angeles are already cheaper on certain deliveries than the human courier equivalent. And I think that’s going to be really exciting both for our company, but also for consumers who can save money.”

In the world of mid-range delivery bots, DoorDash seems to be leading the way with its “Dot” vehicle. The company unveiled Dot in Sep-25, and it is currently making deliveries in Phoenix, like this one:

Dot is roughly a tenth the size of a car, at less than 5 feet tall and 3 feet wide. It is 350 pounds and can travel up to 20 MPH on streets or in bike lanes. It can fit six pizza boxes or up to 30 pounds of food.

Nuro — yes the same Nuro that is rolling out mobility AVs in partnership Uber — was actually working on a similar vehicle, pictured below.

The company eventually pivoted away from a strategy of selling full-stack delivery vehicles (to its current strategy of licensing AV mobility software). As a result, this vehicle is no longer in development. However, undoubtedly other providers will step in to fill this opportunity space should it prove to be a valuable one.

Overall, the opportunity for disruption in delivery through AVs seems lower than in mobility over the medium term: (1) there is less focus on it from the likes of Waymo, Tesla, etc. (for now), (2) there is no Tesla equivalent that already manufactures these delivery AVs at massive scale today, (3) even if there were, in order to provide a holistic delivery offering (and compete with Uber Eats/DoorDash), a provider would need to amalgamate different AV form factors (drone, sidewalk and mid-range), (4) the “last mile problem” as discussed earlier, (5) a robust 1P app is harder to build in delivery due to the huge number of merchants you’ll have to onboard, and (6) as previously mentioned, these delivery networks run at razor-thin margins (and actually make most of their money on advertising, which only makes sense to offer once you reach a very large scale).

However, AVs do still seem to open the door for someone like Amazon or potentially Waymo to compete, as they obviate the need to build a supply of human couriers, which was arguably the largest barrier to entry.

Freight

Unbeknownst to some, Uber actually has a “Freight” division. You can think of it kind of like “Uber for trucking.” Uber Freight matches shippers (companies that need to ship things) with carriers (trucking companies and independent truckers).

Uber Freight was established in ~2017, and in 2025 it generated ~$5.1bn in bookings (or ~2.5% of Uber’s total bookings). Growth has been flat the last couple of years, and the segment operates around breakeven. Because Uber Freight’s revenue is booked on a gross basis (i.e. the amount it charges shippers, not the difference between what it charges shippers and pays out to carriers — or what would be “net” revenue), it also reports a revenue figure of ~$5.1bn. This revenue figure is not “apples-to-apples” with the rest of Uber’s business, where revenue is almost entirely presented on a “net” basis.

Uber Freight initially operated solely within the “brokerage” or “spot market” segment of the freight market. This represents about ~20-25% of the market. ~40-50% of the market consists of “private fleets,” or fleets owned by the companies that are doing the shipping (e.g. Walmart, Pepsi, Sysco). The other ~30% of the market is where mostly large shippers (e.g. Target) sign longer-term contracts with mostly larger trucking companies (e.g. J.B. Hunt, Knight-Swift) for more regular shipping needs.

The broker share of the market has actually increased from ~10% to ~20-25% over the last ~25 years. As the ease of booking shipment through brokers has increased, it has led to smaller shippers increasingly ditching their fleets. Also the carrier base has continued to fragment — there are over 900k trucking companies in the US, and 95% of them operate 10 or fewer trucks.

Uber, not surprisingly, took the approach of offering a highly “digitally native” freight brokerage experience. Instead of negotiating rates by phone which is what a traditional broker might have done, Uber provided instant quotes. If a shipper accepted an Uber quote, and Uber ultimately couldn’t find a carrier to fulfill the shipment at a profitable rate, Uber simply lost money on the transaction. Additionally, Uber Freight plugs right into a shipper’s ERP system. Uber also provided “Uber-style” visibility into where shipments were at any given point in time.

With Uber’s acquisition of Transplace in 2021, Uber broadened its Freight offering beyond the brokerage market. Transplace, which Uber acquired for $2.25bn, was a leading software and managed services provider operating something like a fully outsourced logistics provider for shippers (sometimes called a “4PL”). The company was formed in 2000 when six leading trucking companies (JB Hunt, MS Carriers, Swift Transportation, Werner Enterprises, Covenant Transport and US Xpress) merged (and spun out) their logistics arms.

Transplace’s software platform offered shippers sophisticated supply chain planning, shipping optimization and day-to-day execution. From Uber’s Transplace M&A deck:

At the time of the acquisition, Transplace was doing ~$100m of EBITDA and growing ~15% YoY. It claimed gross retention of 96%, an average customer tenure of ~8-9 years and an average contract length of ~3-5 years.

While Uber has failed to grow the business much in recent years, they remain committed to it for now. Here is Dara from a Mar-26 Semafor podcast:

DK: “I view freight as our first shot at getting into the end-to-end logistics business. You know, what we do with Uber Eats to some extent, we’re doing on-demand, local, last mile logistics. One thing you may or may not know about Eats is, if for example you order an Apple iPhone [through Apple directly], you can get it delivered to your home and that’s an Uber Eats courier who goes and picks it up from the store and [delivers it]. So to some extent we’re already doing third-party last mile delivery.

What freight represents is first mile delivery. Right from the factory to the warehouse. And then there’s this middle mile that we haven’t penetrated yet. Our vision is to build an end-to-end on-demand, next generation logistics network powered partially by truckers moving from factories to warehouses, smaller trucks from the warehouse to the store and ultimately individual consumers from the store to the end consumer.

We’re the only company that’s taking it on. Freight is part of that ecosystem that we’re putting together. It’s a big challenge, but an enormous opportunity for us to build that end-to-end logistics stack, and it’s something I’m really, really excited about.”

Interviewer: “Sounds like you’re firmly committed to it.”

DK: “Oh, yeah.”

My expectation would be that Uber sees AI and autonomous trucking as potential wedges to grow share. AI is a technology that could meaningfully improve its managed services and brokerage offerings (e.g. help reduce costs for shippers, reduce downtime for carriers). Also Uber seems highly focused on securing driverless trucking capacity for its brokerage and managed services offerings. Uber has made meaningful investments in Aurora and Waabi, two leading AV tech providers in the trucking space.

I’m not ascribing any value to Uber Freight at this point. It hasn’t grown in 3 years, it isn’t profitable, it doesn’t seem to enjoy much in the way of synergies with the rest of Uber (at this point at least), and it’s unclear to me if it has much of a moat.

Further, in May-26 Amazon introduced Amazon Supply Chain Services (ASCS), opening up Amazon’s logistics network to any business (with core offerings including freight, distribution & fulfillment and parcel shipping):

My guess is that this was not a great development for Uber Freight. This seems like potentially the type of solution Uber was ultimately going for (and more).

Valuation

Multiples

Uber currently trades for ~12x ‘27e EBIT and ~10x ‘27e EBITDA, pro forma for the impact of Delivery Hero.

Note that the non-GAAP EBIT figures above deduct SBC. The non-GAAP EBITDA figures do not, however. I’m not showing a FCF multiple because, as discussed earlier, Uber’s captive insurer inflates its traditional FCF metric (OCF - Capex).

My EBIT figures may prove to be a bit aggressive (they imply EBIT growth of 40% and 28% in ‘26 and ‘27, respectively, for Uber’s legacy business). However, I wouldn’t expect them to be wildly off the mark — EBIT grew 50% in ‘25 and 42% in Q1’26.

Comps

Below are a selection of comps for Uber, showing certain trading and operational metrics.

Note that the Uber numbers above are not adjusted for Delivery Hero.

Looking at EV / ‘26 EBITDA, it’s interesting that Uber doesn’t trade at any type of premium to its comp set. DoorDash specifically trades at a large premium to Uber (~24x vs. ~14x). Much of the difference is perhaps attributable to (1) less perceived AV risk for DoorDash and (2) DoorDash’s higher recent growth (in bookings and users).

Also interesting is the delta in monthly active users between Uber and DoorDash. Uber has ~202mm monthly actives vs. DoorDash with just ~56mm. ~63% of DoorDash’s monthly actives are already DashPass members. For Uber, the figure is just ~23%. Also DoorDash generates almost twice the annual GMV per monthly active (~$1,822) vs. Uber (~$958). This seems to suggest that Uber has more opportunity to penetrate its existing user base. DoorDash’s numbers are skewed by being US centric vs. Uber’s which include more developing regions such as India and Brazil.

Uber’s reported take rate in delivery is ~19% vs. DoorDash at just ~13%. This would seem like a somewhat precarious position for Uber, reminiscent of Expedia vs. Booking.com. Booking, with a lower take rate and focus on building supply, blew past Expedia which held take rates higher and focused on building demand (a story Dara knows well). However, the delta is not as bad as it looks because (1) Uber is generating more ad revenue as a percentage of total bookings than DoorDash and (2) DoorDash’s bookings figure includes tips and Uber’s does not. Both factors inflate Uber’s take rate vs. DoorDash.

Grab is perhaps the most comparable business to Uber as it has both a mobility business and a delivery business. Grab trades at a much lower GMV multiple than Uber (~0.48x vs. ~0.79x). However they trade in line on revenue and EBITDA. Grab is showing very low take rates in both mobility (~15%) and delivery (~13%).

It should be noted that Grab is projecting $1.5BN of EBITDA in 2028, and it trades for just ~7x that figure. Rumors that Meituan, the largest food delivery player in China, could be entering SE Asia combined with competition from Shopee (the largest ecommerce platform in SE Asia who has launched food delivery) seems to be keeping a (perhaps well-deserved) cap on the multiple. Grab, however, is also much less at risk from AVs. AVs are much further away from being comparable on a price per mile basis to human drivers in the very cheap SE Asia region, where Grab operates exclusively.

M&A Comps

Below are a selection of M&A comps.

The median delivery GMV multiple is around ~0.50x trailing.

Uber paid just 0.43x GMV and ~8x EBITDA for Delivery Hero, a delivery business many times larger than every other comp in the set (other than JET).

Simple “Gut-Check” Model

If we assume:

• ~15% total bookings CAGR through 2032

  • Mobility bookings was 20% in Q1’26 on a CC basis (ex DH)

  • Delivery bookings was 23% in Q1’26 on a CC basis (ex DH)

• EBITDA as a % of GMV grows from ~4.7% in 2025 (pro forma for DH) to 6.50% in 2031

• Uber keeps 30% of the EBITDA it generates as cash to the balance sheet

• A 12x LTM EBITDA multiple

This results in a ~3.0x MOIC and a ~20% IRR over 6 years.

This assumes everything more-or-less goes according to plan. In theory, it gives investors a healthy margin of safety to work with.

Management + Ownership

Leadership

Dara Khosrowshahi (CEO, 56 YO)

Dara has been the CEO of Uber since 2017.

He was previously the CEO of Expedia from 2005 to 2017, and before that the CFO of IAC Travel. Before joining IAC, Dara served as a VP at Allen & Company, a firm he joined as an analyst.

Dara had the good fortune of serving at IAC/Expedia under the leadership of Barry Diller. Dara met Barry while at Allen & Co. Over time, Dara became one of Barry’s go-to bankers as Barry built up IAC through acquisition (and pivoted from a focus on cable TV networks to internet properties). Dara eventually joined IAC as Barry’s “deal guy,” before eventually taking the reins at Expedia (which IAC spun out in 2005 as a standalone, publicly traded company).

At Expedia, Dara was apparently named one of the Highest Rated CEOs on Glassdoor. He still serves on the Board of Expedia and he was previously on the board of the New York Times Company.

Dara was originally born in Iran. He left the country with his family during the Iranian Revolution when he was 9 years old. His family eventually settled in Tarrytown, New York. Dara earned a bachelor’s degree in engineering from Brown University.

Dara’s a likable, self-proclaimed “consensus-builder” type. He strikes me as high quality manager with a strong strategic sense. Interestingly, he’s not only a second-time CEO, but a second-time marketplace CEO. I believe this has its benefits.

Clearly something that was seared into Dara’s brain from his time at Expedia was (1) the power of keeping take rates low and (2) the importance of being a “supply-led” marketplace. While Expedia had a good run under Dara’s leadership, it’s main competitor — Booking.com — did much better. Booking.com overtook Expedia during this period and now boasts a market cap of ~$130bn (vs. Expedia’s of ~$30bn). When Dara took over at Expedia, it had a “demand-led” approach, advertising heavily to build demand while maintaining a premium “take rate.” Booking.com, in contrast, kept take rates lower while focusing on building best-in-class hotel inventory supply.

Dara on his lessons from competing against Booking.com from a Nov-25 HD in HD podcast:

Interviewer: “If we look at the Expedia story…I would say elephant in the room is the whole Booking[.com] situation, how that evolved. Now with the benefit of hindsight maybe explain what happened, what is your version of why Booking is where it is versus Expedia?”

DK: “Yeah definitely, painful story. And you know, during my years both companies did well but Booking did better. I think two lessons there.

One is that if you’re building an aggregation platform, supply is king. Expedia, Hotels.com built more — it was more about audience. “Hey, how do I build audience, how do I build a brand and then if I build an audience and a brand, what supply can I bring on board to fulfill that demand? Booking.com started with supply first. Which is, ‘Hey, I’m just going to go out and add every single hotel in a particular marketplace and each hotel is another opportunity to sell to a new audience.’ And if Expedia had 50 hotels in a market and if Booking.com had 200 hotels in a market, that market would convert better for Booking.com than Expedia. So supply, you know, running a supply business was a lesson for me.

And the second was actually starting from a low margin position. Expedia, at the time when I joined, had a take rate of about 30%. Booking.com had a take rate of 15%. We thought 30% was better because we could make more money per transaction, we could put that money into advertising, build more audience. And we would beat Booking.com at the audience game. But what Booking got right was, because they were charging 15%, they were able to sign up more supply, and again, supply brought them audience.

So the two lessons are, one is start from a low margin position and build up your margins over time and they have at Expedia. We had to reduce our margins over a period of time. And then second is supply is king…I think in hindsight we got the supply memo a bit too late, a couple of years too late. And then once we started organizing against it, we had a great five, six year run against Booking.com and Priceline.

The toughest part was actually the margins. Taking down [take rates] from 30% to 15%. Literally cutting your revenue as a public company, incredibly painful…You can identify what you’re losing, which is a huge amount of profitability when you cut margins. You can’t identify what you’re gaining. You know, a supplier may be a little bit less angry at you, for example. It’s just hard to identify the return from cutting your margins. And that was again 6, 7-year journey that we went through cutting revenue margin, cutting revenue margin, that was really painful as a public company. But we had to do it. In hindsight, I wish we had done it faster.”

On many occasions Dara has stressed the importance of keeping take rates low.6364 Here is Dara on the dangers of raising take rates, from a Jun-23 Acquired podcast:

“High take rates are dangerous. Our job as a company is to grow volume as much as we can as fast as we can and make our shareholders happy enough, minimizing the take rate. Taking as much of that dollar and and giving it to drivers and couriers. Last quarter, gross bookings grew over 22% or so which is really good. The money that drivers and couriers, including tips, made on the platform grew by 30%. And at the same time we were able to expand our margins…Sometimes it is torture. Watch every single nickel and dime. Be incredibly efficient in everything that you do. Automate everything. Get fraud out of the system, etc. So that you can actually operate a business at scale at the lowest take rate possible. Talking about Booking.com…when I started at Expedia — Expedia’s take rate was 25 percent and Bookings’ take rate was 15. And over torturous 13 years we took Expedia’s take rate from 25 to the teens — it was like 17 or so when I left. And those are like pure margin dollars that you’re taking out. Like there’s no goodness that comes out of it and so it’s just really hard work to do…

You don’t want to put yourself in that position. It’s very tempting. It’s very, very easy. This is the temptation. Obviously this quarterly treadmill that you’re on, etc. You can make someone happy by increasing take rate and throwing it to the bottom line and we really, really culturally try to resist that notion.”

Overall, Dara and team have a “shared-savings” bent in how they are positioning the business, which strikes me as a strong positive. Uber is using its scale to reduce costs for the consumer. Here is Dara from a Mar-25 Morgan Stanley conference:

“So I think the team, first of all, is executing really well. But the biggest area that I believe we should do better at is just cost, right?…

Building out a true low-cost product, shared product, getting 2 or 3 people into a car is a very significant challenge in -- operationally, algorithmically. We’re making progress there. So I’m happy about the progress there, but this is -- I think we will be possibly the only company in the world that’s going to solve this at scale. To be clear, we’re losing money in share, but the loss rates are coming down. The efficiencies continue to increase, and I’m quite confident of the activity of that team.

When you look at Uber Eats, for example, we are -- one of our biggest programs is what we call merchant-funded offers. So actually, merchants providing buy one, get one free, if you spend $30 on a basket, gets a $10 item for free, et cetera. These are promotions that these merchants run. Those promotions give merchants more access to our audience, and merchant-funded offers are now running at over $1 billion of, call it, savings for our eaters per year as well.

And then, of course, for us, the membership program is a giant discount program, so to speak. We’re trading higher frequency, higher reliability for discounts to our consumers as well. It is absolutely working. But near term, to be clear, membership is an investment, right? So the discounts come through before the frequency increases come through, the retention comes through. But we’re now seeing memberships. One of the reasons why you’re seeing Uber Eats growth rate actually accelerate in addition to the newer businesses like grocery, et cetera, is that you’re seeing the frequency and the retention from our membership program come into play as the cohorts mature as well.

So for us, just pricing to the end consumer is an area of real focus for us. And I’d like to be making more progress there than we have been in the past couple of years.”

This type of long-term mentality and stewardship is good to see.

Dara seemingly has an almost ideal background for his current role. He has an engineering degree, but also a great deal of finance expertise (banking at Allen & Co, former CFO of IAC Travel). He has extensive M&A experience, which mostly came under the tutelage of famed dealmaker Barry Diller. This experience is coming in especially handy as Uber cuts a myriad of deals across the AV landscape (in addition to other M&A). Dara also has extensive public CEO experience from his time at Expedia (which is also a marketplace, like Uber), where he also replaced a founder-CEO in Rich Barton.

Track Record

Overall, I’d say Uber has executed well since Dara took the reins in 2017. He was taking over a business that was well-positioned, of course. But nonetheless, I think he’s put up a very solid track record.

Since 2017, Revenue has almost 7x’d, a 27% CAGR. Meanwhile, the profitability of the business has inflected majorly. From -$2.6bn in adj EBITDA in 2017, the company generated $8.7bn of adj EBITDA in 2025. In Q1’26, GB grew by 21% (on a CC basis) and the company generated $1.9bn of GAAP EBIT (~15% of rev).

The team has expanded Uber’s products and geos served in sensible ways. For example, Uber introduced new products like “Reserve,” “Wait & Save,” Uber for Business (U4B), Uber Teens, and Moto (in developing countries). These products are growing significantly faster than the core business. For example, U4B is growing more than twice as fast as mobility overall.65 Uber has expanded into grocery, retail and direct on the delivery side. They’ve continued to enter additional countries where it made sense (such as South Korea, Germany, Spain, Italy, and Argentina, which are some of their fastest growing markets today). They’ve quickly grown a large advertising business, which has transformed the profitability of Uber’s delivery segment.

Dara has also overseen the divestments of a number of highly cash-burning segments with murky futures. These included the divestments of (1) Uber’s SE Asia business, (2) Uber Eats India, (3) Uber Elevate (it’s air taxi technology division), and (4) ATG (it’s self-driving tech division). Before Dara joined, the company also divested its China business and its Russia business. Many of these exited markets are notoriously competitive, and I don’t think it’s unreasonable to think that, had Uber stayed, it might still be slugging it out today with various competitors in these regions. Altogether, these moves have contributed to a much healthier level of profitability for Uber.

Perhaps the most controversial of its divestments was that of the ATG division, or its self-driving tech division. There were a number of reasons why Uber elected to divest it, including: (1) the segment was burning a lot of cash (and Uber needed to cut costs, especially in the early COVID period), (2) safety concerns/risk (one of Uber’s self-driving test vehicles hit and killed a pedestrian in 2018), (3) by not having its own self-driving tech (i.e. being a neutral third party), Uber was better positioned to work with almost every other AV provider in the ecosystem, and (4) perhaps most importantly, Uber viewed its core strength as its marketplace business and technology (not cutting edge AV technology), and that it was best for Uber to focus on its core competencies.

Here is Dara from a Nov-25 HD in HD podcast explaining the decision to cut the ATG segment:

“We cut back on AV during COVID, which was an emergency for the company. We went from losing two billion to losing three and a half billion or so. And we only had so much time to get to profitability. We were going to run out of cash. So there was a necessity there to some extent. But the other factor for me is that companies have cores that they’re great at and we were a software company. We’re great at building algorithms, building search, matching, etc. We weren’t a great hardware company and hardware is really tough. The cycles are different. It’s multi-year cycles versus you know multi-week cycles, so to speak, which is how we operate. So I kind of concluded that it wasn’t something that we were truly going to be great at. There was a set of companies that were just working on AV and it was life or death for them. For us it was important but it was a little bit of a side project. And then the necessity of it forced us to get out.

The other element that was really important to me was that because we were building our own AV, no one else in the ecosystem wanted to work with us because they were competitive with us. And so our hypothesis was that AV was not going to be winner take all. There are going to be multiple players who were going to win. And in order to play with the greatest talent base in the ecosystem, we needed to become a neutral party and that we would actually go and actively fund the ecosystem and build up the ecosystem. And help the ecosystem grow versus trying to build our own competitor. So I think it’s the right decision, but time will tell.”

Uber also sold off this division in the “pre-AI” era. Self-driving was a much harder problem to solve in the pre-AI era as, according to my understanding, you basically had to program in rules/logic for driving and to solve edge cases (yikes) vs. “just” using an end-to-end neural net. So this would have made the decision easier.

Regardless, I think the decision to sell was probably the right one for the reasons listed above. This is not to mention that getting the business to a very profitable state unlocked for Uber the ability to secure AV supply for their platform. Getting the business to a very profitable state, in fact, may arguably have been much more important in contributing to Uber’s AV capabilities than anything else.

Bill Gurley would tell you Uber should have taken its self-driving tech and kick-started an open source project (vs. selling it). I’m not one to argue with Bill Gurley, so perhaps that was a miss. From a Feb-26 Stratechery interview:

BT: “Do you have any takes on Uber versus autonomous vehicles? Are they going to be okay?”

BG: “Well, I will give you one take. I have been pushing the company for over 10 years to embrace open source and I think it would’ve been the best thing they could have possibly done. They took the assets they had at the time and sold them into, I can’t even remember the name of the company, and they gave that company money and so if that IP had instead kick-started an open source project.”

BT: “They’d have way more potential suppliers.”

BG: “Yeah, and you commoditize that layer. Part of what we’re seeing in this AI war, and Google got really good at this not only with Android but Kubernetes, like you use open source to commoditize another layer so that your layer is protected as much as possible.”

I think we are sort of getting open-source technology anyway via NVIDIA, however, so perhaps it won’t make a big difference in the end.

I’d probably characterize the team’s M&A track record as “okay.” Uber paid $7bn+ for Cornershop, Postmates and Drizly, which were perhaps overpays but helped Uber kickstart its delivery business in the immediate COVID aftermath. It paid $3.1bn for Careem, which helped cement Uber’s rideshare leadership in the Middle East (this did not translate to delivery, however — talabat and HungerStation dominate the Middle East in delivery). Uber’s most recent acquisitions have been sensible looking tuck-ins: Foodpanda Taiwan (was ultimately blocked), Trendyol GO, Getir’s food delivery biz, SpotHero and Blacklane. Perhaps its most clear miss was paying $2.25bn for Transplace, which doesn’t seem like a long-term winner at this point (Uber Freight is currently ex growth and unprofitable). Most recently, in Jul-26, Uber acquired Delivery Hero for ~$18bn, a deal I like for Uber as discussed previously.

Another area in which you could potentially knock management is their relative tardiness in going after the suburbs, particularly on the delivery side. This opened the door for DoorDash to entrench itself. Uber is now attempting to compete more aggressively in the suburbs, but DoorDash is quite formidable at this point.

Something Uber has done well, and that should greatly assist Uber in the suburbs, is going all in on Uber One. They have smartly priced the membership at the same level of DashPass while bundling in more rewards than anybody else (it seems clearly the superior membership to have in the delivery/mobility world). It’s a “shared-savings” approach, leveraging Uber’s scale and breadth of offerings. I like how they are recycling most/all of the dollars they are generating through their new partnership with Expedia towards additional benefits for Uber One members. I like how they intend to offer additional parking-related benefits for Uber One members by way of their acquisition of SpotHero. They are creating an all-things transportation bundle that has real attractiveness. Even if Uber never makes a profit from hotels for example, these types of moves increase the attractiveness of Uber One, which will feed Uber’s mobility and delivery businesses.

In terms of the autonomous transition, Uber is clearly trying to be on its front foot. I like that they’ve been aggressive using their balance sheet to secure AVs for the Uber platform. Uber Autonomous Solutions was an important step and builds nicely on Uber’s existing capabilities. I agree with the thought that AV fleets will mostly be “financial-ized” at some point, and that remaining an asset-light platform is the ideal future course.

One thing that stands out very positively, inspired by this tweet from K Capital, is Uber’s pace of execution. All of the following things have happened since just February: (1) reported 21% bookings growth (on a CC basis) and nearly ~$2bn of GAAP EBIT in Q1’26, (2) acquired SpotHero, (3) acquired Blacklane, (4) acquired Getir’s delivery business in Turkey, (5) made a highly complex “carve-out” acquisition of Delivery Hero (and executed on a series of machinations to build a stake in the business prior to the deal), (6) expanded its stake in WeRide, (6) unveiled Uber Autonomous Solutions, (7) launched Europe’s first commercial robotaxi service in Zagreb in partnership with Pony.ai, (8) participated in Wayve’s $1.5bn Series D, (9) made a $300m investment into Rivian and committed to buying between 10-40k R2 robotaxis, (10) made a $200m additional investment in Lucid while upp’ing its vehicle commitment from 20k to 35k, (11) announced a partnership with NVIDIA to launch robotaxis in 28 cities by 2028, (11) partnered with Hertz to power fleet management for Uber’s upcoming launch of Nuro/Lucid robotaxis in the SF Bay Area (13) unveiled hotel bookings through the Uber app in partnership with Expedia, (12) announced additional partnerships with Joby Aviation, Baidu, WeRide, Zoox, Motional, MOIA, Wayve and (14) are infusing AI-driven efficiencies throughout its business functions. This ability to do many things well at once surely is a positive indicator in terms of management quality.

Overall, Dara & team strike me as a good stewards of the business. They’ve transformed the profitability of the business, while maintaining very healthy growth. They’re working hard to be on front lines of autonomy. They’re working hard on making the Uber One membership as attractive and sticky as possible. I feel pretty confident they will allocate capital at least reasonably wisely.

Broader Management Team

The broader management team seems pretty good on the whole.

• Andrew Macdonald (President & COO, 41 YO). Andrew was appointed President and COO in Jun-25. He previously served as Head of Global Mobility from 2019 to 2025, and held various general management and regional leadership roles at Uber from 2012 to 2019. Prior to Uber, he was a consultant at Bain & Co from 2007 to 2010 and later co-founded multiple early-stage private companies. In this interview on The Compound podcast, I thought he squeezed in a large number of insights into just 30 min.

• Balaji Krishnamurthy (CFO, 41 YO). Balaji was recently appointed CFO in Feb-26. He previously served as VP, Strategic Finance from 2023 to 2026, and Head of Investor Relations from 2020 to 2023. Prior to joining Uber, he was VP of Equity Research, US Hardware and Communications Technology at Goldman Sachs, from 2011 to 2019. I’ve enjoyed Balaji’s commentary on recent earnings calls

• Sachin Kansal (Chief Product Officer, ~47 YO). Sachin is responsible for the company’s Mobility and Delivery products, overseeing product management, design, and product operations. He also oversees product and technology strategy for some of Uber’s new initiatives such as autonomous vehicles, sustainability, taxis, and Uber for Teens. He joined Uber in 2017 as the company’s first product leader focused on safety technology. Sachin was previously the VP of Product at Lookout, a leading mobile security company, where he managed their Consumer product line and scaled the business to 120M+ users. Before that, Sachin served as Chief Product Officer at Flywheel Software, a provider of on-demand transportation through taxicabs. He spent the early part of his career at Palm (acquired by HP), where he was the Director of Product Management focused on Palm’s mobile operating system webOS and mobile applications. This WSJ interview from Sachin is a good listen, and I’ve also enjoyed his appearances in Uber’s “Go-Get” annual product showcases.

• Jill Hazelbaker (Chief Marketing Officer, 44 YO). Jill has served as CMO and SVP, Public Affairs since 2019. She was SVP/VP, Communications and Public Policy from 2015 to 2019. Prior to Uber, she was VP, Communications and Public Policy of Snap Inc. from 2014 to 2015. From 2010 until 2014, she held senior communications and public policy roles at Google. Prior to joining Google, she served as Press Secretary to Mayor Michael Bloomberg’s re-election campaign in New York City in 2009 and as the Communications Director for Senator John McCain’s U.S. presidential campaign from 2007 to 2008.

• Tony West (SVP, Chief Legal Officer, 60 YO). Tony has served as SVP, Chief Legal Officer and Corporate Secretary since 2017. Prior to joining Uber, he was EVP, Government Affairs, General Counsel and Corporate Secretary from 2014 to 2017 at PepsiCo. Prior to PepsiCo, he served as the 17th Associate Attorney General of the US from 2012 to 2014, after previously serving as the Assistant Attorney General for the Civil Division in the U.S. Department of Justice from 2009 to 2012. From 2001 to 2009, he was a partner at Morrison & Foerster LLP.

Board of Directors

• Ron Sugar (Chairman, 77 YO). Ron has served as the Independent Chairperson of the Board since 2018. He was Chairman and CEO of Northrop Grumman, a global aerospace and defense company, from 2003 until his retirement in 2010, and President and COO from 2001 to 2003. He was President and COO of Litton Industries, Inc. from 2000 until the company was acquired by Northrop Grumman Corporation in 2001. Prior to that time, he served as CFO of TRW Inc. He is an adviser to Bain & Co and was formerly an adviser to Ares Management and to Singapore’s Temasek Investment Company. He is a trustee of the University of Southern California, board of visitors member of the University of California, Los Angeles Anderson School of Management, past Chairman of the Aerospace Industries Association, and a member of the National Academy of Engineering. Dr. Sugar currently serves on the board of Apple. He previously served on the board of Air Lease Corporation from 2010 to 2020, the board of Chevron from 2005 to 2023, and the board of Amgen from 2010 to 2024. Ron and Dara have a standing weekly hour-long call primarily focused on longer-term strategic direction and goals.

• Revathi Advaithi (58 YO). Revathi has served on the Board since 2020. She has been CEO of Flex Ltd., the third largest EMS provider in the world, since 2019. Prior to Flex, she was President and COO, Electrical Sector, of Eaton Corporation plc, a power management company, from 2015 to 2019. Prior to that role, she held senior management roles at Eaton from 2008 to 2015. Between 2002 and 2008, she worked at Honeywell, where she held several senior roles within the sourcing and supply chain functions of the aerospace sector before being named Vice President and General Manager of Honeywell’s Field Solutions business in 2006. She held various other roles at Eaton between 1995 and 2002.

• Turqi Alnowaiser (49 YO). Turqi has served on the Board since 2023. He has served as Deputy Governor and Head of the International Investments Division at The Public Investment Fund, a sovereign wealth fund of Saudi Arabia, since 2021, and previously served as Head of International Investments at The Public Investment Fund from 2016 to 2021. He formerly held several executive roles at Saudi Fransi Capital, a leading financial services firm based in Saudi Arabia, including as Head of Asset Management. Before his career at Saudi Fransi Capital, he specialized in developing, managing, and regulating various financial products across asset classes at Morgan Stanley, the Capital Market Authority of Saudi Arabia, and the Saudi Industrial Development Fund. He has served on the board of Lucid Group since 2019 and on the board of Hapag-Lloyd AG since 2018.

• Nikesh Arora (58 YO). Nikesh has served on the Board since 2025. He has served as the Chairman and CEO of Palo Alto Networks, a leading global cybersecurity company, since 2018. Prior to joining Palo Alto Networks, from 2016 through 2018, he was an angel investor and from 2016 through 2017, he served as an advisor to SoftBank. From 2015 through 2016, he served as President and COO of SoftBank and from 2014 through 2015, he served as vice chair and CEO of SoftBank Internet and Media. Prior to SoftBank, from 2004 through 2014, he held multiple senior leadership operating roles at Google, Inc., including serving as SVP and Chief Business Officer, from 2011 to 2014. He also serves on the board of Compagnie Financiere Richemont S.A., a public Switzerland-based luxury goods holding company. He previously served on the boards of SoftBank, Sprint Corp., Colgate-Palmolive Company, and Yahoo! Japan, among others.

• Ursula Burns (67 YO). Ursula has served on the Board since 2017. In 2021, she co-founded Integrum Holdings, an investment firm focused on partnering with technology-enabled services companies. Previously, she was Chairman of VEON Ltd., an international telecommunications and technology company, from 2017 to 2020 and CEO from 2018 to 2020. She was Chairman of Xerox Corporation from 2009 to 2017 and CEO from 2009 to 2016, prior to which she advanced through many engineering and management positions after joining the company in 1980. She currently serves on the boards of IHS Holding Limited and Taiwan Semiconductor Manufacturing Company Ltd. She previously served on the boards of Endeavor Group Holdings, Inc. from 2021 to 2025, American Express from 2004 to 2018, Nestlé S.A. from 2017 to 2021, and Exxon Mobil Corporation from 2012 to 2023.

• Robert Eckert (71 YO). Rober has served on the Board since 2020. He has been an Operating Partner of FFL Partners, a private equity firm, since 2014. He is also Chairman Emeritus of Mattel, a role he has held since 2013. He was Mattel’s Chairman and CEO from 2000 until 2011, and he continued to serve as its Chairman until 2012. He previously worked for Kraft Foods, Inc. for 23 years, and served as President and CEO from 1997 until 2000. From 1995 to 1997, Mr. Eckert was Group VP of Kraft Foods, and from 1993 to 1995, he was President of the Oscar Mayer foods division of Kraft Foods. Mr. Eckert currently serves on the boards of Amgen, Levi Strauss & Co., and Quinn Company. He previously served on the board of McDonald’s from 2003 to 2023.

• Amanda Ginsberg (56 YO). Amanda has served on the Board since 2020. She has been an Operating Partner of Advent International, a global private equity investor, since 2022. She was CEO of Match Group from 2017 to 2020. Prior to this role, she was CEO of Match Group Americas from 2015 to 2017, where she was responsible for the Match U.S. brand, Match Affinity Brands, OkCupid, PlentyOfFish, ParPerfeito, and overall North and South American expansion. Previously, she was the CEO of The Princeton Review from 2014 to 2015, where she expanded its services to include online services, including tutoring and college counseling. She currently serves on the board of ThredUp Inc. and the board of Universal Music Group. She previously served on the boards of Care.com from 2012 to 2014; J.C. Penney from 2015 to 2020; Match Group from 2017 to 2020; and Z-Work Acquisition Corp. from 2020 to 2022.

• John Thain (70 YO). John has served on the Board since 2017. He is the Founding Partner of Pine Island Capital Partners, a private investment firm, and has served as Chairman since 2017. He was Chairman and CEO of CIT Group from 2010 until 2016. In 2009, prior to joining CIT Group, he was President of Global Banking, Securities and Wealth Management for Bank of America. From 2007 to 2009, prior to its merger with Bank of America, Mr. Thain was Chairman and CEO of Merrill Lynch & Co. From 2006 to 2007, he was CEO and a director of NYSE Euronext, Inc. following the NYSE Group and Euronext N.V. merger. He joined the New York Stock Exchange in 2004, serving as CEO and a director. From 2003 through 2004, Mr. Thain was the President and COO of Goldman Sachs, and from 1999 through 2003 he was President and Co-Chief Operating Officer. From 1994 to 1999, he was CFO and Head of Operations, Technology and Finance, and from 1995 to 1997 he was also Co-Chief Executive Officer for European operations for Goldman Sachs. He currently serves on the supervisory board of Deutsche Bank AG. He previously served on the board of Goldman Sachs Group Inc. from 1998 to 2004.

• Alexander Wynaendts (65 YO). Alexander has served on the Board since 2021. From 2008 to 2020, he was CEO and Chairman of the management and executive boards of Aegon NV, one of the world’s leading providers of life insurance, pensions, and asset management. Prior to Aegon, Alexander began his career in 1984 with ABN AMRO Bank, working in Amsterdam and London in the Dutch bank’s capital markets, asset management, corporate finance, and private banking operations. He currently serves on the board of Air France-KLM SA and the Supervisory Board of Deutsche Bank AG, where he serves as Chairman. He formerly served on the board of Citigroup Inc. from 2016 to 2021.

Notable Departures

• Travis Kalanick & other early employees (Garrett Camp, Ryan Graves, Emil Michael, etc.). Kalanick is now CEO of Atoms, a robotics company focused on food, mining and transportation.

• Sundeep Jain. Former Chief Product Officer and SVP Engineering from 2018-2024. Now President of Mercor.

• Pierre-Dimitri Gore-Coty. Former SVP of Uber Eats from 2021-2025. Joined Uber in 2012. Now GP at VC firm Plural.

• Prashanth Mahendra-Rajah. Former CFO from 2023-2026. Now a Senior Policy Advisor at the US Department of Commerce. Member of the Board at Shopify.

• Nelson Chai. Former CFO from 2018-2024. Now CEO of DailyPay. Member of the Board at Chubb, Thermo Fisher Scientific.

Uber Culture

Uber culture has been the topic of much discussion over the years.

In the early days it was pretty toxic, I think it’s fair to say. Hard-working, but also very much “kill or be killed.” Sharp elbows and stepping on toes were encouraged. HR complaints were not followed up on. Misogynistic behavior festered. Borderline illegal competitive behavior was encouraged. This “win at all costs” culture helped power Uber to great heights as the undisputed global leader in the industry, but it also caused a self-implosion.

With the appointment of Dara as CEO, amongst other things, the Board sought something of a cultural reset. Here are Uber’s official values today:

From interviews, it seems Dara, after joining, tried his best to retain Uber’s hard-working culture. He takes pride in it. He seems fairly unafraid to “push out” those that aren’t willing to put in the hours. From a Feb-26 DOAC podcast:

“So part of working hard is like, you know, sending emails to the team on a Saturday and if I don’t get a response on Saturday, sending them an email on Sunday with a question mark. ‘What’s going on?’

You know, I think at Expedia in hindsight — we worked intensely and we went hard, but but not as hard as I like. Because Expedia was, we were selling vacations, right? The product that we were selling was about turning yourself off. And so we did talk about work-life balance. And in hindsight, at Uber, I don’t.

You come to Uber, you’re going to work your ass off. We’re going to be really demanding. If you’re not performing, we’re going to let you know. And if you don’t fix it, we’re going to push you out.

But while it will be incredibly hard, you will have real agency at the company. We’re a big company, but individuals can make a big difference. And it’s a company that’s making a difference in the world. You’re going to learn so much. And while you will have worked hard, you’re going to have a great time. But this is, don’t come here if you want to coast. And I’m very clear about that. And I should have been more clear at Expedia.”

He’s also tried to retain a culture that encourages disagreement and debate (but where taking that too far, or “being an asshole,” is not). He seeks out truth tellers66, and tries to lead by example along that dimension. He’s candid in interviews almost to a fault.

Uber’s Chief Product Officer Sachin Kansal espouses a deep understanding of the customer (both quantitatively and qualitatively, including frequently dogfooding the product) and shipping quickly (or as he calls it, “ship, ship, ship”).

Glassdoor has Uber rated a 3.7/5.0 on 16,151 reviews (though many of which are from drivers/couriers). 71% “approve” of Dara. Only 53% have a “positive business outlook.”

My sense is that, for the right person, Uber can still be a very exciting place to work. Millions of people use the product in everyday life all across the globe. Navigating the autonomous transition is interesting, challenging, and offers the prospect of large future rewards. There are difficult technical challenges to solve. And it’s still a magical product in many ways.

Ownership

Management and director ownership is quite low on a percentage basis, which is not ideal. That said, the dollar-value of the shares owned by insiders is still quite meaningful.

The relatively rare and always welcomed CFO insider purchase took place in Feb-26, when CFO Balaji Krishnamurthy bought $1.6mm in stock on the open market. This is the only major insider purchase over the last 2 years from what I can tell.

Management Compensation

Management compensation is mostly a combination of (1) base salary, (2) annual cash bonus targeted at 100-200% of base salary (that is awarded based on a mix of company-specific and individual-specific annual goals), and (3) equity incentives consisting of a mix of performance-based RSUs (PRSUs), time-based RSUs, and stock options.

PRSUs and stock options account for ~3/4ths of targeted equity comp for Dara and Balaji. Less so for Andrew (~one-half), Jill (~1/3rd) and Tony (~1/3rd).

Annual cash bonus is derived from a mix of company goals (60% financial / 40% strategic & operational priorities) multiplied by an “individual modifier” (that is based on the achievement of individual/functional goals). Here is how the company goals were calculated for 2025:

PRSU “achievement” (or vesting) is determined at the end of 3 years. 80% of the award is based on the achievement of key financials goals/metrics, and 20% is based on the achievement of long-term strategic goals and metrics. The entire award is then subject to a TSR multiplier (ranging from 0.7x to 1.3x) based on Uber stock’s performance vs. the S&P500.

The company is moving from using an adjusted EBITDA figure to a Non-GAAP EBIT figure (including the cost of SBC) as one of its metrics for measuring financial performance, a positive development.

Thomas Reiner of Altimeter points out that the company has been setting performance targets ahead of street consensus in recent years — certainly a “green flag:”

He also notes that dilution has been relatively tame for the company in recent years.

Capital Allocation

Capital allocation at Uber today mixes a few different priorities: (1) investing in organic growth (less mature products & geos), (2) seeding the Uber platform with AVs and generally helping to advance the AV ecosystem, (3) acquisitions, and (4) returning capital to shareholders in the form of buybacks.

Here’s the company’s “official” capital allocation framework, laid out at its 2024 investor day:

Because of Uber’s recent inflection in profitability, the company can actually afford to do all of the above simultaneously, according to management.67 Here is CFO Balaji Krishnamurthy from Q4’25 earnings:

BK: “Sure. So we did lay out our capital allocation priorities in quite a lot of detail. But just to quickly summarize how we think about this. Our first priority is to ensure that we are making appropriate reinvestments behind the opportunity we’re seeing in our core business. We are in a good position where even as we make those investments, we are throwing off a lot of cash.

As we said, we were already generating about $10 billion of free cash flows, growing 40% as of the last year. So that gives us a lot of room to make investments in ensuring that we are advancing our AV strategy and potentially evaluating any selective bolt-on M&A opportunities as they come along. And then that still leaves us with a significant amount of cash that we can return to shareholders. So this is not a trade-off for us in the sense that we are choosing one or the other. We’re able to do all of these things in parallel.

As to your question on whether we would be returning 50% of free cash flows, based on our current visibility into what we’re seeing as well as the fact that our stock remains really cheap, we will continue to be aggressive buyers of our stock, and you should expect that it continues at a steady cadence, and we are on track to reducing our share count by a healthy amount as we go forward.”

DK: “And I think the good news here is with our free cash flow generation and our expectation of the free cash flow generation increasing going forward, we can do both. We can invest appropriately as it relates to growth. And then at the same time, we are going to continue reducing the share count because ultimately, all of us are shareholders, and we think right now the opportunity to buy back shares is pretty awesome.”

Below is a list of investments Uber has made in the AV ecosystem. This is on top of all of the partnerships that Uber has struck (e.g. WeRide in Dubai/Abu Dhabi, Pony in Zagreb, Apollo Go in London, Nuro in SF, etc.).

By my count, Uber has committed $1.6bn+ in the form of equity investments and another ~$3.5-4.6bn in the form of vehicle commitments (or $5-6bn+ in total). The vehicle commitments are over 5+ years, and are based on partners hitting certain self-driving milestones (so they may not all come to fruition). The FT recently published an article stating that Uber has committed more like ~$10bn (~$2.5bn in equity / ~$7.5bn in vehicle commitments), so I could be short. The FT article does not provide a breakdown by investment.

It should be noted that Uber’s largest investment cumulatively — $500mm and a 35k vehicle commitment to Lucid — has perhaps already gone sideways. It was reported in Jul-26, just two months after Uber’s second investment in the company, that Lucid had brought in turnaround advisors AlixPartners. This sent Lucid stock down by more than half on the day. Lucid denied it was filing for bankruptcy, saying “the rumors are completely false” and that “the company has sufficient liquidity to carry its operations well into next year.” Lucid laid off 18% of its workforce in Jun-26 which followed a 12% cut in Feb-26.

As discussed previously, while Uber is using its balance sheet and cash flow to seed its platform with AVs, the company is clear that the objective is to remain asset-light longer-term. Whether and how soon Uber can dial back the capex remains to be seen.

Buybacks

Uber’s buyback activity has coincided with a material ramp in its adj EBITDA generation, which has gone from ~breakeven in 2022 to almost $9bn in 2025. Uber authorized its first share repurchase program of $7bn in Feb-24 and then a second $20bn program in Aug-25.

The company first began buying back shares in Q2’24, and ramped buyback activity significantly in 2025. With Uber’s share price falling ~25% over the course of Q4’25/Q1’26, Uber ratcheted up repurchases further in Q1’26 (to the tune of $3.0bn in the quarter).

Management has laid out an expectation that Uber’s share count should fall for at least the next few years. Management has also stated that while they intend to buy back stock consistently every quarter, they will be opportunistic and ramp spend on buybacks when they see a “dislocation” in price. From Q2’25 earnings:

“Maybe a couple of points on the buyback to help folks understand how we think about it. As this business has inflected, and we have started to generate meaningful cash flow, returning that cash to our shareholders is a key priority for us. We’ve already executed over 60% of our authorization from last spring, when it was originally authorized.

So today’s $20 billion is in addition to the roughly $3 billion that is yet to be executed. So I know that sometimes folks get confused on that. So think of it as $23 billion to execute over the next couple of periods here. That represents about 12% of our market cap and really is a reflection of how great we feel about the cash flow generation that’s in front of us. So if you look at our history now, we’ve been allocating around 50% of our free cash flow to buybacks.

I think that’s a fair sort of way for you to think about how we will execute the capital return over the coming years. That gives us also a good sort of way to benchmark how we want to design our programs every quarter. So you should expect this to be sort of a multiyear plan. We will be active every quarter. But of course, we always reserve the opportunity that if there is a meaningful dislocation, we’re going to get very opportunistic in the market.

And just a reminder that we made a commitment last year in our Investor Day that we were going to turn the curve and start reducing our share count. And now in the second quarter, we’ve actually taken share count down 1%, and you’ll see that trend continue for the next couple of years.”

Q1’26, I think it’s fair to say, was one such “dislocated” period in management’s view.

Leverage

The company introduced a 2x EBITDA gross leverage target at its 2024 investor day and a strong desire to remain “investment grade” (i.e. rated BBB or higher; Uber debt is rated BBB currently).

Post Delivery Hero acquisition, Uber will have ~$26bn of debt ($10.5bn as of Mar-26 + $16bn new bridge facility) vs. an expected ‘27 EBITDA of ~$16bn. This equates to ~1.6x of gross leverage.

However, if we take into account Uber’s ~$4bn of cash (~$6bn as of Mar-26 less ~$2bn to help fund DH), and we take into account Uber’s ~$8bn of investments (of which $2.6bn, $2.5bn and $2.1bn are in Didi, Grab and Aurora, respectively), Uber’s net leverage will be ~0.9x. The investments, of course, aren’t as liquid as cash. But Didi, Grab and Aurora are all publicly listed.

In all, Uber is positioned conservatively in terms of financial leverage. However, any leverage at all for a company in a rapidly changing industry like Uber’s (and where competitors include the likes of Google, Amazon and Tesla) makes me a little nervous. My hope would be that Uber prioritizes debt repayment over the next couple years.

M&A

Dara has expressed a strong preference for organic growth vs. inorganic growth.68 I can appreciate the sentiment, especially coming from an ex-banker and “deal guy.” From a Mar-26 Semafor podcast:

“I think the best M&A strategy is not to have to buy anything. Deals are hard. The majority of deals fail. So my job, number one, is to build out a path for organic growth. And we’ve grown over 20% for I think four or five years now. That continues for the next 3 to 5 years where I can actively plan for…My job is to set that reality so that any M&A that we do is essentially optional. And on top of that we tend to look at M&A in areas that are adjacent to us. M&A doing the same thing that we do, expanding into different countries for example. We bought one of the leading food delivery players in Turkey, which has been an enormous success. So M&A will always be a complimentary strategy, but my main job is to make sure I never need to buy anything and the business keeps growing at very, very healthy rates.”

Below are Uber’s most notable acquisitions and divestitures.

Most of them seem defensible, though it’s hard to know for sure without seeing the underlying numbers.

As discussed previously, I like the Delivery Hero acquisition for Uber. The deal (1) adds significant scale to Uber (22% increase in bookings), (2) was done at a reasonable multiple (~8x ‘27 EBITDA), (3) adds leading markets (38 of the 50 markets added are in #1 market positions) including crown jewel assets Baemin, talabat and Hungerstation, (4) shifts more profits to delivery, (5) shifts more profits overseas and (6) almost doubles Uber’s number of multi-product markets.

The recent food delivery acquisitions in Turkey seem smart (Uber paid ~$1.2bn combined for Trendyol GO and Getir’s food delivery business). I believe this gives Uber the leading delivery position in the market, and allows Uber to pair its leading mobility business in the country with a leading delivery business (a very important structural advantage, as discussed). Turkey was Uber’s 3rd largest untapped delivery market (out of the countries in which it operates a mobility business), behind Brazil and India.69 This seems like it will pretty clearly go well (so long as Turkey remains a hospitable country in which to do business). In fact, Dara commented at a Sep-25 Goldman conference that the Trendyol Go acquisition “so far has been an absolute terrific success.”

Acquiring the Foodpanda business in Taiwan would have given Uber effectively a monopoly position in food delivery in the country (the market was split ~50/50 between Uber and Foodpanda), so that logic is also easy to follow. Perhaps predictably it was blocked by regulators, however, and Uber had to pay out a $242m break up fee.

Uber spent over $7bn for an assortment of delivery businesses (Postmates, Cornershop, Drizly) in the 2019-2021 period that were likely overpays, but did help Uber jumpstart its delivery business (in the US, in grocery delivery, in alcohol delivery, and in Latam) at a critical juncture.

The $3.1bn Careem acquisition also may have been an overpay, but it cemented Uber as the clear leader in mobility in the Middle East. Uber, however, failed to translate this mobility dominance into delivery dominance. Delivery Hero’s talabat and HungerStation led in delivery in the Middle East by a significant margin prior to being acquired by Uber.

Uber’s most recent deals for Blacklane (a high-end chauffeur service) and Spot Hero (a parking app) seem like good strategic fits. Uber can likely leverage Spot Hero’s parking inventory as part of its self-driving offering (places to idle cars when not in use). Uber also plans to bundle discounts on Spot Hero inventory within the Uber One offering, providing an additional benefit to members.

The most questionable acquisition was likely Transplace. Uber spent $2.25bn for it on the hopes that, combined with its freight brokerage business, Transplace would position Uber to be a force in the freight space. This hasn’t come to pass, with Uber Freight’s growth stalling the last few years.

The divestitures mostly all seem defensible. This effort began before Dara arrived in Aug-17. Uber’s first divestment was its China business, which it sold to DiDi in Aug-16 in exchange for a ~18% stake in DiDi. It later sold its Russia business to Yandex in Jul-17 in exchange for a 37% stake in Yandex.Taxi, a JV (which was later bought out completely by Yandex). After Dara arrived, Uber divested its SE Asia business, Uber Eats India, Uber Elevate and ATG. These divestments led to stakes in Grab Holdings (28%), Zomato (10%; a food delivery operator in India), Joby Aviation and Aurora (26%; self-driving truck tech), respectively. I believe Uber’s stake in DiDi has lost ~67% of its value ($8bn implied value at the time of the divestment vs. a carrying value of $2.6bn today). Uber’s stakes in Aurora and Grab are valued around the same levels today that they were when they were acquired.

The most questionable of the divestments is likely that of Uber’s Indian food delivery business. Uber sold it for an implied value of just $200m, from what I can tell. Dara has subsequently expressed regret that Uber doesn’t have a delivery business in India to pair with its mobility business in the country. Uber does seem on the verge of announcing a closer partnership with one of the Indian delivery businesses (likely Zomato).

While many of Uber’s acquisitions have been tuck-ins and good strategic fits (where Uber is perhaps paying a market-ish multiple), Dara has also espoused the wisdom in the “overpay” for the truly great asset. Here is Dara from a Nov-25 HD in HD podcast, recounting how the best deals at IAC looked like “overpays” at the time, but made up for it (and then some) through many years of strong growth (e.g. Hotels.com, Match.com):

“I do think one lesson that I’ve learned is…the best deals that Barry and I embarked on were deals where we actually overpaid, but recognized that this was a great asset that would have compounding growth for years and years and years. So coming to growth assets with a strict valuation framework of like you know how much is this asset worth now, often misses the point which is it’s not about how much asset could be worth now or next year. How is it going to compound over the next 10 years?

You could argue we overpaid for Match.com…There was an auction for Hotels.com. We were the top bidder. We were convinced that the compounding of offline travel to online travel, it was just very, very early. We were convinced that it was going to continue for 20 years. And that allowed us to “overpay.”

So I think over a period of time as we move from traditional assets that have kind of more traditional metrics, growth metrics to internet assets that can compound for a really, really long time, we learned our lesson that it’s not about identifying what’s cheap. It’s about identifying growth opportunity and compounding on top of compounding. It’s magic when that happens.”

Overall, it seems likely that management will be at least reasonably intelligent when it comes to M&A.

High on Uber’s M&A wish list are likely:

• Instacart. As discussed above, the company is a nice strategic fit in US grocery delivery

• Zomato. Huge market opportunity in India delivery. Could pair with Uber’s mobility business, giving them a key strategic advantage in the country

• Rappi. To consolidate share in Mexico delivery

• Other rideshare — Bolt (Eastern/Central EU and Africa), Ola (India), Grab (SE Asia)

• Expedia. Has been rumored in the past. Dara was CEO at Expedia for over a decade and is still on the board. This would be monumental, and would cement Uber as the all-things travel app (would add hotels, vrbo’s and flights). Expedia currently has a market cap of ~$30bn vs. Uber at ~$150bn.

Management “nuggets”

• Level skipping70

• Bet on people71

Conclusion

It seems unlikely that, in the post AV world, these markets don’t converge back to 2-3 players.

The rideshare/delivery platforms will still benefit from virtuous cycles. The network with the most riders, in theory, will have the best utilization. If you have the best utilization, you can offer the cheapest fares, thereby generating even more riders. Additionally, the network with the most riders, in theory, will have the most vehicles. If you have the most vehicles, you can offer the best ETAs, thereby also generating even more riders.

These types of virtuous cycle industries tend to exhibit a winner-take-all dynamic. Thus, like in the pre AV world, I’d expect only ~2-3 “winners” per geo in the end.

It also seems unlikely that Uber isn’t one of these winners.

By the end of 2028, we could look up and Uber might have:

• ~$16bn of EBIT

• ~100mm Uber One members

• Line of sight to being the largest purveyor of autonomous rides in the world within the year, with a growing set of proprietary depots and charging infrastructure

• Sizable volumes of hotel bookings (maybe even a “flights” product), cementing Uber as the go-to, all-things travel app

Given all this, given Uber’s huge lead in rideshare/delivery, and given that most people won’t use more than 2 apps for either rideshare/delivery, are we really to believe that Uber is going to be somehow disrupted by “AV-natives?”

At this point, I don’t see it.

However, that’s not to say things couldn’t get choppy.

I think you have to believe Waymo is going to ramp the production of its vehicles by a lot. And you have to believe that Tesla will get to L4 (and subsequently also ramp production by a lot). Without even accounting for Zoox, this seems like a recipe for “price wars.” Waymo is backed by Google after all. And Tesla is foaming at the mouth to ramp Cybercab production and aggressively undercut Uber on pricing. Tesla has the manufacturing capacity to do so, and will likely be owned by SpaceX in the not-too-distant future.

While I think Uber has done a nice job playing its cards, AVs reopen the playing field. And unfortunately for Uber, some of the largest companies in the world are entering. Thus, even though I don’t believe Uber gets “disrupted,” I do think there’s a good chance Uber will need to defend market share through price concessions. Price concessions lead to margin pressure.

How bad of an impact this could have on Uber is still murky to me. I could see it perhaps being not so terrible because: (1) these AV natives still have a long way to go before actually impacting Uber’s financials, (2) a flooding of the market with AVs will actually lead to increased demand for rideshare (as is happening today in SF, etc.) which could offset pricing pressure, and (3) perhaps Uber’s not-so-demanding valuation today is doing the heavy lifting of pricing all of this in.

However, I could also easily envision pain as: (1) aggressive actors tend to ruin the party in these industries (commodity products; everyone knows the most market share wins and invests accordingly), (2) three of the largest, most aggressive companies in the world (Waymo, Tesla and Amazon) have entered the space (and aren’t exiting anytime soon), and (3) the number of AVs (and corresponding level of investment) needed to significantly impact the market is not outlandish.

In the end, though, if this all breaks right for Uber, we could be looking at the company as a dominant global transportation infrastructure business. It has the opportunity to be a “tollbooth” business, earning a cut of the transportation spend (mobility + delivery) of a large portion of the global population. If things break right, it will have huge scale, probably absurdly strong moats, likely attractive ROICs (even if it were forced to own many of its vehicles and infrastructure), and a very long runway for growth.

Pat Dorsey, in this talk at Google, explains:

PAT DORSEY: “So when Morningstar initiated on MasterCard, when it went public at $40, $50 a share, I remember our analyst wanted to put like a $100, $110 price target on it. And I was like, dude, you’re nuts, man. This thing is coming public at like $40. We’re not going to stick our necks out that far. Come on. How good a business is this?

And what he did is he just walked through fixed versus variable. That basically, okay — [there’s] very little variable costs in this business. You’ve got the fixed cost of a network, the data processing network. But then of the incremental revenue, tons will flow down to the bottom line. And this is where — and this is a geek thing — when you’re modeling out a high-growth company, don’t model in percentages. Model in dollars.

So it optically looks weird to say, margins are going to go from 13% to 25%. Ah! This is huge! My God, that’s never going to happen! So don’t do it in percentages. Don’t do operating margins doing from x to y. Think about fixed versus variable. How many dollars will they need to spend to get each dollar of additional revenue? And then see what operating margin falls out of that. And you may come to a very different conclusion.”

AUDIENCE: “Is that like operating leverage that you--”

PAT DORSEY: “Yeah, it’s operating leverage. And I have found that operating leverage is frequently one of the most mispriced things in the market because nobody wants to like hand their boss the portfolio manager or the director of research a model that shows something going from 13% to 30% margins. Because they’re going to get it back, and they’re going to say, you’re nuts, man.”

I’ve found this to be very true.

Yes, these high-growth/high-leverage models are more common today than twenty years ago. But still — banking on substantial margin inflection is scary because (1) it rarely happens (especially outside of tech), (2) even if it does happen, timing can be hard to predict and (3) you are sticking your neck out, as Dorsey says, by underwriting to it (much better to underwrite to strong-but-not-crazy operating leverage — you can still get credit for a nice call, but don’t have to look stupid if/when extreme operating leverage ultimately fails to materialize).

I’d say there’s a natural cap on believable margin expansion over a ~5 year window that lies somewhere in the neighborhood of ~1,000 bps (and this would be in an upside scenario, almost certainly not a base case).

So what actually happened at MasterCard? I was curious, so I went back and looked.

Here is the P&L you would have seen at the time of its IPO in 2006:

As we can see, MasterCard demonstrated healthy growth (a 15% 5-yr revenue CAGR) combined with a consistent margin profile of 12-14%.

Looking at this, a responsible analyst taking a cursory look might have baked in something like ~0-500 bps of margin improvement over the next ~five years (in a base case scenario). A better analyst, taking the time to really understand MasterCard’s unit economics/incremental margin profile, might’ve still only forecasted say ~500-1,000 bps of incremental improvement.

Here’s what actually happened:

Quite amazingly, MasterCard added almost zero additional expense from 2005 to 2010 ($2.5bn vs. $2.8bn). In the meantime, revenue grew by $2.6bn. As a result, Mastercard generated in operating income in 2010 ($2.8bn) almost as much as it generated in revenue in 2005 ($2.9bn).

While I’d bet many analysts at the time of MasterCard’s IPO might have correctly forecasted its 5-year forward revenue CAGR within a tight-ish band, there’s no way anybody would have done the same on margins. Even the Morningstar analyst that Dorsey thought was “nuts” was likely way short of the mark. The figures are hard for me to believe, even staring at them today.

But I think the point is, though, that no one is really to blame. Margin expansion like this rarely happens, so you shouldn’t expect it. Further, even if you did expect it, sticking your neck out too far, as Dorsey says, entails career/reputational risk that is scary and unnecessary.

So for the discerning analyst, situations may persist where high-leverage businesses are priced for strong-but-not-dramatic operating leverage, but where fat tail probabilities exist for the latter.

I had a chance to regroup on Veeva over the last few days after writing up the business back in June. Some updated thoughts:

Veeva is going to benefit from AI

“Duh,” you say. And yes, most people aren’t going to argue with this.

But at the same time, I’m not sure this is fully appreciated by the market. One could see how this might be the case: (1) AI revenues still haven’t hit yet (at all) for Veeva, (2) Veeva’s first agentic offerings will be rolling out in CRM, which is an area of more dubious AI utility, and (3) Veeva’s general ethos is under-promise/over-deliver and they don’t make a lot of grand AI statements (other than this1 one, which was pretty grand).

But over the last couple of quarters, some of Veeva’s opportunities in AI have come more into focus. This is because management has been somewhat spelling them out for us. And after a little digging, one can see that some of these opportunities seem pretty straightforward and meaningful. I also think they will hit a bit quicker than I had originally anticipated.

• Safety. Veeva sells a Safety suite that its customers use to monitor the safety of their drugs post clinical trial approval — a practice known as “pharmacovigilance” (PV). This process involves a great deal of “case processing” of adverse events — i.e. something happens to a patient taking a drug, and it must be documented (in a case record with a set of info) and potentially investigated. Case processing involves a lot of outsourced labor sorting through unstructured data (emails, EHRs, etc.) to put together case records for review and to form judgements around causality. The entire outsourced PV market is roughly $6bn today (and is projected to double to $12bn by 2030), with 60-70% of that spend going to “case processing.” Here is a breakdown of the outsourced PV spend:

  

  Case processing seems clearly ripe for AI (as do the other areas listed above, frankly). Here is Veeva management highlighting the opportunity last quarter.2 While Veeva has only 5 of the top 20 pharmas on its Safety suite today, it’s an area where Veeva is gaining ground quickly (and it’s an area where a customer’s entire safety operation will flip to Veeva overnight — it’s not a gradual migration like in EDC, say). If the AI revenue opportunity in PV amounts to just 10% of the ~$10bn outsourced PV market, that’s a ~$1bn opportunity. And it would be an offering that, in theory, customers should aggressively adopt (as it would offer the potential for huge cost savings). This use-case alone would be a significant needle mover for Veeva, which currently generates ~$3bn in total revenue.

• eTMF. Veeva sells an eTMF document management product (eTMF stands for electronic (Clinical) Trial Master File). The eTMF is a collection of documents (like trial protocols, consent forms, investigator brochures, safety reports, site-specific documents) that allow regulators to assess a clinical trial’s conduct and data quality. The process of building an eTMF requires a lot of manual processing as well like (1) collecting documents from various stakeholders, (2) manually reviewing each document to classify/index/assign metadata, (3) manually reviewing each document to check for completeness/quality/accuracy, and (4) properly filing/archiving/managing version control. This is an area that also seems quite ripe for AI. Here is management again spelling it out for us on recent calls.345 The global outsourced TMF labor market alone is estimated to be $1.5-2.5bn annually. Assuming just 10% of this market converts to AI revenue, this would represent another $150-250m of opportunity. And within eTMF, Veeva counts all 20 of the top 20 pharmas as customers.

While I don’t know if the AI use-cases above begin to have an impact in 2026 or 2028 or later, I’m almost 100% sure they will have an impact at some point and be meaningful.

And these are just two examples. Veeva’s aggregate opportunity in AI is obviously much larger — after all, the Veeva suite spans over 50 applications across the entire commercial and development functions of pharma companies (i.e. sales, marketing, customer support, clinical ops, clinical data, regulatory, safety, quality). Other potential AI applications that I put in my original write-up on Veeva (Safety Signal, Regulatory Content Writer, Clinical Trial “Co-Pilot,” CRM “Co-Pilot”) I would still expect Veeva to execute on, amongst many others. Just the fact that something called the CRO industry exists (which is a labor-intensive set of outsourced R&D services for pharma companies), and that it’s enormous (projected to be a ~$125bn industry by 2030) might be all that one needs to gain comfort that there is much to do.

As a side note, another perhaps underrated aspect of Veeva’s AI offerings is that they shouldn’t have to deal with much competition. Other potential competitors aren’t in the workflows, don’t have access to the data and would have to overcome extreme regulatory hurdles + ultra-high conservatism from buyers (around systems that deal with Safety, etc.) just to compete.

I’m not worried about the commercial business

Veeva traded down ~20% post Q3’FY26 earnings partially as a result of losing another top 20 pharma CRM customer to Salesforce (making it 4 losses total) as part of transitioning its CRM from Salesforce’s Force.com platform to it’s own “Vault” platform. This transition meant Veeva had to “re-win” all of its CRM customers and migrate them all onto this new platform — a process that started in 2022 and won’t be fully completed until 2030. By losing 4 top 20s, Veeva’s count of top 20s will drop from 18 to 14 in CRM.

It’s not ideal, but it’s also not detrimental by any means. CRM is just 20% of revenue at this point, and so losing a few large customers in that line isn’t going to make a huge dent (call it ~$60m of lost revenue on a base of $3bn, or ~2% of total revenue). And the losses will happen gradually over time.

Also, Veeva never fails to mention on calls that these lost customers can be (and that some probably will be) won back. The process of fully transitioning to Salesforce will be long/protracted (extending through 2028-2029 or longer) and will happen in select geos/brands at first before any kind of enterprise-wide roll-out. It will require a lot of custom development work (Salesforce doesn’t have a life science CRM product today) which is risky and oftentimes doesn’t go well. In the meantime, while Salesforce is focused on just building a CRM, Veeva’s entire commercial suite will advance, will develop seamless integrations with IQVIA (something it didn’t have before) and will offer an increasingly large set of new AI capabilities tailor-made for life sciences. So I’d say the likely case is that 1-2 of these lost customers actually come back. Further, below the top 20, Veeva’s win rate in CRM is much higher — these customers simply don’t have the risk appetite to take on custom-build projects.

For the customers that do transition to Vault CRM, Veeva can now cross-sell a new marketing offering, a new customer service offering and new AI offerings (all of which Veeva couldn’t offer while on the Force.com platform). Veeva can now also target MedTech customers in CRM, a significant opportunity (MedTech represents 1/4th of the life science industry). These forces will create a new tailwind for the commercial business going forward.

Lastly, management reaffirmed its $6bn revenue target by CY30 on its latest earnings call67, likely on account of (1) the strong continued momentum in Crossix, (2) the continued strength of Veeva’s R&D suite and (3) perhaps some comfort that AI will be a contributor to revenue in future years.

Other notes

• IQVIA partnership. IQVIA (the leading life sciences data provider) and Veeva settled their almost decade-long dispute in August. This seems like a positive development for both Veeva and IQVIA. I don’t think it limits too materially Veeva’s own data opportunity (e.g. Compass) — Veeva is still pursuing this just as aggressively. But it (1) helps cement Veeva’s position in commercial (it resolves a big issue with Veeva CRM which was that IQVIA data couldn’t be fed into it seamlessly), (2) allows for IQVIA’s CRO (the largest in the world) to use Veeva products (not an insignificant development), and (3) opens up some incremental revenue opportunities for Veeva (e.g. Veeva Nitro, which could be as big of an opportunity as Veeva’s other CRM add-on products, per management).

• Crossix. Management seems to have gotten even more bullish on Crossix in recent quarters. It seems the shift to digital marketing (through channels like Doximity, OpenEvidence) is now fully underway for the pharma industry, and Crossix is the platform of choice for spend measurement & optimization.

“Crossix is becoming more of a standard…that’s going to be a well-growing business for us. You should think of that as a well-growing business for the foreseeable future, 3, 4, 5 years type of thing.” — Q3’FY26 earnings

“It’s not unforeseen that Crossix [could] be as big as the whole CRM suite by 2030.” — Q3’FY26 earnings

• LIMS. Veeva won its first top 20 LIMS customer in Q3’FY26, which was a nice development.8 LIMS is a large category — I estimate in the neighborhood of $1bn.

The next couple of years should be exciting for Veeva. Over the course of 2026, Veeva will introduce AI capabilities inside of each of its applications in a phased rollout. By 2027, my hope is we’ll start to see some meaningful AI revenue contribution.

Further, I think there’s a reasonable chance that some of Veeva’s AI products will be powerful enough that they spur more rapid market share gains in Veeva’s “legacy” software offerings (while also contributing separately as consumption-based revenue).9 If you are a large pharma company and your rival is eliminating half of its outsourced PV expense through Veeva’s safety suite, how could you justify not adopting Veeva? For any of these AI solutions to work, in fact, a customer needs to be on modern cloud software — many of the legacy point solutions that still exist across R&D today (and that Veeva is displacing) are not that.

In all, Veeva’s downside continues to seem relatively limited while it’s upside potential, in my eyes, has grown. The recent weakness in the stock seems like a good opportunity to add, which I’ve been doing.

Important Disclaimer

This post is intended for informational purposes only and you, the reader, should not make any financial, investment, or trading decisions based upon the author’s commentary.

Although the information set forth above has been obtained or derived from sources believed to be reliable, the author does not make any representation or warranty, express or implied, as to the information’s accuracy or completeness, nor does the author recommend that the above information serve as the basis of any investment decision.

Before investing in a security, readers should carefully consider their financial positions and risk tolerances to determine if such a stock selection is appropriate. At any time, the author may trade in or out of any securities that are mentioned in the report as long or short positions in his portfolio without disclosing this information.

This report’s estimated fundamental value only represents a best efforts estimate of the potential fundamental valuation of a specific security, and is not expressed as, or implied as, assessments of the quality of a security or an actionable investment strategy for an investor.

Monday.com is many things to many people.

It’s a task/process management tool. It’s also a CRM. It has a Dev offering. It has a Service offering. It’s an offering for SMBs, mostly. But it’s also a tool for the Fortune 100 (its largest customer has 80k seats). It serves tech customers. But mostly it serves non-tech customers (70% of revenue is spread across 200+ non-tech business verticals). It has customers in 200+ countries, with about half its revenue coming from outside North America. Did I mention it has over 250k customers?

Being many things to many people, however, is not something I generally look for in an investment. I would say most investors tend to look for the opposite — a business with great strength in a niche.

Further, Monday’s primary offering — task/work management — isn’t considered particularly sticky, as far as software goes. It’s not vertical SaaS or an SoR provider in an area like ERP, CRM, etc. It’s not the worst thing to move from one project management tool like a Trello or an Asana to another. Morningstar went so far as to list Monday as a “no moat” company (in Sep-25 it was upgraded to “narrow” moat). Here’s another way to put it with a little more vitriol:

Nonetheless, I am intrigued by Monday.

I am intrigued because (1) it does all these many things in a surprisingly hyper-efficient manner and (2) Monday is still able to create real differentiation despite a lack of “nichey-ness.”

Hyper-Efficient

Monday is surprisingly efficient. It has best-in-class gross margins of ~90%. It has a very efficient R&D function at ~22% of revenue, despite a relatively small scale and an expansive product offering. It launched CRM, Dev, and Service — three major new categories — in the span of three years. It launched its initial CRM offering using a team of just four developers and its initial Dev offering with a team of just three. Oh and Monday’s on track to generate $1bn+ in free cash flow from 2023-2026 vs. an enterprise value of ~$6bn today.

Real Differentiation

Monday differentiates through the best-in-class flexibility of its solutions. Monday is able to offer this flexibility in each of its solutions, because each solution (Work Mgmt, CRM, Dev, Service) is built as a “thin layer” of code atop its core platform, and Monday’s core platform is highly flexible. The platform is built to act like a modular set of building blocks — easily customizable into whatever a customer needs.

A company that can be many things to many people — but that does it (1) very efficiently and (2) while creating real differentiation — is actually kind of great??

This is a company that’s done this:

And this:

(Overall NRR has increased meaningfully over time, if we exclude COVID-impacted 2021-2022)

And this:

And at its current price, Monday trades at a huge outlier multiple vs. its fundamentals (see below EV/ NTM gross profit vs. Rule of X across the software universe):

Yes, Yes — But What About AI?

I can understand that all of the above is meaningless if AI is going to significantly disrupt Monday’s business. The market is clearly particularly concerned about Monday, given where it trades (i.e. way out of line vs. fundamentals). Monday is not a SoR business nor a VMS juggernaut, qualities that would, in theory, give you a fighting chance in the AI transition.

I can understand the consternation from investors. Watch this Gemini at Work showcase from Google, and Monday feels behind the times. How far will these agentic capabilities from the likes of Google go and how quickly? Can Monday really be expected to compete with Google and Microsoft, who are encroaching on Monday’s turf? For Monday, AI introduces significant uncertainty and even existential risk. So I think it’s reasonable that Monday, despite being a very healthy business today, has been relegated to “prove it” status.

However, fears of Monday’s demise are being greatly exaggerated in my opinion. In fact, I’d argue Monday is set up better than most to compete in the AI era.

Some of Monday’s AI fit + capabilities:

• Suite of solutions. Having a full set of offerings (WM, CRM, Service, Dev) is potentially quite powerful in the AI era. It should allow agents to more seamlessly perform actions across business functions and keeps customer data together in one place. Further, for many years, I believe agents will do best sitting alongside more traditional software workflows (in many/most areas), and Monday has this “scaffolding”

• Healthy initial set of AI capabilities. Monday already has the bones of a fully blown AI orchestration platform. While certain elements are missing that you see from leading competitors, Monday seems to be catching up

• Good product fit for AI era. Monday is the leader in Work Management, which is a natural fit for an orchestration layer. If you want a “closed/won” deal in your CRM to kick off a bunch of other processes automatically (including in other applications), Monday is one of the best tools out there today. It is digital “connective tissue” spanning an organization — tailor made for agentic applications

• Good platform fit for AI era. Monday’s platform has best-in-class flexibility and shines at process management and automation. Further, it’s “schema-less” database architecture should help Monday excel as a no-code app platform

• Large user base. Monday has over 250K customers. This should drive more robust feedback loops and ultimately better agentic offerings, all things equal. The products that win in AI will likely be those getting lots of usage/feedback/refinement (see Cursor, OpenEvidence). A large customer base should also drive a more dynamic marketplace of third-party agents (which Monday is introducing), which would extend Monday’s agentic capabilities

• Great track record of product development. Monday seems to know how to design great products. With WM they threaded the needle between user-friendliness, high flexibility, and powerful capabilities (and emerged as the top-rated WM offering). Their CRM is highly-rated despite launching just recently and with relatively bare-bones capabilities. Like the rest of their products, Monday’s AI solutions will be designed for the non-technical user (which seems like a fruitful niche and where they could be a leading offering)

I’m by no means crowning Monday an AI winner. But I also think Monday’s chance of success with AI is much higher than zero. In H1’26, Monday will be broadly releasing it’s agent platform, and I think at that point Monday’s AI vision will come a lot more into focus.

All-in-all Monday is at a crossroads of sorts.

Pre AI, Monday seemed to be on a reasonable trajectory towards building something like a ServiceNow-lite for SMBs + mid-market, with a high-quality platform and a growing suite of offerings.1 As the AI era dawned, it put Monday in an all-out race to hit enough critical mass across its product + AI offerings so that it can withstand (and ultimately benefit from) the coming agentic wave.

Below I’ll more closely examine the case for Monday.

Company Overview

Monday.com was founded in 2012 by Roy Mann and Eran Zinman as a provider of project/work management software.

From the beginning, the founders sought to create a highly flexible software platform. Their thesis was that the market didn’t need another, “better” workflow for project/work management. Rather, the founders believed that there was no one best workflow — that the best workflow for a business depended on the unique needs of that business. Thus, their solution was to empower customers with the ability to easily build/tailor their own solutions.2

Further, the founders felt a gap existed in the market between (1) SMB software, which, though attractively priced, tended to be “rigid” and (2) enterprise software, which, while much more customizable, tended to be costly & painful to set up and maintain.

Monday gained initial traction with its “boards” offering in 2014. Once launched, the company leaned heavily into a product-led growth strategy focused on performance marketing and a self-serve GTM motion. While heavy performance marketing (on platforms such as Facebook and YouTube) was looked upon suspiciously by the VC/investor community, by meticulously tracking the impact of this spend, Monday was able to optimize it and grow rather efficiently. Monday also shrewdly optimized their spend around cash on cash return. As a result, Monday was able to quickly recycle cash from customers back into more marketing, which greatly accelerated growth relative to dollars raised.

From inception through their IPO in Jun-21, Monday burned ~$121m of capital to finance their operations. Monday has been cash flow positive every year since the IPO. Monday achieved this while emerging from a space that was already mature and highly competitive (project management).

Starting in 2022, the company launched three new major product categories within the span of just 3 years — CRM (2022), Dev (2022) and Service (2024). CRM, its most mature new offering, hit a $100m revenue run rate in Q2’25, just ~3 years post launch (best-in-class SaaS start-ups would do this in more like ~5-8 years). And Monday’s Service offering is actually growing faster than CRM at the same point of development.

More recently, Monday has become significantly more aggressive targeting the enterprise segment — adding enterprise level functionality, premium support offerings and a more traditional enterprise sales motion (to pair with its PLG motion). These efforts have generally been quite successful — Monday now counts ~4k customers that spend over $50K annually, up from ~500 at the time of its IPO.

Monday today has a revenue run rate of ~$1.3bn. It’s grown by ~4x since its IPO in mid 2021, representing a CAGR of ~40%.

Product Overview

Monday was born out of the idea that prevailing SMB software was frustratingly rigid. The founders wanted empower non-technical users with the ability to easily build custom software solutions.

Thus, from the company’s inception, the Monday platform (called WorkOS) was intended to act like a set of “lego blocks,” or underlying capabilities, that users could leverage to assemble robust “product solutions:”

What are the WorkOS “Building Blocks?”

• Boards. It starts with Boards, which is Monday’s core user interface. They have rows and columns like an Excel sheet. Monday calls them a combination of “a sophisticated spreadsheet and a no-code database.” It’s where you hold and process information. Users can start Boards from scratch or from using one of 150+ Board templates built for various use cases.

  • Columns can include things like numbers, text, dates, locations, formulas, links to other Boards, attachments/files, tagged users, etc.

  • Rows, or “items,” can represent virtually anything a user chooses, including an entity, workstream or campaign. Users populate items with actions, steps, leads, contacts, or other elements of a workstream. Within an item, users can collaborate with each other, write text, attach files, tag other users, etc. Users can communicate within the updates section of the item, keeping all communication within a specific context.

• Views. Monday offers a variety of views — table, kanban, gantt, calendar, etc. These allow a Board’s contents to be visualized and interacted with in different ways

• Forms. Users can capture additional data through web-based forms, including from non-Monday users. Each response can create a new item within a Board, with all of the data supplied by the recipient automatically populated in the relevant columns. Forms can be adapted to suit a variety of use cases, including ticketing and order management.

• Automations. Monday’s platform can be used to easily build automations for complex workflows. Automations are triggered based on actions that occur or thresholds that are reached and can include notifications, item creation, assignment of ownership and due date, moving an item to another Board, etc.

• Integrations. Integrations allow users to connect with external tools to share data, automate actions, etc. For instance, users could set an automation that when a charge event occurs in Stripe, an item is automatically created in a specified Monday Board and a user is notified via Slack. Monday offers integrations with Gmail, Outlook, Jira, Salesforce, Google Drive, Dropbox, Stripe, Slack, and Zendesk amongst many others.

• Dashboards. Users can create dashboards to summarize/track/report on progress across any/all of their workstreams.

• Workspaces. Users can organize and centralize Boards and dashboards using workspaces, which are defined by any category the user chooses. In large organizations, workspaces can represent a department or a team working on a project, for example. Users with workspace-level permissions can also restrict access and protect private information.

Here is a short overview demonstrating the capabilities of WorkOS.

Atop these building blocks sits Monday’s product suite — Work Management, Dev, Sales and Service:

Work Management

Work Management is a platform that enables teams and organizations to plan, execute, and monitor all types of work, projects, and processes.

Work Management was introduced in 2014, and is Monday’s original and most mature offering, responsible for ~90% of Monday’s total revenue today. It can be used by 1-2 person teams all the way up to the Fortune 100 (its largest customer has ~80k seats). Currently ~42% of ARR is SMB (<100 employees), ~34% is Mid-Market (100-1,500 employees) and ~24% is Enterprise (1,500+ employees).

It’s priced in the $9-19+ range, per user per month. The $12 “standard” option is highlighted as the most popular, which includes guest access, timeline/Gantt/calendar views, automations (250 actions per mo), integrations (250 actions per mo), and dashboards that can combine up to 5 boards.

Below is pricing for Monday’s primary competitors in WM. It’s not an apples-to-apples comparison because the functionality and tiers vary by competitor. But based on my read of functionality/tiers, I’d say Monday is priced at a slight premium to Asana and Trello, in line with ClickUp, at a slight discount to Notion (hardest to compare apple-to-apples), and at a meaningful discount to Airtable and Smartsheet.

The product is generally very highly rated.

Gartner has Monday as the top-rated solution in both the “Adaptive Project Management” and “Collaborative Work Management” Magic Quadrant categories:

It also scores highly in the “Marketing Work Management” category:

It’s worth noting that Monday has been working its way “up and to the right” in Gartner’s Magic Quadrants over time. Here is the Adaptive Project Management quadrant from 2022-2025. Monday just recently took over the top position in 2024.

G2 has Monday as the highest rated solution (by satisfaction) in their Project Management category, slightly ahead of the likes of Smartsheet, Trello, Asana, Notion, etc.:

While Monday’s offering often gets lumped in with primarily project management tools like Asana and Trello, it is built to be more work/process management vs. project management (though it works just fine for project management).3

Below is a list of pros/cons for Monday WM based on reviews from around the web:

Strengths

• Visually Appealing. Clean & colorful design, intuitive, variety of views

• Customizable Workflows & Templates. It’s a very flexible platform — just about everything is “drag-and-drop/editable.” Users can select from over 200 templates for various industries, functions (e.g. sales, product, marketing, HR, finance, etc.)4

• Automation & Integrations. Automations are easy to generate/“no-code” using natural language. Deep integrations with tools like Slack, Google Drive, Microsoft Teams, Zoom, GitHub, Zapier, etc.

• Collaboration. Great for keeping everyone on the same page. Real-time comments, file sharing, mentions, activity logs, alerts, and guest access

• Scalability. Suitable for small teams up to enterprise-level organizations, with rapid pace of updates and new features

Weaknesses

• Pricing Model. Higher priced vs. certain other project management tools (Asana, Trello), must buy in seat package increments (i.e. 3 seats, 5, seats, 10 seats, 15 seats) that can lead to paying for more seats than is needed

• Learning Curve & Setup Complexity. While overall intuitive, mastering custom automations, boards, dashboards, and automations can be “overwhelming,” especially for newcomers

• Performance of Large Boards. Handling very large or data-heavy boards can introduce lags or slowdowns (Monday has been very focused on fixing this — more on that later)

In terms of actual use-cases for the WM suite, Monday hasn’t disclosed this. However, at their 2025 investor day, they did disclose the breakdown of use-cases for Monday Magic, one of its AI solutions. And I imagine this is reflective of the overall use-cases for the Monday WM platform:

As we can see, Monday is used for a range of functions that touch all aspects of a business. While most popular are project management (16%), marketing campaign management (13%) and creative project management (10%), you also have areas like event management (5%), employee management (onboarding, HR, development; 5%), sales management (3%), finance management (3%).

Overall, Monday has said that ~40% of its Work Management users are from sales, developer or IT teams:

According to a recent Business Breakdowns podcast with Ben Hensman (a PM from Square Peg), 12% of the customer base uses Monday for HR, 47% use it to manage clients, 14% use it for finance management and 21% use it as a ticketing system (I’m not sure where he gets these numbers).

Sales CRM

Monday Sales CRM was released in 2022 and thus represents Monday’s second most mature product. As of Q2’25, CRM was generating $100m of ARR, just ~3 years since its launch (and up from $25m of ARR as of Q4’23).

Monday CRM is suitable for just SMB to mid-market customers at this point. It’s not trying to compete against the likes of Salesforce — it lacks key functionality that would be required to target enterprise-class customers.

With ~33.5k customers and $100m of ARR as of Q2’25, Monday CRM is generating ~$3.0k of annual revenue per customer (which implies ~10 seats per customer). This is up from ~$1.9k per customer as of Q4’23 (which implies ~6 seats per customer).

Currently 74% of ARR comes from the SMB segment (<100 employees), 20% from Mid-Market (100-1,500 employees) and 7% from Enterprise (1,500+ employees). CRM currently skews the most towards the SMB segment out of all of Monday’s offerings, including WM (which is ~42% SMB).

Below is a look at quarterly growth in Monday’s CRM customers. In Q2’25, customer accounts grew 61% YoY.

(Monday stopped reporting this figure in Q3’25)

Monday CRM is priced in the $12-28+ range, per user per month.

Below is pricing for Monday’s primary competitors in CRM. Dynamics & Salesforce lead the way on pricing, followed by HubSpot, then Zoho/Freshworks. I’d say Monday is the cheapest option, or close to it, which is appropriate as a newer, less feature-rich offering.

While Monday CRM made it into Gartner’s Magic Quadrant for CRM for the first time in 2025, as you can see below, Monday was Gartner’s lowest rated CRM.

G2 rates Monday CRM much more strongly, with slightly higher customer satisfaction than all of the other most prominent CRM solutions (Salesforce, HubSpot, MSFT Dynamics, etc.).

Monday would tell you that they win in CRM the same way they win in other categories — by offering best-in-class flexibility and operating in a middle ground between customizable, but expensive options like Salesforce/Dynamics and lower-priced but rigid options like Zoho/Freshworks.

From the Dec-24 NASDAQ conference:

(Note that Monday management is Israeli, and they speak in a somewhat broken English, hence some of the poor grammar in the quotes that follow)

“So a good example would be that -- we started CRM and within 6 months, we went from 0 to being the top 4 in G2 Crowd in the review site, and there are hundreds of CRMs there.

And the reason because of the power of the platform that like -- and in CRM specifically, like we see a huge gap, like if you either have like rigid small CRMs and the large ones like Salesforce, where you can customize them, but it costs a lot and there is an implementation cost as well. And we’re kind of like solving that problem in the middle. It’s still very easy, very simple, but you can do anything you want. We brought something new to the -- like a trade-off. There is no trade-off.”

More quotes on why Monday wins.567

Here is a clip of a Monday CRM partner — CRM Crew — a former partner of Salesforce, Dynamics, Hubspot and Zoho, commenting on how much easier Monday is to manipulate than these other platforms.

Here are the Pros/Cons taken from the Gartner report:

Strengths:

• Highly customizable no-code interface: monday CRM’s intuitive approach and drag-and-drop functionality allow teams to quickly create and adapt workflows without technical expertise, thereby improving familiarity with the solution. Its core user interface resembles adynamic spreadsheet with columns and row equivalents, also known as boards within monday.com’s features.

• Ease of management: monday.com differentiates itself with competitive license fees, lean implementation and maintenance costs, and a short deployment time, all contributing to a lower total cost of ownership (TCO). The platform’s no-code automation and intuitive UI contribute to rapid deployment and high adoption, driving faster return on investment compared with more complex SFA solutions offered by other vendors evaluated in this Magic Quadrant.

• Collaboration and sharing capabilities: The platform excels in collaboration features, including mutual close plans, document co-editing and granular sharing permissions (down to the cell level). These capabilities support both internal team collaboration and external partner engagement, making it suitable for partner relationship management (PRM) use cases.

Cautions:

• Limited AI and ML depth: Despite the presence of generative AI, monday.com lacks machine learning capabilities across critical areas such as predictive lead scoring, opportunity health scoring and forecasting. All AI-requested demonstrations to Gartner were rule-based and formula-based or relied on simple prompt-driven automation using generative AI, limiting their strategic value.

• Manual data entry: The platform heavily relies on manual data entry and spreadsheet-like boards. This design can become unwieldy and inefficient, especially for larger organizations managing complex sales processes. The need to predefine numerous columns and formulas adds to the setup burden.

• Weak forecasting and analytics capabilities: Forecasting features are minimal to nonexistent. monday.com does not support predictive forecasting, what-if modeling and advanced analytics. Reporting is limited to simple dashboards with limited visualization options, which may not meet the needs of data-driven sales organizations.

Campaigns Module

In Q4’25, Monday released its “Campaigns” module for CRM. Monday Campaigns is an “AI-powered” product that helps marketers create, launch, and optimize campaigns. This is a nice walkthrough of its capabilities.

Key features include:

• AI-powered creation. Generate copy, explore audience segments from CRM data, and keep campaigns “on-brand”

• Drag-and-drop email builder. Build professional campaigns with no coding required

• Intelligent automation. Trigger personalized campaigns from CRM updates and customer actions

• Optimal timing. Send at the best times for maximum impact

• Segmented audiences. Target the right customers with precision list management

Campaigns is designed to integrate deeply with Monday:

• With Monday CRM. Campaign performance and lead conversion tracking live in one place, giving teams a complete view of the customer journey, from lead to customer

• With Monday Work Management. Campaign planning, briefs, approvals, and creative execution connect seamlessly to day-to-day workflows, ensuring marketing aligns with wider business goals

One of the largest contingents of Work Management users are marketing personnel. So this is a solution that is both a natural/necessary extension to CRM, but also a great cross sell fit on the Work Management side as well.

Monday introduced this product in Sep-25, and already had 200+ accounts as of early Nov-25.

Dev

Monday Dev launched in Q4’22. It has not enjoyed the same level of traction as CRM. As of Q2’25, it’s grown ARR to $14mm and its number of accounts to 4.7k.

With 4.7k customers and $14m of ARR, this implies Monday Dev is generating ~$3.0k of annual revenue per customer (implying ~20 seats per customer). Currently 49% of ARR comes from the SMB segment (<100 employees), 31% from Mid-Market (100-1,500 employees) and 20% from Enterprise (1,500+ employees).

Below is a look at quarterly growth in Monday’s Dev customers. In Q2’25, customer accounts grew 72% YoY.

(Monday stopped reporting this figure in Q3’25)

Monday Dev is priced in the $9-20+ range, per user per month.

Below is pricing for Monday’s primary competitors in Dev. Jira is the dominant player in the space, and Monday is actually priced at a bit of a premium to Jira which of course seems backwards. But this is something Monday felt was necessary.8

Here is the DevOps Magic Quadrant for 2025 (Monday is not included):

And the G2 reviews, where Monday scores well.

In reading expert call transcripts with Monday partners, a theme regarding Dev is just how hard seems to displace Jira.

From Monday partner Omnitas Consulting (Tegus, Jul-25):

“It’s not performing, and it’s not because of the product. It’s because of the culture and the stickiness of the competitors. It’s so difficult to push Atlassian out of a company to replace it or Azure DevOps. We’re not even bothering anymore. We’re not marketing it.”

From Monday partner Elevate (Tegus, Aug-25):

“Dev was not as successful as a product. Tough to go against someone like Jira, who has just such a cultural grasp on that demographic.”

It seems Monday Dev is best positioned for customers focused on (1) user friendliness and/or (2) collaboration between all stakeholders in the development process (beyond just developers).9

Service

Service is Monday’s latest product, released in Q2’24. The product was doing ~$7mm of ARR as of Q2’25.

Service can be used across all company functions including IT, HR, finance, legal, etc. — any area where requests/tickets are created and need a response. Using Service as a customer facing solution is not currently a core use-case for the offering (though it is a possible use-case), from what I can tell.

Service is currently Monday’s most upmarket-oriented product, with 16% of revenue coming from enterprise (1,500+ employees), 52% coming from mid-market (100-1,500 employees) and just 32% coming from SMBs (<100 employees).

This is also Monday’s highest ACV product, with an average ACV of ~$7k as of Q2’25 ($7mm ARR / 1.0k accounts)

While generating relatively little ARR today, it is actually Monday’s fastest growing product to date — outperforming both WM and CRM (slightly) through its first ~15 months post launch.

Here’s a look at customer growth since launch.

(Monday stopped reporting this figure in Q3’25)

Service is priced in the $26-38+ range. It is Monday’s highest ASP product.

Below is pricing for Monday’s primary competitors in Service. While ServiceNow doesn’t list pricing on its website, it’s ITSM product reportedly starts in the $60-90 range, per user per mo. Zendesk and SolarWinds occupy the higher end alongside ServiceNow. Freshworks is a tier below, followed by Jira, Zoho and Monday in the lowest tier.

As Monday enhances the product, there should be room to add a tier above Pro, which as we can see from Monday’s peers, is generally quite lucrative (for Freshworks and Jira the highest tier is ~2x the price of the next highest). Currently Monday is lacking in functionality relative to a Zendesk or a Freshworks.10

Service is perhaps the most natural product fit for Monday’s platform (after Work Management). Service requires flexible and robust process management combined with strong automation capabilities. Monday management from Q2’24 earnings on the early success of Service:

“Look, it’s very early days -- how this will compare to Monday CRM or Monday Dev. What we can share is that the feedback is very positive from customers. A lot of excitement. One of the things that really resonates compared to other players in the market, again, is the flexibility, kind of similar to CRM and Dev products. But I think specifically for Service, it kind of -- the effect of this is even more significant to our customers because they want to automate a lot of the processes within the IT service management product, and they want to have more customizability on how they catch data and how they process workflows. So this is exactly where Monday shines.”

Not only is the Monday platform well positioned to execute on a Service product, but Service is also a natural fit to sit alongside Work Management for many of Monday’s current customers. Monday management from the Dec-24 NASDAQ conference:

“So the reason we come out with new products is we really believe in them. We see the demand. We see customers want it, but it’s actually every one of our products have a different go-to-market…And what we’re seeing with Service, which shows like great and promising early results, is that we’re able to go to market to existing customers because it’s -- service essentially it’s ESM, enterprise service management. It’s like everywhere, you need tickets, you need requests. So it might be legal, it might be finance, it might be IT as well. So wherever we land within work management, the same person that we talk with can also take service. So we’re really optimistic about it and show great early results, but we’ll have to see.”

AI Capabilities

Monday has been rapidly introducing AI capabilities.

While Monday is seeing adoption — primarily of its AI Blocks product and Monday Vibe — I wouldn’t say its AI offerings have hit some sort of “escape velocity.” Monday’s agent platform will be a very important capability (coming H1’26), and it would be a great sign for this platform to show meaningful traction/engagement.

Below is an overview of Monday’s AI offerings today:

• AI Blocks (released Sep-24). Modular, customizable AI actions. These capabilities include actions like categorizing items, extracting info, summarizing text, translating text, etc. that you might embed in a Board

• Monday Sidekick (released Jul-25, now generally available). A “digital worker,” aware of a user’s company, role, and responsibilities. Proactively suggests and takes action to move work forward. Like a ChatGPT overlay within Monday

• Monday Magic / Monday Vibe

  • Monday Magic (released Jun-25, now generally available). Like vibe coding within Monday’s core offerings, users can create and modify workflows or boards with prompts in plain language

  • Monday Vibe (released Aug-25, now generally available). A broader vibe coding platform that enables users to build secure, custom business apps natively connected to all their Monday data

• Monday Agents (scheduled rollout H1’26). Specialized digital workers that can be built from scratch or by building off a Monday “template” using natural language

• Product Power-Ups (scheduled rollout H1’26). Agentic capabilities embedded throughout Monday’s suite of products (i.e. an SDR for Monday CRM, ticket resolution for Monday Service, etc.)

AI Blocks

AI Blocks was Monday’s first AI product, released in Sep-24. AI Blocks allow users to embed AI functionality within Boards (mostly as AI-powered columns).

For example, AI Blocks would allow users to translate, summarize, assign a label, extract information from a file, improve text, generate a doc, assign people to an item, detect sentiment, prioritize, etc.

Here’s an example from Monday where an HR department loads a bunch of resumes into a Board. The team can then perform a number of different functions using AI Blocks:

• Pull out a candidate’s years of experience in a certain industry or overall

• Summarize a candidate’s work experience

• Dynamically assign interviewers based on certain criteria

• Provide an overall score to a candidate based on a natural language prompt like “a strong hire would be someone with 3+ years of experience, held a leadership position in the past, and prioritizes effective communication with their team. An average hire has at least one of those qualities”

The use of AI Blocks has been, to put it in management’s words, “exploding” across the platform. Here’s a chart of “Total AI Actions” from their Q2’25 shareholder letter:

Monday Sidekick

Monday Sidekick was Monday’s first foray into the world of digital “agents.” Monday describes Sidekick as a personalized, context-aware digital worker.

You can ask it to do things both inside Monday as well as outside, like researching a venue for a conference in NYC and drafting emails to check availability.

Here is an overview.

From expert transcripts, my sense is that Sidekick is not that great of a product today.

Monday Magic

Monday Magic is like “vibe-coding” but contained within Monday’s core offerings. Monday Magic allows users to build and modify work solutions in Monday (e.g. boards/dashboards/forms/etc.) by just describing what they want using natural language.

Here’s an example of building & iterating on an event tracking application with Monday Magic.

From it’s launch in Jun-25 through mid Sep-25, users built 2,000+ solutions with Monday Magic. While that’s not exactly explosive growth, Magic seems like a useful capability to add to the platform, and I’d expect it’s usage to grow.

Monday Vibe

Monday Vibe is a fully blown “vibe-coding” offering from Monday. Except with Monday Vibe, vibe-coded apps will sit atop MondayDB and come natively connected to all of a customer’s data, workflows and integrations within Monday. Apps will also be backed by Monday’s governance, security and compliance.

Here’s an overview.

Monday Vibe has been used to build 60k+ apps between its limited launch in late August and early November.

Below are some examples of apps that you can create with Monday Vibe: a customer segmentation tool, executive KPI app, event page, supply chain tracker, time tracking, OKR dashboard, sales commission tracker, social media content calendar, org chart, a campaign health tracker.

Monday Agents

At its Sep-25 investor day, Monday unveiled its Agents offering, which will be generally available in H1’26.

The offering will consist of (1) ready-made agents built by Monday as well as (2) a platform for building custom agents, where bespoke agents can be designed by customers using natural language. Bespoke agents can be designed from scratch, or built off of existing templates.

Here is an example of a bespoke agent created to collect feedback from customers. As you can see, the agent can be built using a user-friendly, no-code interface (similar to the style of Monday’s automation platform).

Product Power-Ups

One such “ready-made” agent Monday highlighted at their Sep-25 investor day was Chris, their SDR agent. This agent is designed and built by Monday, and will come embedded within Monday CRM. Chris has the ability to contact new leads by phone, qualify leads and set up calls with sales reps.

Monday’s Service offering will also feature embedded AI capabilities. AI is critical in service where tickets can be auto-assigned to reps, auto-categorized, auto-scored for sentiment, or ultimately just answered/completed by AI.

In Service, as we can see from the slide above, Monday will offer the ability to more granularly fine-tune AI’s level of autonomy.

Monday’s AI Agent Frameworks & Vision

In this talk, the head of AI at Monday, Assaf Elovic, outlined some key frameworks that Monday has used in the design of its agents:

• User control. It’s important to give users control of an agent’s “level of autonomy.” Do you want an agent to purely suggest actions? Do you want it to operate independently with no human in the loop? Do you want it to preview actions, and require human approval before taking them?

• Entry point. Monday is able to weave AI into its offerings in creative ways that help drive adoption. While they have a chatbot-like window, they also give users the ability to say, assign a task to an AI agent within a board just like how you might assign a task to a human. In that case, using an agent is a completely seamless experience within an existing workflow. Monday has also embedded Sidekick into Boards in a way where users can “just click the star” (which is the icon for Sidekick) which will immediately engage Sidekick on an item.

• Preview. An engineer wouldn’t push code generated by Cursor straight into production without looking at it. In the same way, a user on Monday shouldn’t be expected post content to a Board without also looking at it. So this requires a preview function, where a user can see what actions an agent will take before “approving” those actions. After this preview function was introduced, Monday saw a dramatic increase in AI usage.

• Explain-ability. Give users an understanding of why outputs from an agent are happening. This way, users are given an ability to change the outcomes.

Agent Marketplace

Monday will be opening up its marketplace to agents built by third parties. If a robust marketplace develops, this could nicely extend Monday’s agentic capabilities. Imagine a third-party legal agent that could draft or review contracts. Or an agent that might be really good at enriching leads. Or an agent that could verify someone’s work history (by actually calling all past employers). Or even hyper vertical-specific agents like handling collections for dental practices or scheduling viewings for property managers.

What AI Capabilities are Missing?

• Agent-to-Agent. A missing piece for Monday is the ability for a Sidekick to connect to other apps natively (i.e. from within the Monday application). This would enable things like summarizing a document in Box, searching across Slack, prompting Canva to draft a set of designs, asking your inventory management system to order more of some item, etc. — all without leaving Monday.

• AI-Infused Workflow Builder. Monday already has a workflow builder product. The logical next step would be to infuse this product with AI in useful ways. I think over the next couple years, we’ll start to see this. But as Assaf shared in this Oct-25 post following the OpenAI dev day, even OpenAI isn’t doing this yet:

  

• AI Notetaker. Monday introduced an AI notetaker for all Enterprise & Pro accounts a few weeks ago. This is a helpful feature for users and also creates critical “unstructured” call transcript data that can be leveraged by AI

How is Monday Charging for AI?

Monday currently charges on a consumption-basis for most of its AI products, as we can see below.

Right now Monday provides a one-time set of AI “trial” credits for all accounts. Once credits are used up, accounts will need to buy additional credits. Certain minor AI features are free to use (e.g. summarize updates with AI, filter with AI, build Workforms with AI, etc.).

For Vibe, Monday charges $100/mo for up to 10 active apps and $250/mo for up to 25 active apps (I assume this is per account). Users can “draft” an unlimited amount of apps for free. This seems priced appropriately cheaply to drive adoption (and enhance the attractiveness of the overall Monday solution).

Go-To-Market

From the inception of the company through 2018, Monday had essentially no sales people. The GTM motion was pure performance marketing (which it leaned into heavily) + self-serve.

Also from practically inception (when Monday was <10 people), Monday decided to build a custom BI engine, called Big Brain. Big Brain’s original focus was performance marketing, and it provided Monday with in-depth data on the success of every campaign it would run. Monday credits Big Brain with much of its success in performance marketing. From the Logan Bartlett Show (Dec-24):

“Our initial intuition [was] let's use something off the shelf. An analytics software off the shelf. But, I don't know, since day one we had this huge inspiration that we want to build a big company. And we said that if we want to be successful, we have to control our own destiny. We have to control information and we cannot fly blind…

Because back in the days, unlike what happened in the later years, we didn't have a lot of resources to raise money from. Our Series A was actually $1.4 million…We have to be very efficient in how we spend money, so since day one we tracked everything, literally everything. Every campaign click, every visit to our website, every banner view, every YouTube ad being watched.

And then, because we built it on our own platform, we could track a user all the way throughout the sign-up funnel — converting to payment, eventually expansion, and so on. So you could see a complete user journey — years back — with all the data, all the touch points, everything that converted that user into a paid account. And eventually how they expand over time... And then on top of that, what we built is a very efficient performance marketing engine that can crunch all that data and let us know, in a very specific way, which campaign that we’re running is very effective and which ones we're losing money on.”

Another unique thing that Monday did in its early years was spend heavily on certain non-Google ad platforms, in particular Facebook and YouTube. While this was a relatively unusual tactic, Monday knew that it was making a healthy return on this spend, which Big Brain could prove definitively. This is partly what Monday credits for the fact that its customer base became mostly non-tech companies — not something that management had originally anticipated. To this day, Monday’s customer base is ~70% non-tech (as well as ~50% non-North American).

Another unique decision Monday made early on was to optimize its marketing spend for cash-on-cash payback (i.e. minimize the time it took to recover their marketing spend, in cash, from customers). From the Logan Bartlett Show (Dec-24):

“Almost nobody cares about cash…Nobody talks about what actually goes in and goes out out of the company bank account. But at the end of the day, I always say to people you cannot go to the supermarket and pay with LTV. People need cash in order to operate their business. So we always try to optimize for that. How much did we spend — actual spend going out. When do we get the cash back? Do people pay us annually or monthly? Do they upgrade? And so on. I think that eventually made Monday a very efficient cash machine over time…

I think a lot of people, when they think about, ‘We want to be more aggressive.’ We’ll just raise another round, and we can spend more in acquiring customers. Let's say we got a great LTV of, I don't know, $100,000. So we could spend more to acquire those customers. But that's not the point, because when you recycle cash quickly you can reinvest that into the business. I always say that our best investor [has been] our customers because of the fact that they pay for the tool up front…allowed us to get the cash back to reinvest that into marketing, essentially. Let's say out of that $1.4 [million Monday raised] in our seed [round], we spent, I don't remember the exact number, but about 5 million in performance marketing using that 1.4.”

This also meant tweaking the prospect journey in ways to optimize for receiving as much cash upfront as possible. They reportedly spent to an 8 month payback initially, “and not a day longer.”11

Today, Monday markets through over 55 different customer acquisition channels, with roughly 10% of new ARR coming through Google AdWords.

In 2018 the company decided that a sales org was necessary after all. If they wanted to go up market, they found, these types of customers wanted to speak with somebody. So Monday began building a sales team that ended 2022 with 774 employees (across sales, partners and customer success), up from 166 employees at the end of 2019. Since 2022, the overall employee base has doubled, so the GTM team has very likely more than doubled to 1,600+.

So since the end of 2019, over 5.5 years, this team has 10x’d. Somewhere likely a little over 1/3rd of the team was hired in just the last 2 years, and would therefore not yet be fully ramped contributors.

The GTM motion today is a combination of “top-down” and “bottoms-up.” Monday is unique in that their “top-down” motion so far has only been applied to existing customers, who have already adopted and are actively using the product.12 From Q1’22 earnings:

“Usually the way customers adopt Monday is “bottom-up.” And we see customers starting using Monday from all companies, all sizes and different roles, different positions within the company.

The way our sales algorithm work is that, whenever a customer signs up, we ask that customers advances detail during the final process about company size, their role in the company and how large a deployment they want. And also we're using third-party tools to enrich that information. Using all that information, we classify accounts that have potential to grow. So once we see paying account with engagement within the account and the potential to grow to, let's say, 1,000 people or more. Our sales team is engaging that customer.

And basically, it's a different process because the customer is already using Monday, happy with a product and then we try to expand the usage…And usually, it's a short sales cycle, which leads to future sale cycles as well.”

We can see below that Monday’s spend on performance marketing as a % of total S&M spend has fallen from 60% in Q1’21 to 33% in Q2’25.

As of 2019, self-serve channels represented 71% of new ARR added. As of 2023e, this number was down to 37%, with the sales team and the partner channel making up the difference.

It’s worth noting that Monday’s suite build-out has also helped with their GTM efforts as Monday can now target a broader array of buyers. Not only can Monday target prospective users of Work Management, but they can also target prospective users of CRM, Dev and Service. While CRM in particular is much more competitive/expensive in terms of performance marketing, this gives Monday greater optionality on how to spend its marketing dollars.1314

In Apr-25, Monday brought in Casey George as CRO, replacing Yoni Osherov who left to join Israeli VC firm Entree Capital. Casey laid out his priorities for the GTM function at Monday’s Sep-25 investor day: (1) multi-product growth, (2) accelerating upmarket and (3) customer retention.

• Multi-product growth. Monday wants to sell the full platform — WM, CRM, Dev and Service. This means cross selling to customers where appropriate (not something that’s been a focus of Monday’s to date) and bundling products together (bundles officially launched in Q4’25). It also means staffing product specialists (i.e. sales reps specializing in CRM, Dev, Service and WM), which Monday calls “product overlays.” Monday ended FY24 with 33 “product overlays,” and anticipates increasing this number to ~125 by the end of FY26

• Accelerating upmarket. To Monday this means targeting the mid-market. They feel this is where their platform resonates most strongly, and is the next logical step up from SMB. To the enterprise, Monday can really only sell WM today. But to SMB and mid-market, Monday can sell their full suite.

• Customer retention. Here Monday aims to enhance retention through (1) comp plans that incentivize retention, (2) more clear accountability over retention through “retention owners” and (3) by using AI to increase the number of “touches” that customers receive.

Financial Performance

Revenue Growth

Monday has 15x’d revenue since 2019 — a 58% CAGR over this period.

The growth rate has slowed over time from 100%+ in 2020 to ~26% projected for 2025.

Here are the quarterly numbers. Monday has truly grown like clockwork.

Monday enacted a price increase beginning in Jan-24 that added about ~$30m and ~$40m in revenue to 2024 and 2025e, respectively. In the absence of this price increase, YoY growth would’ve been ~29% and ~22% in 2024 and 2025e, respectively.

Customer Growth

Monday has 2.8x’d its customer base since 2019 — a 23% CAGR over this period.

Overall customer growth was very healthy until 2024. In 2024, customer growth was 9%, a big drop from 22%/21% in ‘22/’23, respectively.

Offsetting the slowing in overall customer growth has been strong growth in large customers. From almost none in 2019, Monday now counts almost 4k customers with $50k+ in ARR as of Q3’25. This segment has grown at a 111% CAGR since 2019, and grew 39% in 2024.

While overall customer growth has slowed, Monday suffers from the law of large numbers more so than its peers. Monday has a huge number of customers…

Monday’s total customer count of 250k+ is way above its average peer. Monday is ~tied with HubSpot for the third highest customer count out of the peer group behind just Shopify and Atlassian.

But while Monday has a huge number of customers, within their base of customers (especially enterprise customers) they are severely underpenetrated relative to their potential opportunity.

Their experience with the Fortune 500 is illustrative. As of 2023, Monday had users within 61% of the Fortune 500, but its average penetration was <1%.

So while customer growth has slowed, we’ve continued to see solid NRRs…

Net Retention

Monday’s overall NRR has actually risen over time from 100% in 2019 to 112% in 2024.

Below are quarterly figures through Q3’25.

Below are NRRs across the peer group. The only peers with better overall NRRs (that report the figure) are Snowflake, Datadog and Crowdstrike (three companies that trade at an average LTM revenue multiple of ~25x vs. Monday at ~5x).

While Monday doesn’t disclose gross retention, management stated that gross retention was in the mid-eighties in a Sep-23 SaaStr interview. That, combined with the following chart, provided at their 2025 investor day, would indicate that gross retention is currently in the high-eighties range.

Impressively, as we can see above, Monday’s current gross retention is at an all-time high, which no doubt has to do with their move up market.

Margins

Gross margins sit at 89% as of 2024. They have climbed ~450 bps since 2019, representing an average annual increase of ~93 bps.

Monday’s GMs are in line with best-in-class SaaS peers.

Overall opex as a % of revenue has declined from 203% in 2019 to 91% in 2024.

R&D as a % of revenue is surprisingly low for a relatively smaller scale SaaS business. It sits today at 22%, which is down from 32% in 2019 (representing a decrease of ~191 bps per year).

At 22%, it is well below the median of its SaaS peers (~27%). The only SaaS peers with <20% on this metric (Salesforce, Shopify, Intuit, and Adobe) have at least $10bn of revenue, or 10x+ that of Monday’s.

G&A has declined ~500 bps since 2019 and sits at 15% of revenue. This is on the high side vs. peers (peer group median is ~10-11%) and should decline over time with scale.

The biggest decline by far has come in S&M, which has dropped as a % of revenue from 152% in 2019 to 55% in 2024. Around a third of S&M spend is performance marketing, with the balance consisting of payroll for sales, customer success, partner channel, and related.

While S&M spend has fallen quickly, Monday still spends well above the median peer. Monday spent 55% of revenue on S&M in 2024, which was the highest amongst the peer set other than Asana.

However, while S&M as a % of revenue is elevated relative to peers, S&M spend has been relatively efficient. Monday’s GM adjusted CAC payback on its S&M spend has hovered around the mid 20s (average of 27 months since Q3’19).

For 2024, Monday generated a CAC payback of 29 months, which is slightly below the horizontal SaaS peer median:

And this is despite a significant ramp in Monday’s S&M spend in 2024 that would have been a drag on efficiency (as new reps take 12+ months to fully ramp).

Monday slowed hiring significantly in 2023, which led to an inflection in margins. The company has since reaccelerated hiring, but that pace is expected to slow again through H2’25 and into 2026.15

Overall, the operating margin has inflected sharply over the past 5 years from deep in the red to breakeven on a GAAP basis and 14% on a non-GAAP basis.

Monday really shines when it comes to FCF margin:

Monday outperforms on FCF because they put a strong emphasis on collecting cash upfront from customers. As we can see below from their Sep-25 investor day, ~83% of customers pay upfront (I believe all yearly and multi-year contracts require upfront payment).

Here is Monday’s FCF margin vs. the peer set:

Over time as growth slows, FCF margin will converge with Monday’s non-GAAP operating margin (less tax expense).

Investment Merits

Highly Attractive Financial Model

Monday benefits from 100% recurring revenue, ~90% GMs, high operating leverage and strong excess cash flow generation.

But in addition to these fine characteristics (shared by best-in-class SaaS peers), there are some unique aspects to Monday’s model that make them potentially even superior to other high-quality SaaS companies.

• Very efficient R&D function. Monday is “platform first” in its product development, which has led to a highly efficient R&D function without sacrificing (actually improving16) the quality of its products. Once the platform is in place, each additional offering (thin “layers” built on top of the platform) is much quicker and less costly to develop1718

• Better growth “durability.” Monday is like a tapeworm. Monday doesn’t generally make big upfront sales of licenses, like a CRM might or an ERP might. It makes a small initial sale, then it spreads — from one division to another, from one geo to another, and eventually to other products. Thus, Monday’s growth should, in theory, demonstrate more “durability” with more sustainable NRRs (which we’ve seen) than perhaps your typical software company.

Huge TAM

Monday’s TAM is enormous. At its 2025 investor day, Monday estimated its total TAM is $158B, growing 12%. This consists of WM ($58B), CRM ($32B), Dev ($23B) and Service ($45B).

Just about all of WM seems addressable, in theory, given Monday can serve SMB through the Fortune 100 (and any business vertical, and any geo). If we assume all of WM is serviceable, but only ~30% of CRM (mid-market and below), ~30% of Service (mid-market and below) and ~10% of Dev (SMB), that leaves a SAM of ~$80B.

Using a $80BN SAM, this would imply Monday is <2% penetrated into its market opportunity. Within just Work Management, where Monday is arguably the market leader, they are just ~2% penetrated.

Naturally Differentiated Products

Monday wins the same way across all its products — flexibility. What’s great about this, is that this flexibility is inherent to the Monday platform. So any product built on top of its platform (i.e. all of its products) benefits from this.

From 2023 investor day:

“Another important point about products that are built on top of the platform, is that every product that we build on top of the platform actually inherent all the advantages that people love about Monday. So if you look at reviews on CRM, for instance, and Dev, you won’t see specific features. You won’t see customers love it because of a specific feature or deep feature within the CRM. Rather, they’ll say, they love the customization abilities, the first time to adopted the flexibility, and of course, there is a fuse.”

More quotes on the uniqueness of what Monday provides.1920

While being more customizable is great, it goes without saying that for that to work, prospective customers need to want this. And here I’d say what Monday management preaches seems generally correct: that SMB products tend to be cheaper but more “rigid,” while enterprise products tend to be more customizable with all the bells & whistles, but more expensive to set up and maintain. Thus a gap of sorts exists in the market which Monday fills nicely. From a May-24 JPM conference:

“What I’m trying to explain to investors is that we offer something that’s very different from the current solutions in the market today. For example, in the CRM space, right now, we -- when we compete on deals, we might see a Pipedrive or Zoho or HubSpot. And I actually did quite a lot of user interviews like why are you choosing Monday over one of those players? And the answer is always the same.

Although we’re missing features compared to, let’s say HubSpot. The main reason is that they want that customizability to their product. At some point, when you use what we call a rigid product, you want to have your own workflows. You want to create something that’s kind of unique to how you work as a sales team.

And then usually, the next stop for most of those customers was to use Salesforce because it’s an enterprise software, you could build anything you want with their API. But it’s very costly to set up and very costly to maintain, and the license cost is very high. So for them, seeing Monday as an alternative was something they didn’t have before because Monday gives a product that is as customizable, let’s say, as a Salesforce as a ServiceNow. But it’s very easy to use. It doesn’t cost a lot to customize and you can do it yourself. And that’s a huge benefit in each one of those categories.

And even though we might be competing with a dozen other vendors in that space, nobody can offer the same level of customizability that we offer as a platform in any one of those verticals. And customers get it. They want to buy a platform. They want to have that power over their destiny, and this is why we win most of those deals.”

We see Monday’s differentiation show up in its reviews from G2 and Gartner, where Monday is amongst the highest rated products consistently (if not the highest rated).

A nice thing about differentiating through core flexibility (vs. say certain features), is that it’s less readily copied by competitors. Having this flexibility requires a certain architecture to the product from the outset. It is very, very difficult to re-architect your products once you’ve already scaled them up.

From the Dec-24 NASDAQ conference:

Roy Mann (Co-CEO)

“I think it’s true for a lot of companies like everyone want to be a platform, but the question is like are you built in that way? Like for us, it’s not -- we didn’t build a project management tool or a CRM. We built like a database that can be adapted to anything. We build like expandable UI set that everyone can build on top of. And essentially, that’s like how our infrastructure is built. So it’s really hard to become a platform post building stuff because you need to then go back and like tear out stuff. So we did it like day 1. And yes...”

Robert McCooey (Analyst)

“That probably is one of the keys to the success, right?”

Roy Mann (Co-CEO)

“Definitely. And customers -- so it also gives us different go-to-market, which is like allows us to grow faster. It gives us -- it gives customers the knowledge they’re never going to hit the wall, whatever they want to develop further on or customize, they know they can do it, and that wins deals.”

Strong Initial Traction in the Enterprise

Monday has only recently become a viable product for the enterprise.

And this only applies for WM. Monday seems pretty far from enterprise-ready in the rest of the suite — CRM, Service, and Dev.21 There’s no clearer example of their non-enterprise readiness in CRM than the fact that Monday itself uses Salesforce for its CRM, not Monday.

But going up market Monday is. And it does seem to be working quite well. See chart of Monday $50K+ ARR customers below:

And quarterly customers across the $50K+, $100K+ and $500K+ categories (YoY growth actually accelerated in Q3’25 across each of these buckets):

~40% of Monday’s ARR now comes from the $50K+ cohort, representing a massive shift from 2019, when just ~5% of ARR came from the cohort:

Up until ~2 years ago in Dec-23, Monday’s largest customer was 7k seats (which had been Monday’s largest customer, I believe, since its IPO in Jun-21). In Dec-23, Monday closed a 25k seat deal. Then a couple quarters later, in Q2’24, Monday closed an 80k seat deal for a healthcare business. Meanwhile its new 25k customer, a quarter later in Q3’24, upped their seat count to 60k. So in the span of less than 1 year, Monday went from a top customer of 7k to having two customers with seat counts roughly ~10x larger (60k and 80k).

Enterprise readiness has resulted from (1) new features/functionality geared towards the enterprise (e.g. permissions, audit logs, domain block and IP restriction) and (2) perhaps most importantly, a reworking of Monday’s underlying database (an initiative called MondayDB) to allow for faster load times while working with large sets of data.22

Here’s a rough, non-exhaustive timeline of enterprise-oriented feature releases:

• Q4’22: Custom role based account permissions

• Q2’23

  • MondayDB 1.0: large boards loaded at <4 seconds

  • Multi-regional support (EU & Aus)

• Q3’23: Premium support

• Q4’23: Adds 25k-seat customer

• Q1’24: MondayDB 2.0: 100k items per board, 500k items per dashboard

• Q2’24: Adds 80k-seat customer

• Q1’25:

  • MondayDB 3.0 Premium: 1mm items per board, 5mm items per dashboard

  • Enterprise Project Management (EPM) offering: sits atop work management, allows for managing projects at scale, allows CIO to enforce more standardization/structure in Monday through a company, better resource planning capabilities

In addition to the product not being ready for the enterprise, the GTM motion wasn’t either. As discussed in the GTM section, prior to 2018 in fact, Monday didn’t have much of a sales team at all. The company had been pure self-serve, PLG. Starting in 2018, the company made a rapid about-face.

CRM + Service

CRM was launched in Aug-22. Just three years later, in Aug-25, CRM crossed $100m in ARR. For a SaaS business, as we can see below, that’s quite good. In fact, if it were a standalone start-up, it would’ve been one of the fastest growing SaaS start-ups ever.

Of course it had a big advantage in that it could cross sell into Monday’s existing customer base, some of whom were already using Monday as a CRM. But regardless, I think it’s fair to say that CRM has been a solid win for Monday thus far.

In a Feb-25 interview with SaaStr, Eran details some of the reasons behind Monday’s success in CRM. This is a long exchange, but I think illuminating:

Eran:

“You mentioned CRM. We have great momentum with CRM and we’re super proud of the growth. It’s actually growing faster than what Monday did back in the day, which is pretty impressive.”

Jason Lemkin:

“Broadly speaking, CRM is the one of the oldest and most crowded categories, right? And so your smallest customers they might have nothing, right? But as you get just a little bigger, there’s some conflict with another vendor. So why is it working at that level because it’s not greenfield in some cases, right?”

Eran:

“Yeah so I would say it boils down to two main reasons.

One, which I think is a little bit underestimated, is the flexibility of the Monday platform. People, when they buy, whether it’s a CRM or Service or whatever it is, I think the SMB segment usually or the mid-market segment, is usually the players are, what I refer to as, rigid software. It’s software that probably has a great user interface, but it’s very hard to customize. And if you want to further customize it, you have to go enterprise. You have to buy like the big enterprise companies that offer 100% flexibility. But the problem there, is the cost of doing that and the complexity of doing that, in those enterprise companies, is far greater than any other platforms. We offer the best of both worlds. You can customize Monday to infinity, basically. But it’s easy to use and you can do it yourself. So that’s reason number one…

And the second one, which is a big part of our strategy going forward, and also I think maybe a little bit underestimated, is that people want to connect between the different softwares. I think softwares today inherently creates silos within the organization. Somebody have access to Salesforce, usually the sales team — the marketing team or the product team doesn’t have access. Same goes for Zendesk or any other SaaS tool you’re using. And then you essentially creating silos of information between different teams, just the nature of using different SaaS tools. And people, when they buy multiple products from Monday, and because it’s kind of multiple products for different departments, they love the fact that data flows between different products. They love the fact that they create a dashboard that shows them a picture of what’s going on all the way from the Service to the Work Management to the CRM tool. And I think over time as we scale those products and gain more momentum in enterprise companies, we’ll be able to sell bundles of those products and people will buy the Monday brand, hopefully.”

Jason Lemkin:

“Yeah I think that’s a great point. I think there’s been a lot of discussion about platforms versus standalone vendors as there’s been consolidation. But I think it’s not really that — it’s if you could have it all work on one platform, you’d want to. Who in their right mind wants — now sometimes you have to and Monday got started as a point solution because you had something special and so it’s not an either or — but, no one in their right mind wants 11 vendors. Especially today, when it’s harder to get them to integrate, when prices are up everywhere, right? Why would you want that?”

Eran Zinman

“And especially now that AI becomes more dominant. You want AI to have more access to more information across the different products. And you get more benefit from that, so definitely. And because of that, you don’t necessarily have to have the best CRM in the world. It might be second place, but because of that integration and because of that data flows between the different products, you get such huge benefit from that, that it’s a huge advantage.”

At the same time, Service (launched Feb-25) is actually growing faster than Sales CRM was at the same stage of development.

And there are reasons to believe that Service is actually a much more natural cross sell opportunity for Monday than CRM. Here is Eran talking about why in a Feb-25 SaaStr interview:

“We just recently released Monday Service. And what we found is that we have a big overlap — not just in the use case between Monday Service and Work Management, but also in terms of the buyer, which is as important as the use case. So usually the person who buys Work Management is the CIO, COO. Same goes for Monday Service. So you’ve got their attention, you’ve got the relationship, and it just make it easier to do the cross sell.”

Given that (1) Service is actually growing faster than CRM, (2) the cross sell opportunity is more natural, (3) the Monday product is very well suited for Service, and (4) the Service TAM is larger and ASP higher than CRM, I think it’s fair to be optimistic re Service.

So where does that leave us? Well, this is what the Bessemer chart above, rolled forward from $100m to $1bn, looks like for best-in-class SaaS start-ups.

Best-in-class SaaS start-ups have gone from $100m to $1bn in 3-7 years.

I would say that if CRM is a winner (appears that way) and if Service is a winner (seems probable, but early), it would be hard for Monday as an investment overall to not be a winner. CRM and Service could potentially add ~$1bn of revenue over the medium term, say 7 years, on a revenue base that is ~$1bn today. In this scenario, Monday would double revenue before even accounting for continued growth in WM (where Monday is the market leader and less than 2% penetrated) or AI revenue of any kind.

Showing material adoption in CRM and Service would also considerably enhance Monday’s competitive positioning/stickiness and contribute to a re-rating of the stock, one would think.

Flywheels

I think there is real evidence Monday is building virtuous flywheels that should help it sustain its momentum:

• Flywheel #1: Product quality. Monday is a true platform-first company. This is very hard to do after the fact; you need to build this way from the beginning. As more investment goes into Monday’s platform, every product gets better simultaneously, and each successive product is of higher and higher quality out of the gate. There is good evidence that this flywheel has taken hold at Monday. It’s CRM, developed with an initial team of just 4 people, has immediately become a top rated option on G2. CRM grew to $100m in ARR in just 3 years. Service, Monday’s latest offering, is on a similar trajectory.

• Flywheel #2: Distribution. The more customers you have, the more cross sell opportunity you have. Every product you release can grow faster than the last. See CRM outgrowing Work Management, and Service potentially outgrowing CRM. Small tuck-in acquisitions can be dramatically scaled up through cross sell to a large install base. Monday has 250k customers, one of the largest customer bases in the SaaS universe. And it is having real success going upmarket as well.

• Flywheel #3: Bundling. Monday just started bundling its products in Q4’25. I think this will be a material tailwind for Monday’s cross sell efforts. As the bundle gets larger, pricing should become more of a moat.

  

• Flywheel #4: Integration + AI. In general, customers want more integration/consolidation of their software stacks, not less. During the SaaS era, we may have gotten use to the idea that customers preferred best-of-breed apps. But before SaaS, consolidated suites were more common. It may be the case that SaaS products were so superior to on-prem, that a window in time opened for a large number of best-of-breed SaaS players to thrive. Now, however, well into the SaaS era, it may be the case that environments will again consolidate, as product suites from the leading players grow. Here is Parker Conrad from Rippling making this point.23 Furthermore, I suspect AI will be an additional “consolidating force,” as the benefits of having (1) all your data together in one place and (2) agents that can natively dip into various applications to perform actions, will be increasingly important.

Other Merits

• Strong management team. Monday is still led by its co-founders, Roy Mann and Eran Zinman. To this point, they’ve done an excellent job. See management section

• Monday can be VMS. While Monday is horizontal SaaS, Monday can also be vertical SaaS. Because Monday is so easily customizable, for some verticals, Monday can make sense as vertical SaaS. For some verticals, the VMS provider(s) may be quite subscale, and thus wouldn’t have the resources that a Monday would to invest in their platforms. Thus, even though Monday’s offering might not be tailored to a certain industry out of the box, Monday’s offering might still be the superior solution. As Monday’s offering becomes even more easily customizable — for instance through Monday Magic/Vibe — this opportunity should grow larger. Here’s a nice quote on the subject from Parker Conrad of Rippling.24 Here’s additional commentary on how Monday can be a great VMS fit.2526 An example of Monday replacing a VMS provider in the RIA space.
  
  I recently came across this BMO report on Constellation software. The report discusses some of the niche offerings from Constellation-owned companies, like forestry software in Sweden and software to manage K-12 school lunch programs.

  

  While I’m not sure what kind of impact platforms like Monday could have on these businesses, I do not think all these companies would have existed if Monday-esque platforms had always been available. Further, given how important AI is likely to be, and the fact that platforms like Monday will have much more resources to invest in AI, this likely gives the platforms even more ammunition to compete with subscale VMS

• Additional applications. Monday could release additional applications over time. They’ve hinted at an HRM a couple times in the past2728, though this seems hard to do.29 Perhaps a knowledge hub product (like a Confluence from Atlassian)? CFO Eilran Glazer on a recent podcast referred to ServiceNow as a “North Star” for Monday due, in part, to ServiceNow having eight businesses with over $250m of revenue.30

• Lack of obvious non-starters. No customer concentration. $1.7bn of cash, no debt. Strong cash generator. Strong track record. Not wildly overpriced

Investment Risks

Future of SaaS Risk #1: Cost of Software Development Going to Zero

The rise of codegen and vibe coding apps is a trend that isn’t going to reverse. In theory, over a long enough time frame, the cost of software development could approach zero. As such, it seems very possible that greater competition will eat away at pricing power, reducing margins.

At this point, nobody knows for sure how this will play out, including folks much smarter than I on software. My guess would be that yes, there will be greater competition and maybe pricing power erosion, but that it can be mitigated somewhat:

• I would say the vast majority of companies are not going to develop custom apps for there core workstreams. Issues such as security, permissions, reporting, data integration, software maintenance/support are real things and companies will not want to bring them in-house.

• Further, I would say there will still be degrees of “lock-in:”

  • If a company is running many/most of its core workstreams on ServiceNow say, they will do everything in their power to stay on ServiceNow. Lifting and shifting all your data and workflows and know-how to some other platform still seems like it will be incredibly painful, even in this new world

  • Companies will come to have potentially hundreds of agents embedded within their software workflows. These agents will come to understand how your business operates, and be customized in many thousands (hundreds of thousands? more?) of ways (including likely the insertion deterministic logic where appropriate). Trying to replicate this agent layer on another platform would likely involve significant pain

As for Monday more specifically, I think they benefit relatively speaking from the fact that their software is lower priced. Satya Nadella pointed to this advantage recently on a couple of podcasts. I believe his point is essentially just that there is less room to undercut Microsoft on price which should help insulate them from potential new competitors. Monday is the cheapest out of all the platforms (except perhaps Atlassian).

Further Monday also has much software development to do, across a whole suite of applications. So Monday itself is a beneficiary of this, perhaps more so than many others who might have a more limited scope of potential offerings. Monday has recently taken a full month to reset its R&D function around better leveraging AI.

Future of SaaS Risk #2: Business Model Uncertainty

(1) AI workloads are much more costly than SaaS apps to deliver — this will lead to gross margin erosion. (2) Additionally, seat-based pricing models will have to make the difficult transition to consumption-based or hybrid pricing.

In terms of gross margins, Monday management has stated that GMs are likely to be pressured over the medium term. Management said at their Sep-25 investor day that GMs might contract into the “mid-80s” as a result of “incremental AI-related compute costs.”

They were asked about it during Q&A and had the following exchange:

Analyst

“I was wondering as you drive adoption of AI agents and AI utilization across the customer base. We’ve seen some others talk about the potential impact on margins when you’re in this early curve of investment. And before monetization starts to kick in. So I was curious, can you talk a little bit about how you’re thinking about driving margin expansion with this next investment cycle in AI?”

Eliran Glazer (CFO)

“Maybe I would start, in the presentation, I shared the model about what would be the impact on our gross margin. So I said potentially, we are a very high margin company in terms of gross margin. We have achieved 90% throughout the last four years. We believe there is going to be an impact of a few hundred basis points. And what we have assumed is probably that we are going to be in mid-80s, but we think there is an upside on it because we are very efficient in the way we run our business. So I would say this is probably the range I would think there is going to be potentially some impact between 85% to 90%.”

In my model I have GMs contracting to 86% by 2031e, which seems relatively conservative based on this exchange. Further, I’m taking GMs down while not really giving Monday any credit for any AI-derived revenue. I assume Monday (and others) will be able to monetize their AI offerings (in the form of consumption pricing or higher seat-based pricing) to at least some extent.

I concede however that there’s a relatively wide range of potential outcomes on longer-term GMs (across all of SaaS).

In terms of seat-based pricing vs. consumption vs. hybrid, maybe the biggest risk would simply be that Monday will be overly concerned with temporary GM compression. This might lead them to overprice their AI offerings relative to competitors, which would potentially cause churn or stunt growth. I do foresee Monday management being responsive to these types of competitive dynamics, but it is something to monitor. Monday currently charges on a consumption-basis for many of its AI offerings, and I’d say people generally don’t like this type of model (unless the ROI on this spend is very clear). I would expect the “Product Power-Ups” from Monday to be included in all Pro or Enterprise plans for no additional cost, but we’ll see.

Overall I think TAM should expand, however. And that companies that do AI well will be able to capture a portion of the value they create — whether that comes in the form of consumption based pricing or higher seat based pricing.

Future of SaaS Risk #3: AGI Renders the Application Layer Obsolete

AI will become so powerful over time, that most (all?) workflows needed today simply won’t be needed in the future. The “agentic layer” will displace the “application layer.”

I think consensus from leading experts today is that we’re perhaps a decade away, in ballpark terms, from just getting most agents to really work.

This exchange on 20VC from Oct-25 with Atlassian CEO Mike Cannon-Brookes seems to be generally reflective of our current situation:

Mike Cannon-Brookes (CEO, Atlassian)

“The speed with which AI can be understood and deployed successfully by companies/customers is something that is going to take a lot longer to really play through. Because the delta between magical demos and actual value delivered is quite high in a lot of cases.”

Harry Stebbings

“Why is that? Is that adoption and education? Is that data cleanliness and availability? Is that security and permission-ing in large enterprises? Why?”

Mike Cannon-Brookes (CEO, Atlassian)

“Every single one of those and more…You have to understand your problem well. You have to change the way you look at the problem — often the workflow, the business process. You have to understand the technology and what it is and is not capable of at the current time. You have to have the security, the data, you have to have the cleanliness, you have to have all these things. The magical demo of just like a push a button and like some app pops out or this thing happens. It doesn’t exist, right? It doesn’t mean the technology is not useful at all, but I think there’s a lot there’s a lot more work to do to to really deliver on the promise.”

Here is commentary from Salesforce CPO Steve Fisher from their Oct-25 investor day on some of the challenges they are having building agents at a more granular level:

“Why has it been so hard to get agents that actually deliver the goods?…

So we would have customers tell the agent in natural language through prompts like everybody is doing, first do this and then do this and then do this, but never do this or always do this. And if this is true, then do this, but otherwise do this. If you looked at the prompts for all of our customers, they’re riddled with that. But what happened was that the LLM would maybe most of the time do what you told it to do, but not all of the time would it do what you told it to do. And that’s kind of been the experience is that the reasoning ability, the ability to always follow instructions, that’s not really the LLM strength. And the realization was, if you actually know what you want the agent to do, don’t go to the LLM. Just say, do this and then do this. And if this, then do this, else do this, kind of the basics back to the traditional software role that we’re all used to, but I think maybe we thought the agents could kind of just figure it all out on their own. But they get very confused. They can’t do that. They’re designed not to do that. They’re designed really to figure out the next word to say…

So then they would redo the prompts and redo the prompts and get into what Srini calls the prompt doom loop…then if it’s not listening to you, you’d say, never do this and you put it in all caps and repeat it 3 times. Literally, that’s what our customers were doing…

And the first thing we learned was actually, if you don’t constrain the agent to maybe no more than 6, 7, 8 things that they should be paying attention to, it gets very confused. And then we learned, well, even within those 6, 7, 8 things, use the LLM where it’s appropriate, do not use it where you actually know what to do. That’s this deterministic capability with Agent script. And all of that has sort of said, okay, Software is still software. In the world -- in business, when you actually know what you want, whether it’s for compliance reasons or you just want to execute your business process 100% the way you want to execute it, then just use the enterprise software you’ve been using all along.”

Of course Atlassian and Salesforce are “incumbents” so you might disregard such statements. But it’s also just a reality that AI-natives in these “messier” areas of software (where not everything is text-based; where data is not self-contained or there is external context; where evals don’t necessarily exist) are not growing as fast as AI-natives in other areas that are more ripe for immediate AI disruption (like software development, legal, creative arts, customer service).

Here is Andrej Karpathy (founding member of OpenAI and former Director of AI at Tesla) in an Oct-25 interview with Dwarkesh Patel. He makes the case that we are still about a decade away from workable agents.

Dwarkesh Patel

“Why do you say that this will be the decade of agents and not the year of agents?”

Andrej Karpathy

“…The quote you’ve just mentioned, ‘It’s the decade of agents,’ is actually a reaction to a pre-existing quote. I’m not actually sure who said this but they were alluding to this being the ‘year of agents’ with respect to LLMs and how they were going to evolve.

I was triggered by that because there’s some over-prediction going on in the industry. In my mind, this is more accurately described as the decade of agents. We have some very early agents that are extremely impressive and that I use daily —Claude and Codex and so on — but I still feel there’s so much work to be done.

My reaction is we’ll be working with these things for a decade. They’re going to get better, and it’s going to be wonderful.”

Dwarkesh Patel

“What do you think will take a decade to accomplish? What are the bottlenecks?”

Andrej Karpathy

“Actually making it work. When you’re talking about an agent, or what the labs have in mind and maybe what I have in mind as well, you should think of it almost like an employee or an intern that you would hire to work with you. For example, you work with some employees here. When would you prefer to have an agent like Claude or Codex do that work? Currently, of course they can’t. What would it take for them to be able to do that? Why don’t you do it today? The reason you don’t do it today is because they just don’t work. They don’t have enough intelligence, they’re not multimodal enough, they can’t do computer use and all this stuff.

They don’t do a lot of the things you’ve alluded to earlier. They don’t have continual learning. You can’t just tell them something and they’ll remember it. They’re cognitively lacking and it’s just not working. It will take about a decade to work through all of those issues.”

Dwarkesh Patel

“…Why a decade? Why not one year? Why not 50 years?”

Andrej Karpathy

“This is where you get into a bit of my own intuition, and doing a bit of an extrapolation with respect to my own experience in the field. I’ve been in AI for almost two decades. It’s going to be 15 years or so, not that long. You had Richard Sutton here, who was around for much longer. I do have about 15 years of experience of people making predictions, of seeing how they turned out.

Also I was in the industry for a while, I was in research. I have a general intuition that I have left from that. I feel like the problems are tractable, they’re surmountable, but they’re still difficult. If I just average it out, it just feels like a decade to me.”

Later he describes the ways in which codegen apps mostly fail to be useful to him. He likens codegen apps, one of the best use-cases we have today for LLMs, to “slop:”

“I think overall the models are not there. I kind of feel like the industry — it’s making too big of a jump. And it’s trying to pretend like ‘this is amazing,’ and it’s not — it’s slop. And I think they’re not coming to terms with it and maybe they’re trying to fundraise or something like that — I’m not sure what’s going on. But we’re at this intermediate stage. The models are amazing. [But] they still need a lot of work.”

We can see agents “failing” in the enterprise from a number of recent studies as well:

So I’d say it seems as if we are in for a steady, many-year grind of turning these LLMs into “applied” use-cases.

New technologies also tend to be “asymptotic.” Meaning the pace of change is extremely fast at first, and much more gradual later on. Here is Peter Fenton of Benchmark on the Uncapped podcast (Jul-25):

“If you get to a speciation event, 90% of the phenotype change occurs in the first 10% of the species life. Why is that interesting? Well, you look at the iPhone. Okay, so for the first 10% of the iPhone’s life, 90% of the change — of course — like that’s when the camera gets added on the front. That’s when you have video and then it sort of goes to an asymptote. So we’re in a zone now — and I think maybe it’s a three to five year period, could be seven. Who am I to say? Where we’re having radical variance and what will happen is we’ll look back I think in 2040 and and you blur your eyes and say it all kind of looks the same like it did in 2030. But from 2025 to 2030 it’s going to look so different every 6 months.”

This doesn’t mean the application layer is safe, necessarily. But I do think the rapid pace of change we are witnessing now in AI won’t go on forever.

Extreme Competitive Intensity

There is intense competition across all of work/project management, CRM, Dev and Service. However, what is most spooking investors seems to be the specter of groups like Google, Microsoft and OpenAI making even greater inroads into the enterprise. And doing so in ways that increasingly encroach on Monday’s turf.

Much of the fear seems revolve around new “AI layer” offerings from these companies that many believe will come to sit atop of the enterprise stack. Companies will use this layer to manage their businesses and their agentic workflows (amongst other things). This encroaches on Monday’s turf, where automations and workflows are key selling points. Further, the natural course for Monday as a WM provider is to lean into agentic capabilities, which will put them even further into competition with these groups.

So should we be spooked by these offerings?

I think the answer is yes, we should probably be spooked to some extent. Google/Microsoft/OpenAI are making it a very clear priority to attack this AI layer, putting them into competition with Monday (even if it’s still mostly indirect competition), and that’s clearly not great. It has the possibility of putting Monday onto a treadmill where it needs to spend more to compete, but it’s ultimately in a battle it can’t win.

The flipside of this argument is that fears of these large companies just completely dominating the enterprise are likely overblown. Microsoft does not dominate CRM, ERP, etc. despite having offerings in those areas — far from it. There is still a lot of power in specialization and room to build better products for all kinds of different use cases.

Below is an exchange from an Aug-25 a16z podcast touching on the points above:

Aaron Levie:

“I think that people are probably way too afraid of the model providers eating them. I think it was basically a phenomenon in the first wave, which was if you were just doing like…a text interface that produced more text, like yes ChatGPT ate you…But basically since then most enterprises want, kind of, applied use cases for AI and AI agents. And so it’s not obvious that the current crop of companies — if you’re doing AI for healthcare, if you’re doing AI for life sciences, if you’re doing AI for financial services…We’re just in a 5-year period right now where you’re going to have to build agents for every vertical, every domain and there’s a playbook that’s starting to emerge of what that needs to look like.”

Martin Casado

“I think there was kind of a technical head fake that happened early on which was pre-training. So the pre-training really was a 10 out of 10 technical innovation…Like two years ago if somebody — I had a friend that was building their own aging model, post training aging model…and then of course the next version of Midjourney whatever comes out and it does a better job of it. And the thing with pre-training was you’re just kind of consuming all of the world’s existing data. You’re draining all of that energy and it perfectly generalized, right? But it feels like technically that’s passed and now we’re more in post training, in RL which is a lot more domain specific.”

Aaron Levie:

“Well and the moment that you have access to some set of data that is just for that enterprise — who gets permission to access that data, who gets permission to do the workflow on it? It’s going to be applied companies.”

Martin Casado:

“Yeah. If we had an infinite number of tokens then the models would just continue to generalize, but it’s pretty clear that that’s not happening. And so now we’re going into — which we all understand very well — which is now companies have to choose which domains to go into and they’ve got to solve the long-tail problems there and get access to the data, etc.”

Steven Sinofsky

“And I also think that there’s this shadow…cast by large companies over we’re going to put you out of business and stomp you. It’s ridiculous. And it has never in any technology wave lived up to the fear that people have…

The thing that’s not being said now, which we should do a whole [podcast] on, is like what is the actual platform? Because it’s all well and good to say that the large models will go subsume every application. The thing is, the minute they start doing that, no one will be in their platform, right? …There’s a phrase ‘Sherlocking’ on the Mac and the Apple world to this [phenomenon]. And so it does it have a real chilling effect and that’s one of the things all the model people are going to learn very, very quickly.”

Aaron Levie

“There’s a chilling effect, but there’s also just a problem of like it’s hard to go deep in 50 categories...At some point it becomes purely just an execution issue which is like, I don’t know how anybody would set up a company to be to be able to beat 50 startups across 50 different domains.”

Steven Sinofsky

“No, it’s ridiculous. And in fact, like it’s only good because what happens is that the big company raises the awareness of a whole category, and then you just swoop in and you go, ‘To them you’re just a feature, but to to you this my whole life,’ right? And and you’re gonna win. Look, I just always come back to — there’s a whole company that just signs things, right? [DocuSign] Like I cannot believe there’s a whole company that just signs things.”

I tend to agree with these takes. I also tend to think this layer will be very technically difficult to accomplish, as discussed in the section above.

Another group of competitors I’d expect to grow are ServiceNow copycats, of which Monday is one. These integrated suite providers seem to be positioned well in the AI era. We were already seeing the large horizontal SaaS incumbents (Atlassian, Salesforce, Workday) extending their suites. Rippling is an example of a start-up that’s building a broad suite. Reevo is another taking an AI-native approach to building an all-encompassing GTM suite.

AI Native Competition

Will Monday and the other incumbents be outcompeted by AI natives in their core offerings like WM, CRM, Dev and Service?

You’d be right to be a little spooked by the AI natives. Some AI natives are growing shockingly fast, much faster than the top SaaS companies of yesteryear:

However, I think some areas are clearly more ripe for immediate AI disruption than others, as discussed earlier.

These fast-moving examples may lead folks to believe that owning workflows/customers/data doesn’t matter. Or that doing so may even hamper an incumbent (as it locks them into a product or revenue model that isn’t adaptable to the AI era). And that will be right in some cases. But I think owning the workflows, data and distribution will still advantage incumbents in many categories.

I think the categories that Monday touches are such categories more or less — we have not seen any high flying winners in areas like CRM or Work Management. So while AI natives are certainly a risk, I think it’s more of heads-up competition, where Monday has some clear advantages.

Monday Agents Won’t be Very Good

Perhaps Monday’s agents won’t be very good. Monday is not a specialist in CRM nor a specialist in Service. In an area like customer service, there are customer service-focused, AI-native competitors like Decagon and Sierra that will very likely have better customer service agents than Monday.

If Monday knew they were going to have to develop all these agents, would they have still launched all these apps? They don’t even have all the core software functionality they need, never mind all the agentic capabilities…

So how much of a problem is this? Well, it’s probably agent-specific:

• Is the agent super important? Say there is a sales outbound agent that emails prospective customers. It is great at researching and drafting custom emails that hit just the right notes. This could be quite valuable. If some competitor’s agent was much better than Monday’s, that would certainly be reason to go with a competitor.

• How much worse is Monday? Say there is a customer service agent that handles 95% of issues autonomously, while Monday’s agent handles 90% autonomously. And say customer service for the customer is of relatively lesser importance. Well then Monday being a little worse is probably not reason for this customer to go out of its way to use this competitor (especially if Monday’s offering were cheaper, say).

I could certainly envision a world where Monday’s agents are “pretty good” in many cases, but where “best-of-breed” specialists exist as well. Monday customers could integrate with best-of-breed providers where they feel the need, and use Monday where Monday is “good enough.” In this world, custom Monday agents could also handle the bespoke agentic needs of a business that wouldn’t necessarily be the territory of a best-of-breed specialist — of which there will be many such use cases.

Here’s an example discussed in an expert call with Omnitas Consulting (Tegus, Nov-25):

“I have, for example, a client that does new product launches. One of the big things there is that the photo shooting team — they’re always the bottleneck. They get really bad briefs, they need to send them back days before the shoot. There’s a lot of frustration around it.

As soon as I get my fingers on an agent, I would like to try that use case and see — can we actually train the AI on what’s a good and what’s a bad photo shoot brief? Once the brief is submitted, it can automatically revert back to the one who sent the brief and say, "‘Listen, you need to put more color here. This information is missing.’ We know that AI is really good at doing this because we’ve done something similar for another client but with OpenAI, and we had 98% good results. It was only 2% that needed to be manually reviewed. In that way, it can be crazy. Imagine having a full-time photo shoot assistant that works 24 hours?”

Monday is also offering third-parties the ability to sell agents in Monday’s marketplace. If this marketplace turns out to be relatively robust that could extend Monday’s capabilities by an order of magnitude. This would provide even more reason to stay within Monday’s immediate ecosystem.

Ownership Structure / Founder Share

Co-Founder and Co-CEO Roy Mann holds a “founder share.” This gives him the right to veto the sale of the company.

From a Nov-24 article:

“When asked whether the company’s current market valuation makes him less worried about a potential sale, Mann explains: ‘When we were a private company, we blocked any possibility of an acquisition. Since going public, we’ve added a special section to the prospectus that explicitly outlines our commitment to remain independent. We always rejected offers outright to eliminate any temptation. You must stand firm in your intention to build a big company.’”

So do I think Monday would sell if a buyer came along and offered a big premium? Based on the commentary above — no, I do not. Does this make me less comfortable as a shareholder? A little bit — the presence of a valuation floor below which a take-out becomes likely is not a given with Monday. This might eliminate a bit of downside support.

However, to date, building as if they were going to be a big company (and never sell) has really worked out for Monday. The investment in Big Brain, the investment in the platform, going multi-product — this is what’s going to save Monday.

Other Risks

• Not a True SoR; Monday is “Bundle-able.” There’s a long list of non-SoR productivity tools that have gotten “bundled” by a Microsoft et al (Slack, Zoom, Dropbox, Calendly, Miro, Netscape, Winamp, Lotus, WordPerfect, etc.). You could make the argument Monday is such a tool. However, I believe Monday belongs in a higher tier. I think there is enough differentiation to Monday, that it cannot be “bundled” in a way that impairs the business (in fact, Microsoft already offers “Projects” and “To Do,” which are work/task management tools). The more that Monday can expand into other product categories, the less “bundle-able” it will be. So while Monday is not as sticky as a Salesforce or another pure SoR solution, I would say it’s much stickier than someone like Asana. And the numbers back this up. NRR is best-in-class for a SaaS company of its size. GRR is at an all-time high.

• Dev Struggles. Dev is not penetrating well. The cause seems to be that Atlassian is highly entrenched.

• CRM (and Service) Product Limitations. Monday’s CRM placed dead last in Gartner’s magic quadrant. Service is also quite limited at this stage. We need to believe that these products will improve at a healthy clip over time. If CRM and Service growth were to slow materially, that would introduce a significant headwind to the short-term and long-term story. Monday stopped reporting customer count figures for CRM and Service as of Q3’25 — this is generally not a positive indicator.

Valuation

Multiples

Monday currently trades at ~5.1x revenue / ~5.7x GP / ~19x FCF on 2025e management figures and ~4.2x revenue / ~4.7x GP / ~15x FCF on 2026e analyst consensus figures.

This is the lowest level Monday’s ever traded. Below is a chart of Monday’s NTM rev multiple over time. Since Mar-22, it’s generally traded in the 7-11x range (vs. ~4.5x today).

Shortly after the IPO, shares traded as high as ~50x fwd rev, which explains why shares are still well below peak PPS levels of ~$370.

What’s Causing the Drawdown?

Monday shares are in a steep drawdown, falling ~50% over the past 4 months.

The drawdown is resulting from some combination of AI fears (covered above) + slowing growth.

On its Q3’25 earnings call last week, management guided to 22-23% growth for Q4’25, which is down from 26%, 27%, 30%, and 32% in Q3’25, Q2’25, Q1’25 and Q4’24, respectively. In other words, growth has slowed to ~23% from 32% just 4 quarters ago. When a company goes from growth of 30%+ to flirting with teens growth, a re-rating of some sort is to be expected.

One much discussed cause of the slowdown has been a slowing of traffic from Google search (organic and paid) due to the rise of ChatGPT and Google’s own AI Overviews. Google AdWords (paid search) represents about ~10% of new ARR for Monday, according to management. From reports I’ve seen, Monday’s search traffic growth turned negative right around the end of 2024, with traffic cratering by as much as 20% overall (and over 25% for organic search) by mid 2025. These trends seem to be stabilizing, which management also highlighted on Q3’25 earnings.

Monday is attempting to mitigate this impact by shifting some marketing spend to other marketing channels as well as into more sales resources. However, (1) shifting Google AdWords spend (which pretty quickly translates to new revenue) to sales personnel (which have a much longer payback) will introduce a bit of an air pocket, even if Monday can generate similar ROIs on the spend, and (2) you can’t really “shift” budget around when you lose organic search traffic (because organic search is of course free, less a little SEO expense).

The slowing growth calls into question whether the $1.8bn of revenue the company guided to for 2027 at its Sep-25 investor day (which implies a 21% CAGR from 2025e) is still achievable. On the Q3’25, management reaffirmed the $1.8bn number after being pressed on it multiple times:

Steven Enders (Analyst)

“And then I guess on the guidance, again, just -- I understand Q4 is coming down a little bit. I guess on the back of that, I guess, what maybe gives the confidence as we think through ‘26 and ‘27 that you still feel good about that $1.8 billion number, especially as we think about lapping some of the price increases going into next year. Just yes -- how should we think through those factors and what gives you all the confidence behind those numbers?”

Eliran Glazer (CFO)

“So when we think about a few reasons why we are confident in next year numbers. First, demand and expansion from our larger customers. So we’re accelerating year-over-year growth for all upmarket customers. So you have seen $50,000, $100,000 and $500,000 I mentioned earlier. We’re accelerating year-over-year growth of RPOs and we are improving $50,000 NDR.

In addition to that, we have the multiproduct adoption that continues to trend positively. CRM is becoming very significant with more than $100 million in ARR and we are seeing customers increasingly adopting multiple solutions.

We are only in the early innings of customers that are adopting more than one product. And only, I think it was 6% that we said in Investor Day, and now we’re doing much, much better. AI Product engagement is accelerating. We were focused on educating, and we are focused on adoption of customers in the market, and we see a very healthy adoption of AI products that we believe are going to monetize next year in a more significant way. And as Eran mentioned, we have signs of stabilization in top of funnel at the end of the quarter that we are encouraged by that going into the fourth quarter and into next year.”

Here is another exchange. In this one, management also affirms their comfort with “consensus” revenue figures for 2026.

Aleksandr Zukin (Analyst)

“And then maybe Eliran, and just for you, if you’ve had a couple of questions regarding how you feel maybe about next year specifically. I think it’s just given some of the changing dynamics that you’re mentioning around channels and how you’re going to market and shifting spend what seemingly is a little bit of a change in terms of your guidance for Q4 versus previous years and periods in terms of passing through the beat and this timing adjustment. Maybe just help us pace how we should think about the growth? Are you comfortable with where consensus is for next year? Is it something where it may be a little bit more back-end loaded, and you could actually see acceleration in ‘27 because of some of these timing adjustments that you’re calling out, I think it would be really helpful for us to just understand the pacing of growth given some of these evolving dynamics.”

Eliran Glazer (CFO)

“Alex, Eliran here. So I think I mentioned it earlier, but if you think about where we’re going to be in fiscal year ‘27, we said that we are going to achieve $1.8 billion in revenue, we feel very confident with that. We feel very confident with the number based on everything that we see today. And this is something that we -- when we made the assumptions we took into account the trends that we see today we made some assumptions about the cross-sell motion, the new products that we are launching to the market. The fact that AI is going to be monetized to a certain extent, and the fact that we are going to expand within existing customer base.

So having all of this into -- taking all of this into account, we feel that the $1.8 billion in fiscal year ‘27 is achievable in the interim, we are confident with the consensus number for next year as well.”

With management being adamant that revenue growth remains on track and because I think this team is pretty honest (I’m naive like this), I do think they probably feel they are still on track (maybe they are lying to themselves). Nevertheless, in my model I assume a slight haircut to the $1.8bn for 2027.

One interesting trend to keep an eye on, which might be part of what management is seeing, is that as Monday goes more upmarket, new bookings will be more heavily weighted towards Q4. From Q3’25 earnings:

“As it relates to the quarter. All I would say is that when you move upmarket, obviously, you create somewhat of a hockey stick in the quarter and in the year, as we progress upmarket, we continue to see that phenomenon play out.”

So between (1) the fact that new bookings should get a natural boost in Q4, (2) Monday initiating its first set of bundled offerings in Q4, which should provide a boost to their cross sell efforts, and (3) stabilization in top of funnel search trends, I could see how management might believe that growth trends will firm up a bit.

It should be noted that while the slowing of top of funnel traffic is a real headwind, it’s basically unrelated to Monday’s strength as an offering. Some investors may be conflating the two, seeing slowing growth as an indication of a weakening value proposition perhaps. But I don’t think we should assume this is the case. If it’s not, Monday should be able to start “lapping” this top-of-funnel impact next year, which will alleviate a bit of the headwind.

Trading Comps

A set of “high-quality” SaaS comps currently trades around ~9-10x LTM revenue / ~12-13x LTM gross profit / ~30-40x LTM FCF.

Monday trades at ~5.4x LTM revenue, ~6.0x LTM GP, and 18x LTM FCF, well below this peer set despite a significantly higher growth profile.

Below is a chart showing software industry multiples over the last 10 years from Altimeter. The industry average multiple is currently ~4.8x fwd revenue, which is around the lowest its been over this 10 year stretch.

However, the dispersion between the highest multiples and the average multiple has grown significantly, as we can see above.

While the average multiple being on the low side might imply the industry overall is undervalued, growth is also the slowest its ever been. So the lower multiple would be justified, in theory, all else equal. We can see by this chart of EV / revenue / growth, that multiples are actually a bit elevated perhaps vs. historical figures, when taking growth into account.

Additionally, just about every public software company is not an “AI native.” They are the “incumbents,” and while some are considered AI beneficiaries, some are not. This is undoubtedly contributing to the dispersion in multiples.

Here’s what my “high-quality” software comp set looks like, charting (1) LTM GP multiple vs. Rule of 40 and (2) LTM GP multiple vs. revenue growth:

As we can see, Monday trades at a pretty extreme outlier of a multiple, especially when plotting multiple vs. growth. Based on the latter chart, Monday should trade in the vicinity of 3x+ its current multiple.

If we use Meritech’s regression analysis, plotting EV/GP vs. Meritech’s Rule of X (Meritech modifies rule of 40 by multiplying the growth component by 2 to give it extra weighting), we can see that Monday is even more of an outlier.

Based on metrics alone, we can make a solid case that Monday should trade 2x+ where it does today.

M&A Comps

Below is a set of software M&A comps.

The median multiple for this set is 8.3x LTM revenue and 12.5x LTM gross profit.

Charting GP multiple by Rule of 40, we see a similar correlation to our trading comps. According to this chart, Monday’s ~50% Rule of 40 would warrant a ~15x GP multiple (vs. ~15-17x according to the trading comps, so roughly in line).

If we chart GP multiple by revenue growth, Monday’s growth rate would warrant a ~13-15x GP multiple (vs. 20x+ according to the trading comps).

Where Should Monday Trade?

Going off metrics alone, Monday should trade at twice its current valuation — or at a mid-teens LTM GP multiple (vs. where it trades today around ~6x).

For the model I use a range of 4.0-6.5x NTM revenue at exit. Based on where multiples are today and based on the fact that I have Monday converging to a ~37% Rule of 40 in the outer years, this would represent a conservative range, in theory. We can see from the regression analysis below that the trendline for a ~37% Rule of 40 company equates to a ~6.5x NTM revenue multiple — which I use as the top of my range.

5-Year Model / Returns Analysis

For my simple model, I estimate growth using (1) estimated revenue growth from existing customers (i.e. NRR applied to prior year’s revenue) + (2) estimated revenue from new customers.

In the forward years, I assume:

• NRR steps down gradually from ~111% today to ~106% in 2031

• Revenue from new customers stays constant at ~$140m, a haircut to the $154m average of the 2023-2025e period (we can back into this historical average by removing the growth from existing customers implied by Monday’s NRR from overall incremental revenue)

• Non-GAAP EBIT margins climb to 26% by 2031 (vs. 14% in 2024)

If we assume Monday trades at 5.0x NTM revenue at the end of 2030 (or ~18x NTM FCF), that equates to a 2.2x MOIC and a 18% IRR:

Below is a sensitivity table showing theoretical IRRs at different entry prices and exit multiples.

Can Monday Generate a 106% NRR in 2031?

Let’s take a closer look at this assumption.

To start, it’s worth noting that Monday’s overall NRR is actually still technically growing (if we exclude 2021-2022) — from 100% in 2019 to 112% in 2024.

However, it seems NRRs are on the cusp of beginning their decline, which is inevitable.

Management provided this commentary around NRRs at their Sep-25 investor day:

“We believe we have a lot of room for upside with NDR, which means that in the next few years, we think that potentially NDR for all customers, maybe 110% to 115% and NDR for customers above $50,000 or the enterprise customers, as we call them, will be in the range of 115% to 120%.”

In other words, management expects that NRRs are likely to stay at current levels for “the next few years,” with perhaps even a little upside.

It’s worth noting that management guidance in past has been accurate with regards to NRR. At their 2023 investor day, in the midst of NRRs dropping rapidly from the 140-150% level for the $50K+ cohort, Monday stated that they believed NRRs would stabilize in the 110-120% range:

“We went to 140s, 150s. But then I think the normalized number that we are looking at is between 110 to 120. And we’re very consistent with this message…In addition to that, our gross retention profile remained very stable and our churn remains very stable, and it provides us with a lot of confidence that we are going to see this number in the range of 110 to 120 over the next few years. And potentially with everything that we are doing — the product, mondayDB, the add-ons, upsell, cross-sell — we’re also going to see some uptick in the future.”

This is exactly what happened, with NRRs for this cohort stabilizing at ~115% for 2023 before ticking back up a bit to 116% in 2024.

Furthermore, Monday benefits from certain factors that should help buoy its NRRs over the next few years (and beyond):

• Monday is like a tapeworm. It gets into an organization and then spreads. It’s not software where a bunch of licenses are sold upfront— it’s the opposite of that. This should support a naturally higher NRR than the average software business. Here’s a case study that management presented at its 2023 investor day.31

  

  While not necessarily the typical customer experience, this customer started with 5 seats and expanded steadily to 7.5k over 5 years. Below is another case study from their 2025 investor day. This customer started with 15 seats and expanded seat count to 9.6k over 4 years.32

  

• Huge TAM

• Upmarket opportunity. Monday has been successfully moving upmarket and this trend is still in the very early innings.

• Cross-sell opportunity. Monday can cross-sell CRM (+ Campaigns), Dev and Service. CRM has seen nice uptake. Service, while early, is showing promise. Monday just launched bundled offerings in Q4’25, which should be a tailwind

• AI monetization. While not something I’d necessarily bake into projections at this point, this provides additional upside (through consumption pricing or higher seat-based pricing).

Based on (1) the NRR guidance from their 2025 investor day and (2) the positive forces that should help sustain NRRs, I think my NRR assumptions are reasonable.

As a sanity check — my assumptions result in revenue of $1.77bn in 2027, which is slightly below the guidance management provided at Monday’s Sep-25 investor day:

Can Monday Generate a 26% Margin in 2031e?

My model calls for a 26% non-GAAP operating margin by 2031, up from 14% in 2024. Here is the model scenario for opex as a % of revenue:

And margins:

A 26% margin would be around average for the peer group:

I think this margin profile achievable:

• Monday has best-in-class GMs of ~90%. Only Asana and Autodesk have better gross margins in the comp set.

• Monday has a very efficient R&D organization. Monday spends just 22% of revenue on R&D.

• Margins really come down to S&M. Monday spends a lot on S&M, which clocks in at 55% of revenue for 2024. But a key distinction is that Monday is pretty efficient with its spend — it just spends a lot. Thus, as growth tapers, S&M should settle in a zone that provides for healthy overall margins.

Here again is Monday’s CAC vs. its horizontal peers for 2024: