The recent news that a Chinese-made immersion DUV lithography tool, linked to SiCarrier and its affiliate Yuliangsheng, has entered limited production has been framed by much of the financial press as a breakthrough capable of eroding ASML’s dominance and revenue base. The implicit assumption behind that framing is that price is the deciding factor: ASML’s machines cost tens of millions of dollars, a domestic Chinese alternative costs less and cheaper eventually wins. A closer, numbers-driven look at the economics of fab operations shows the opposite. Even in the extreme case where a Chinese DUV machine were handed to a fab manufacturer completely free of charge, a financially rational operator not swayed by political pressure would still choose to buy ASML’s machine at full price. The reason is that the purchase price of a lithography tool, however large it looks on an invoice, is a rounding error next to the revenue that tool’s performance determines over its operating life. Price is not the variable that decides this contest. Performance is, and the performance gap is large enough to make “free” irrelevant.

Start with a simple observation about how fabs actually make purchasing decisions. A lithography scanner is a capital expense, paid once. The wafers it produces are a revenue stream, collected every hour of every day for the machine’s entire operating life. Over that life, the two are not remotely comparable in scale, which is exactly why the acquisition price of the DUV machine (I’ll be using the NXT:1950i as the case study as it is the most relevant) , $30-45 million for the ASML machine or zero for the Chinese DUV is close to irrelevant to a financially rational buyer. Using representative figures for 28nm-class production, where wafer revenue runs around $3,000 per wafer, a machine running at 175 wafers per hour generates approximately $525,000 of theoretical hourly revenue. If an ASML tool operates at 97% uptime and a China DUV machine operates at 90% uptime, that 7-percentage-point gap alone costs approximately $36,750 per hour. Multiplied across 24 hours a day and 365 days a year, that single variable costs a fab approximately $322 million annually in missed revenue. That figure is roughly 8x larger than the entire $30-45 million ASML machine would have cost to buy outright. Put plainly, taking the free Chinese machine to avoid a $30-45 million purchase would cost the fab about $322 million a year in lost output. No fab manager who has done this arithmetic accepts a “free” machine on those terms given that the free machine is not a discount but rather a $322-million-a-year bill disguised as a gift.

Uptime is only part of the story and arguably not even the larger part. Yield, governed primarily by overlay accuracy, is the more consequential variable because it scales with every wafer produced rather than only with downtime hours. Overlay describes how precisely each lithography exposure aligns to the layers beneath it; lower overlay numbers mean higher precision and more functional dies per wafer. ASML’s currently shipped DUV tools for China, the TWINSCAN NXT:1980Fi, offer machine-matched overlay of 2.5nm. The domestic Chinese tool now entering production is architecturally comparable to the NXT:1950i, a design ASML first shipped around 2008 for 32/28nm-class single-exposure work. Reporting on advanced-node production in China using older-generation tools stretched through multi-patterning techniques indicates yields around 50%, compared to near 100% yields achieved by leading-edge fabs using modern single-exposure tools. A gap of that magnitude means a domestic machine, even if it were free and even if it somehow matched ASML on uptime, would still produce dramatically fewer sellable dies per wafer, a loss that compounds across every single wafer run, not just downtime windows. Combined, the uptime and yield gaps mean the true revenue penalty of operating a free domestic tool is considerably larger than the $322 million uptime estimate alone, which makes the “but it’s free” argument weaker still, not stronger. The zero purchase price buys the fab a bigger annual loss, not a smaller one.

This is also why the sticker price of a DUV machine is genuinely the least important variable in a fab’s decision. A lithography bay carries enormous fixed costs beyond the scanner itself which include cleanroom filtration and vibration isolation, metrology equipment, track coaters, etch and chemical-mechanical-polishing tools calibrated to that specific scanner’s alignment marks, and the trained engineering staff required to run it. That infrastructure is sunk regardless of which vendor’s machine occupies the bay, and it is only justified economically if the machine inside it maximizes revenue per square foot of cleanroom space per year which is a function of wafers/hour × yield% × uptime% × wafer price. Give a fab manager a free scanner with meaningfully worse overlay and reliability and the same expensive supporting infrastructure now produces less sellable output than it would with a $45 million ASML tool sitting in the same bay. The purchase price of the scanner itself is a small fraction of what that bay costs to run over its lifetime, so “free” does nothing to fix the underlying problem which is the fact that a free but underperforming machine still wastes the much larger, already-committed investment surrounding it. A rational fab manager evaluates the whole bay’s revenue per year, not the scanner’s price tag in isolation and by that measure the free machine is the more expensive choice.

A further factor, easy to miss because it lives in accounting conventions rather than headline specifications, is the gap between a lithography tool’s depreciation schedule and its actual operating life. Fabs typically write these machines off over a 5-year useful life for accounting purposes but ASML’s installed base routinely remains in active production for 15-20 years, with units from far older generations still running today decades past their book value. This matters in two ways. First, ASML’s 97% uptime figure comes from a track record proven across an operating horizon far longer than any domestic Chinese tool has existed (the domestic effort is only now entering limited production, with roughly 5 units expected this year). A fab taking the domestic alternative, even for free, is betting on 15-20 years of unproven reliability instead of a demonstrated multi-decade record and a reliability or overlay-drift problem discovered in year 8 or 10 is far more costly than it would appear under a 5-year depreciation lens. Second, this reframes the price comparison itself: if a 5-year accounting life dramatically understates a real 15-20 year service life, the effective annualized cost of the “expensive” ASML machine amortizes the $45 million down to roughly $9 million per year which is trivial next to the $322 million annual uptime penalty and the additional yield losses stacked on top. Viewed this way, ASML’s machine was never really the expensive option. Spread over its true operating life, its annualized cost is a tiny fraction of the annual revenue penalty the free Chinese machine would impose. The fab isn’t choosing between cheap and expensive but rather it is choosing between a $9 million a year tool and a $322 million a year loss revenue.

A further point often lost in coverage of this development is that the domestic machine entering production is not actually competing against ASML’s current product line at all because ASML does not sell that class of machine to China anymore. The domestic tool is being compared, implicitly, to a 1950i-class design from around 2008. ASML’s actual recent shipments to China have consisted of newer-generation immersion DUV tools such as the NXT:1980Fi (330 WPH, 38nm resolution, 2.5nm overlay), offering meaningfully higher throughput and tighter overlay than the 175 WPH, 2.5nm-overlay, 2008-vintage design the Chinese tool resembles. In other words, China’s new machine is not displacing a sale ASML would otherwise have made, because ASML was never going to sell a 2008-generation tool into that market in the current period regardless of whether a domestic alternative existed. The domestic tool is substituting for legacy capacity ASML has already been regulated out of supply, not winning business away from the machines ASML is actively selling.

The pace and scale of actual adoption so far leaves more to be desired given that reporting indicates the domestic tool is expected to ship in roughly 5 units this year, scaling to around 20 units next year with initial deliveries going to SMIC, Hua Hong and CXMT primarily for production-line validation rather than volume manufacturing. A separate domestic effort from Shanghai Micro Electronics Equipment has reportedly sold around 10 units of a comparable series to date. These are pilot scale deployments measured in single digit and low double digit unit counts, not a fleet level replacement of ASML’s installed base, which numbers in the thousands of systems globally. Industry analysis of similar transitions notes that moving a new lithography platform from early production into high volume fab use typically takes months to years, because uptime and yield can only be proven under sustained real world manufacturing conditions. Some reporting also indicates domestic tools still depend on imported components, including parts sourced from Japan, meaning full supply chain independence has not yet been achieved either.

Taken together, the numbers converge on a single, deliberately extreme conclusion: even setting the price of the Chinese machine to zero does not make it the rational choice over ASML’s machine at full price. An approximate $322 million/year in lost revenue as an uptime penalty alone dwarfs the entire $30-45 million purchase price of the ASML machine it would replace, meaning the “savings” from taking the free machine are wiped out roughly eight times over in the first year of operation alone. Layer on a likely larger yield-related loss from materially worse overlay (2.5nm on a 2008-era-comparable design, versus 1.4nm on ASML’s current lineup) and the case against the free machine only strengthens. A 5-year depreciation schedule that masks a real 15-20 year service life shrinks the effective annualized cost of the “expensive” ASML machine to roughly $9 million a year. The domestic tool is being measured against a machine class (NXT:1950i-era) that ASML no longer even sells to China, with shipment volumes of roughly 5 units in 2026 scaling to 20 in 2027 against an ASML installed base numbering in the thousands.

Every one of these facts points in the same direction. A rational fab operator, free from political pressure, does not compare a $30-45 million ASML machine against a free Chinese machine and conclude that free is the better deal. It compares an approximately $9 million annualized ASML cost against a $322-million-plus annual revenue penalty and concludes, correctly, that the free machine is by far the more expensive option on the table. The domestic Chinese DUV tool is a meaningful supply chain resilience milestone for China. But it is not an economically rational substitute for ASML’s machines at any price point, including zero, and it does not represent a material near-term threat to ASML’s revenue.

To conclude, these numbers point to a clear set of conditions China’s domestic lithography industry would need to satisfy before it could genuinely displace ASML in the Chinese market rather than merely supplement it at the margins.

First, the domestic tool would need to close the reliability gap entirely, moving from today’s roughly 90% uptime to something in the 95-96% range that ASML tools already deliver.

Secondly, it would need to match the yield fabs currently extract from ASML machines at the 28nm node yields that sit close to 100% for leading fabs like TSMC which is not the product of a single engineering breakthrough but of decades of accumulated trial and error over decades formatting the fab layout and process so that it can achieve the highest possible yield. If this new machine were to be implemented, the whole fab layout would have to be rerouted to suit the new machine which will incur unnecessary costs.

Third, even after closing both gaps, the domestic tool would still need to be priced significantly below ASML’s machines by enough to entice fab operators to walk away from a platform with roughly 30 years of DUV history and an unmatched track record of compounding reliability, process stability and yield performance. This is a switching decision few rational operators make lightly even when the economics look close, let alone when a new, unproven entrant is asking them to bet a fab’s output on it.

Each of these three conditions is individually difficult. Achieving all three simultaneously:matching uptime, matching yield built on decades of accumulated learning and still undercutting ASML on price by enough to justify the switching risk would be a genuinely remarkable feat and one with no real precedent in the history of the lithography industry. Given the scale of the gap on each dimension individually and the fact that all three must be closed at once rather than one at a time, it would be a considerable surprise if China’s domestic DUV effort managed to clear all three bars in the foreseeable future. Until it does, the domestic tool remains a supply-chain resilience achievement rather than a genuine competitive threat to ASML’s position in China.

Macro

So the markets this week saw a move out of hyperscalers into semiconductor companies, following Google’s earnings. As usual, we will cover some Macro, then zoom into Micro.

For Macro-economic data, nothing much this week, Manufacturing and Services PMI came in slightly higher than expected, which are coincident indicators of strength in the economy. The main “Macro” story this week is probably still the war.

Oil prices have surged significantly over the week, with crude oil creeping up to $90/Bbl again. US and Iran, at least for now, do not look like they want to resolve the war, with Iran holding the leverage and US playing the “showman game”. Trump wants to look victorious, but he will end up having to give way to the Iranians.

What this means is that inflation expectations have gone up again, with yields spiking over the week, back to levels seen during the Trump tariff crisis. Yield spikes are generally not great, it means higher discount rates for equities and it pressures asset prices. That’s partly why the markets tanked during the week. So until the war is resolved, there will still be weight in the markets pushing it down.

On the hyperscalers front, slight macro again since they affect so much in the markets, Credit Default Swap spreads for these Hyperscalers have been creeping up steadily over the week, and with some of them turning FCF negative, there is expectation for this to continue. CDS spread increases means higher risk, and higher future borrowing cost for these hyperscalers.

Oracle, for instance, has seen its its rating down to BBB-. So one more downgrade and it is in the junk/high yield territory, and it would make up roughly 7% of the entire high yield market. And there will be forced selling as well if they fall out of IG status. Then there’s their equity price that could keep falling with higher refinancing cost and lower confidence, the deals with OpenAI that they have signed.. Sounds like a doomsday event, but the markets have always collapsed when (1) excessive leverage is taken and when (2) there is no more investor willing to provide liquidity.

So all in all, bet on a hyperscaler that has a defensible balance sheet. Hopefully this doomsday event doesn’t play out, we shall see.

Earnings this week

Defense companies:

Northrop Grumman and Lockheed Martin reported, both of which beat earnings and raised guidance. It's clear that the war has been accretive to these companies, with backlog increasing significantly from the recent quarter. Expect the same for the other defense companies reporting soon. These are high quality companies with a (1) steady backlog and (2) strong recurring revenues. The House also just passed a $60 billion Iran war budget which will directly flow to these companies.

GE Vernova:

GEV is one of the strongest AI related companies in the space, with Gas Turbines for utility companies that have a bottleneck up till 2030. EPS was a whooping 30% growth, with revenues up 22% from last year. Orders have also increased by 88%, with the company increasing EPS and revenue guidance, though its a slight miss from consensus.

The business is great, with only 3 companies that can produce these Gas Turbines for data centre utilisiation: GEV, Mistubishi, and Siemens, but market valuations are too high for any meaningful consideration.

Moody’s:

Moody’s reported earnings, a strong beat on both the top line and bottom line driven by increased financing from data centre buildup (which we covered in the SPGI short writeup). EPS was up 30+%, though a portion of it was attributed by non-operating income.

The company has fallen this year due to fears associated with the Saaspocalyse, but I view this business as a strong duopoly with S&P, with a ratings business that is structurally durable.

Google:

Arguably the biggest earnings this week. Google reported revenue up 24% y/y with Google cloud revenue up 83%. Operating income was up 30% y/y and the company has seen accelerated growth on all fronts, with search revenue up 17% y/y. The business is strong, doing well, but the streets were hyper focused on two things: (1) FCF is now negative for the business (2) The company has raised Capex guidance again, up to 195B this year, with the buy side expecting a 325B print for next year’s Capex.

Google is relatively covered for this year as well as next year’s capex, with 240B sitting in its balance sheet and additional offerings from its recent equity shelf program. The markets, rightfully or wrongfully, are fearful of reduction in ROI on Capex spend, with these increased Capex number.

There are a few tension points here: (1) Capex is cyclical, harvesting phase comes after the initial Capex ramp (2) Proper assessment should be made on the company’s ROI on the capex spent, rather than just the headline number (3) The beats on the Capex guidance have slowed, which in itself might be a positive.

ROIIC is still strong, at 17%, which is above their WACC of about 12.8%, even with these strong capex numbers. Backlog has grown significantly for the cloud, majority of which is incremental revenue for the company, with some parts serving as renewals. The company mentioned that 50% of the backlog will flow to the topline over the next 2 years, which is about 250B. At its current run rate of 99B, it is a material step up.

So there’s a company here that is still earning strong returns on capital, that is growing rapidly and will potentially displace Azure in second place, that has a fully vertical AI stack and the distribution capability to do it. I mean, the company is simply firing on all cylinders, except one. Model intelligence is subpar now. The company recently released Flash 3.6, which came out to be below Meta’s new model. But does the company really need higher intelligence if it is able to (1) push for lower token prices with internal chips (2) have this massive market share lead ahead of peers?

So the key metric to watch over the next few quarters is their margin profile. Management mentioned that margins will be suppressed due to 3P compute usage in the coming quarters, so we’ll have to track those once they normalise. Should suppression of margins come in, and if model intelligence is weak, then Google has probably lost its edge. Otherwise, the company is doing well.

The company mentioned during the earnings call that they will continue pushing for frontier models, that means, continue with their push towards greater intelligence, towards Artificial General Intelligence. We shall see how this plays out, but this could be another narrative driving the stock price down in the near term.

All in all, for Google, the company is now trading at 20x Forward EV/EBIT, around its average over the past 10 years, down significantly from its recent high. While not particularly cheap, I don’t think the stock is expensive here after this rerating.

But I would caution overconcentrating in these bets, as the market continues these hyperscalers as”losers” and semiconductor as “winners”. The tune of the market has been like this over the past 7 months, so play along with it perhaps? Anyways, may the markets be with you.

ASML reported a strong beat across the board, with CEO Christophe Fouquet and CFO Roger Dassen delivering bullish commentary on every segment. Once again, ASML has exceeded my expectations and updates will have to be made to my model for the company.

A bit of context for my position in ASML. Full disclosure, I had intially trimmed my stake in ASMl at around 1600 given that my model (with my prior expectations for the company), gave it as the bull case and I saw limited upside. I trimmed my stake in ASML as a % of the overall portfolio from 15% to 13%. Currently, I hold ASML in 2 of my 3 portfolios with the concentration of the holdings as shown in the following snapshots.

Readers of the substack are well aware that both guanrui and my entry prices in the company are in the mid 600s and we have seen significant appreciation in both of our holdings of the company. However, we had differing views on the outlook of the business. Despite my bullish assumptions, ASML has managed to surpass my expectations in their recent earnings call.

Without further ado, lets dive into the details of the most recent Q2 earnings call. I will provide some back of the envelop valuation for the company towards the end of the writeup which should be taken at face value given that it is only an approximate. Do note that all the valuations provided will be in USD as I am invested in the US-listed ticker.

ASML Q2 actuals vs. guidance:

The installed base overrun was the main driver of the margin beat. Dassen noted that upgrade demand is increasingly software-led, which means customers can boost productivity with minimal machine downtime which is exactly what fabs want when they’re capacity-constrained.

What Management Said

1. Demand Has Re-Accelerated and Customers Are Committing Further Out

Fouquet described a demand environment where customers are not just investing more but accelerating their plans ahead of schedule. The reason: their own end customers (hyperscalers, AI chipmakers) are giving them longer-term purchase commitments, giving fabs the confidence to pull forward capex. This is a meaningful qualitative shift proving that it’s not just a cyclical bump but customers structurally extending their planning horizons.

2. Logic

ASML is seeing simultaneous capacity additions at 5nm, 4nm, 3nm, and a “as aggressive as possible” ramp of 2nm. Customers are already beginning to look at 1.4nm. Advanced foundry/logic revenue is expected to grow ~25% in 2026. This is not a one-node story it’s broad-based and AI-driven at multiple levels.

3. Memory

Memory revenue is expected to grow ~75% in 2026, driven by two forces: HBM (high-bandwidth memory for AI) pushing customers to build out aggressively, and advanced memory nodes requiring higher lithography intensity (more EUV and immersion layers per wafer). Memory customers are essentially sold out and are telling ASML they need as much capacity as they can get.

4. Installed Base

The installed base business (upgrades, services, spare parts) reached €2.8B in Q2 30%+ growth expected for the full year. Software-led upgrades that boost existing tool throughput are particularly attractive: high margin for ASML, minimal disruption for customers. As the EUV install base compounds, this line becomes increasingly powerful.

5. System Volumes

Management was unusually specific on capacity expansion which is surprising given that they love to sandbag guidance (lol). Fouquet stated that both EUV and DUV immersion capacity will increase by approximately 30% in 2027 versus 2026 and another 30% in 2028 versus 2027. That compounds to roughly double the 2026 shipment volumes by 2028.

Crucially, this is supply-led capacity expansion, not just demand speculation. ASML is actively adding manufacturing capability. Fouquet confirmed ASML is “close to receiving all the orders it needs” for 2027 even at the expanded level and for 2028 already has “a large number” of EUV orders in hand. The supply expansion is being built to meet orders that already exist and not speculative capacity.

6. High NA

Intel is now using High NA EUV in production on its most advanced products. Fouquet called this “proof of the maturity of the tool.” High NA enables going from 3 masks to 1 and reduces process steps from ~100 to ~10 which is transformational for fab economics. As more customers adopt High NA, it becomes an additional revenue layer on top of the Low NA and DUV base.

7. China

China remains ~20% of total revenue. Importantly, Dassen noted this 20% now applies to a much higher revenue base (€43–45B vs. prior €36–40B guidance), so the absolute dollar contribution is actually higher. Current China demand is primarily domestic logic, not advanced nodes subject to export restrictions. Management stated the current export control environment is accommodated within guidance.

Full-Year 2026 Guidance

The midpoint of the new 2026 guidance (€44B) is 15% above the midpoint of the prior guidance (€38B).

Q3 2026 guidance: €11–12B revenue, 55–57% gross margin, €2.9B installed base. The gross margin guidance of 55–57% implies continued expansion and suggests H2 2026 will be materially stronger than H1.

EPS Model: 2025A → 2028E

The 30% compounding capacity expansion management explicitly guided for is the critical driver of the 2027 and 2028 revenue build. EUV going from 65 to 85 to 110 units, and DUV immersion from 130 to 170 to 220 units, creates a step-change in shipment volumes that combined with a growing Installed Base and rising ASPs from High NA mix pushes revenues materially higher.

Revenue Build

P&L Model

2026E:

• Revenue: €44B Gross margin: 55% → GP = €24.2B

• R&D €5.0B + SG&A €1.6B

• EBIT: €17.6B Net income: €14.7B Shares: 390M

• 2026E EPS: €37.7

2027E:

• Revenue: €57B Gross margin: 56.5% (High NA mix + scale) → GP = €32.2B

• R&D €5.8B + SG&A €1.7B

• EBIT: €24.7B Net income: €20.5B Shares: 385M

• 2027E EPS: €53.2

2028E:

• Revenue: €71B Gross margin: 58% (continued High NA ramp + absorption) → GP = €41.2B

• R&D: €6.8B + SG&A €2.0B

• EBIT: €32.4B Net income: €26.9B Shares: 380M

• 2028E EPS: €70.8

EPS Growth Summary

2-year CAGR (2026→2028): 37%

Forward P/E Analysis

At the current price of $1,900:

ASML has historically traded at 30–40x forward earnings during growth periods. On the base case which fully reflects the 30%/30% capacity expansion guidance, the stock trades at just 24.8x 2028 earnings, which is cheap for a monopoly rising a secular tailwind. The pre-Q2 consensus at 29x 2028 will likely be revised upward as analysts digest the capacity expansion guidance, which could drive further multiple re-rating alongside earnings upgrades.

Expected Returns to End-2028 (~2.5 Years)

Bull Case EPS: $82, Exit P/E 40x
30% capacity expansions fully execute, High NA accelerates to multi-customer production, no export escalation. Revenue approaches €75B+.
2028 Price Target: $3,280 Total Return: +73% +24% annualized

Base Case: EPS $76, Exit P/E 35x
30% capacity ramps execute as guided, modest margin expansion, stable export regime. Revenue ~€71B.
2028 Price Target: $2,660 Total Return: +40% +14% annualized

Bear Case: EPS $55, Exit P/E 30x
Export controls escalate and restrict DUV to China; capacity ramp faces supply chain delays; semi cycle softens in H2 2027.
2028 Price Target: $1,650 Total Return: -13% -6% annualized

Key Risks

Export controls remain the primary binary risk. China is 20% of revenue. Any escalation in DUV immersion restrictions would directly impact €8–9B of annual revenue. Management says current guidance accommodates the ongoing discussions, but the situation is politically driven and inherently unpredictable.

Valuation leaves little room for disappointment and multiple contraction will eat into annualised returns

FX exposure: ASML reports in EUR; USD investors take EUR/USD risk on every line item.

High NA ramp risk. The bull case partly depends on High NA scaling into multi-customer production by 2028. Intel’s production use is an important data point, but broad adoption is still early albeit the current AI wave will boost adoption.

Meta’s Cloud Business and the Semis Sell Off

The markets are finally broadening out, away from semis, into lagging sectors, given most of the S&P’s returns this year have been contributed by semis.

The trigger for this semis sell off was the Meta news. Meta’s decision to sell their “excess compute” led to narratives of supposed excess capacity and overbuilding, seemingly planting the seeds for lowering of Capex guidance. Zuck is sending conflicting signals and playing both sides. On Thursday, he mentioned that Meta is currently not optimising for AI, and that they will likely expect to see benefits in the next 4-6 months.

On Meta’s decision to become a neocloud, it’s a slight positive for the company. The issue with the stock is that the markets don’t understand the need for the company to spend all this Capex on its advertising arm, and the whole debate has been around the Capex. Revenue was up 30+% in their recent earnings, yet the stock fell.

In the case of Meta, their Capex has been used to drive automated ad creatives, generations, recommendations, etc, which don’t require a high end model. Pivoting away towards a cloud business protects profitability, potentially, but the next question should be on whether or not the company needed all that Capex in the first place?

As for the other hyperscalers, it’s a slight negative with more competition, though there is a balance to this as models become more efficient and token usage falls. Tokenmaxing is a term used to describe companies that have not optimised for token usage. A lot of companies haven’t optimised their token spending. So, as companies become more sensitive to token pricing, as token pricing falls, the use of compute increases, benefiting these hyperscalers.

So, in the case of Meta, I am confused by Zuck, and I’m sure the market is as well with the stock running on Wednesday, and fell on Thursday.

Jobs market

Weak jobs data this week suggests a potentially slowing labour market, supported the rate cuts narrative.

My feeling is that many individuals underestimate the ability of AI, and that most small scale enterprises are still not actively using it yet. There are plenty of entry level jobs that will be replaced by AI, those tied to data entry. Business owners are now reducing the need for staff that provide low value work, such as design, data entry, documentation. Perhaps entry level investment bankers in the future will also slowly be phased out.

And so for companies facing disruption risks, e.g. Adobe, Intuit, and other software companies, while valuations are cheap, I am leaning towards avoiding that whole bucket. Non-essential or non-regulated tasks will eventually be replaced by AI, with work that can be done in minutes, without the need for an actual human.

Ideas this week, and some of my moves

In the short run, markets are focused on a few key debates. In the long run, it is always the ROIC of the company that leads to excessive returns, that means, the company’s ability to maintain its high market share, in a high growth market, with high barriers, for a sustained period of time. The key is to find an “amalgamation” of both.

In the case of these 2 companies here that I’ve allocated capital into again, there are both short term, and long term narratives driving the stock. Ideas here are kept short, and actionable, though no specific price targets or time horizons are included, as I view them as long term opportunities.

1. Honeywell

Honeywell, current EV/EBIT of 19.4x is trailing RTX (23x), Safran (22x), TDG (30x) despite strong aftermarket exposure (~cumulative 60+%).

Aerospace industry is currently in a prolonged recovery cycle, with pre-covid recovery that only recently occurred in 2024. Delivery from OEMs are still constrained, which is bullish for OE providers with repairs, because planes fly longer and are older. Airlines have refuted against retiring fleets early, scared that they might make the same mistake as during covid.

Commercial flights have since recovered post Iran War. GE Aerospace previously projected for flights to be flat to LSD growth this year. Data suggests that flight growth recovery has already began, with recent commercial flights surpassing that of last year. Barring anymore shocks, the trend should continue.

Key debate points for the company in the coming quarters:

(1) Airbus and Boeing production volumes: Production volumes has a slight negative effect on OEM providers with pricing, including Honeywell. Deliveries have remained constrained which lowered retirement rates and increased ages for fleets. Delivery this year still lags that of the last year.
(2) Defense spending and confirmation: Trump’s proposed 1.5T defense spending primarily on replenishing Iran War munition
(3) Aftermarket, flight hour growth: Management recent projection of 6-8% sales growth suggests conservatism, with commercial flight hour growth already projected to be at 4%, and pricing likely to be above that of inflation. Adding on RMU’s strong growth (18% FY25, +DD FY26) suggests management is setting a low bar for a beat
(4) Standalone costs, too high? too low?
(5) Capital allocation post spin off: Management has pledged 1. dividends 2. strategic acquisitions 3. investment in R&D for growth. Wait for consensus to determine conservatism, if any

Potential Catalysts:
(1) Inclusion in the S&P 500 will drive further institutional buying
(2) Earnings report that beats their conservative projections
(3) Analyst coverage

Risks:
(1) Economic recession
(2) War starts again.

2. S&P Global

With the capex theme, hyperscalers will need to raise more debt to fund capex. The likely beneficiary is SPGI. While rates have remained high, capital markets have remained resilient.

Key debate points for the company in the coming quarters:

(1) AI Risks: Ratings + Platts and Indices make up about 70-80% of company EBIT. So 20% is tied to Market Intelligence, of which the CEO thinks that only 5% is subjected to AI threat.

(2) Software moat weakening: The carving out their mobility business to further streamline business quality weakens this argument.

(3) Rates: Maintain rates, or hike rates? Economic data suggests both are almost equally probabilistic. Rate hikes will weaken SPGI’s economic outlook.

(4) Is the company really 7x P/E less worthy than Moody’s?

(5) Can margins continue to expand? Are they using AI to improve efficiency? Or are they at risk of AI?

Potential Catalysts:
(1) Spin-off of mobility business (Already done)
(2) Earnings reporting reduced software contributed EBIT, and raising of guidance tied to AI debt.

Risks:
(1) Economic recession, rate hikes

Recently, I decided to sell 70% of my shares in Meta and instead reinvest it back into Google. My initial rationale for trimming Google was due to several reasons that can be found in the Substack writeup titled “Logging My Rationale For Trimming Alphabet.”

In that article, I stated my worries of Google being unable to increase their advertisement prices to cover the increasing costs of search queries via the LLM, which was approximated to be 10X the cost of delivering the traditional search query. However, recent updates at the investors’ presentation, as well as Alphabet’s Q1 earnings call, has alleviated my worries, as Alphabet has stated that their cost of meeting the search queries has actually dropped by over 30% thanks to their increased utilisation of their TPUs. Monetisation of their ads shown on LLM are as effective as their original search queries and asking around my fellow SMMA owners and my ex-clients also fits the narrative that google ads have been working as well-if not better than before as Google ROAS still remains higher then Meta and the gap is widening.

Furthermore, Google’s ability to be the lowest-cost provider of computing will likely be entrenched, given that the company has had a tremendous headstart in being able to design their own TPUs, which puts the company at least a decade ahead of their peers. Google can be considered the grandfather of LLMs, given that it was their technology that enabled the development of LLMs. In a paper titled “Attention Is All You Need,” Google introduced the transformer architecture to the world that enabled developers to move away from sequential processing (like RNNs and LSTMs) to a fully parallelizable approach, which is the foundational breakthrough enabling modern Large Language Models.

If Meta were to get into the cloud computing industry due to oversupply and crush margins in the industry, google will likely be able to tide through given that it is highly vertically integrated.

The core reason why Google will be able to be the lowest-cost producer in the compute industry is due to two simple reasons:

Hardware Specialization and Memory Optimization

The core of Google’s cost-efficiency lies in the evolution of their Tensor Processing Units, such as the specialized TPU 8i designed specifically for inference. Unlike general-purpose GPUs, these chips are purpose-built to handle the Transformer architecture by maximizing Matrix Multiplication Unit utilization. To tackle the memory bottleneck that plagues LLM inference, where the system is often stalled waiting for model weights to be fetched from high-bandwidth memory, Google has dramatically increased on-chip Static RAM. By keeping a larger portion of the Key-Value cache in this fast, on-chip memory, the TPU reduces the frequency of trips to slower high-bandwidth memory, significantly lowering latency and power consumption per token. Furthermore, architectural advancements like the Boardfly networking topology have flattened the communication structure between chips, reducing the latency overhead of multi-chip coordination in large model serving.

Software Stack and Algorithmic Efficiency

Beyond the hardware, Google leverages a tightly integrated software stack, including the XLA compiler, which generates highly optimized machine code specifically for the TPU architecture. This allows for advanced techniques like paged attention, which dynamically manages memory to prevent wastage, and speculative decoding, which uses a smaller, faster draft model to predict tokens that the larger target model then verifies in parallel. By optimizing the full stack from the JAX-based modeling libraries to the physical data center fabric, Google can maintain higher productive compute time and achieve better performance-per-dollar. These efficiencies allow them to pack more inference requests into the same infrastructure, fundamentally lowering the cost to serve each individual token while simultaneously improving response times for the end user.

Google’s competitive advantage stems from owning the full stack that is vertically integrated, and currently, no other company has a fully integrated stack (other than possibly Amazon). Their custom Inter-Chip Interconnect and optical circuit switching allow them to scale thousands of chips into TPU Pods that function as a single, massive computer. This level of interconnect efficiency is difficult for standard off-the-shelf ASICs or GPU clusters to match without significant specialized networking overhead. This cost advantage can be clearly seen in controlled testing for specific LLM inference tasks, where Google has reported significant gains, often cited as 2x to 4x better performance-per-dollar compared to general-purpose cloud inference offerings.

While technical proficiency is essential for navigating the current landscape, focusing exclusively on Google’s specific hardware or software architecture may be a short-sighted strategy. Because AI hardware cycles are incredibly rapid, rendering today’s state-of-the-art accelerators obsolete within a few years, investing too much cognitive capital into the mechanics of current TPUs offers diminishing returns and provides little ability to understand whether Google will succeed in the future. Instead, I believe that a more robust approach is to evaluate the underlying organizational culture. Based on my previous posts, I am sure that readers are aware of my admiration for Sundar and his team, and how they have been able to spot opportunities decades before others caught up to them and invested heavily in those ventures. By serving as an early, pivotal investor in visionaries like SpaceX, Anthropic, and DeepMind, the company has consistently placed capital behind the future’s most critical infrastructure, whether in aerospace, foundational AI safety, or general intelligence. This pattern of high-conviction and successful investing in different areas clearly shows that Google’s team has a knack for capital allocation and a disciplined approach to spend their bets evenly, as compared to Zuck’s all-in approach to AI, currently without a track record of good capital allocation (we all remember the Metaverse).

Google also has a tremendous culture that has benefited them incredibly. In 2013, Demis Hassabis and his co-founders were looking for a partner with the computational scale necessary to achieve their ultimate mission, which is to build Artificial General Intelligence. Both Google and Facebook (now Meta) emerged as major suitors.

Mark Zuckerberg was eager to acquire DeepMind, viewing it as a powerful asset for his growing empire. Reports indicate that Meta actually offered a higher financial valuation for the company, hoping to lure the founders with enormous signing bonuses and immediate capital. (Sounds familiar? History does not repeat, but it sure does rhyme.) During negotiations, Demis Hassabis famously tested the two leadership teams. When he met with Zuckerberg, he steered the conversation toward the long-term potential of AGI. However, he noted that Zuckerberg seemed equally excited about a wide range of disparate technologies and treated AGI as just one item on a list which included VR and 3D printing, whereas Larry Page demonstrated a deeper understanding of the foundational importance of AI and granted them the autonomy to pursue high-risk, long-cycle scientific research (which later culminated in AlphaFold, that enabled Demis to be awarded the Nobel Prize), whereas Zuckerberg refused to fund technology that could not be commercialised. Despite the lower financial offer from Google, Hassabis chose them and prioritized a partner whose vision for the scale and impact of AI matched his own. Looking at Zuck’s recent acquisition of “talent” via huge signing bonuses, as well as large and costly acquisitions of nascent AI founders that could be lured away by another large paycheck, shows how important culture is in an organisation. Google’s enduring competitive advantage lies in its cultural commitment to a purpose that transcends short-term economic gain. The top-tier researchers who are currently pushing the boundaries of AI and fundamental science are rarely motivated solely by financial incentives and are instead driven by the pursuit of discovery and the ambition led by a visionary leader like Demis Hassabis. This mission-centric ethos defines the core of Google’s workforce, setting a precedent that is ultimately guided by leadership whose life’s work is rooted in a vision for societal progress rather than mere capital accumulation. Demis will undoubtedly be one of the key players that will lead the future of AI.

In the midst of a massive technological paradigm shift, it is a company’s ability to foster radical innovation, maintain high-velocity engineering, and pivot its internal infrastructure that dictates long-term success. Technology will inevitably be commoditized or eclipsed by the next innovation, but a culture designed to thrive in uncertainty and adapt to shifting requirements remains the only sustainable competitive advantage, and that is what Google clearly has.

The other reason for increasing my position in Alphabet is the company’s strategic exposure to industry-leading private enterprises, specifically SpaceX and Anthropic. While I am generally hesitant to invest directly in private companies due to their elevated valuations, I am comfortable gaining exposure to these assets through an exceptionally well-run entity like Alphabet.

Meta conversely lacks these strategic stakes. It is worth considering the immense value creation that might have occurred had Mark Zuckerberg allocated capital toward such high-growth external ventures rather than focusing exclusively on the internal development of the Metaverse.

As of early 2026, Alphabet maintains an ownership stake of approximately 6.1% in SpaceX, following dilution from its merger with xAI. With SpaceX targeting an IPO valuation between $1.75 trillion and $2 trillion, Alphabet’s 6.1% stake is estimated to be worth roughly $106.75 billion. Furthermore, Alphabet holds an estimated 14% ownership stake in Anthropic. Following Anthropic’s recent Series H funding round, which valued the company at $965 billion, Alphabet’s 14% stake is valued at approximately $140 billion. Combined, these two positions represent roughly $246.75 billion of Alphabet’s market capitalization.

When factoring in Waymo, which was recently valued at $126 billion, alongside Alphabet’s interests in Stripe and various other private ventures/subsidaries (most crucially Demis’s Isomorphic Labs) that have yet to go public, the total value of the company’s non-core stakes could be worth as much as $450 billion. This valuation still excludes the upside of Google’s quantum computing research, the clear leader in the field. By maintaining such a diverse portfolio of foundational technology, Alphabet offers a level of strategic optionality that distinguishes it from Meta and provides a unique right-tail upside potential in nascent industries that no other company is able to provide, while its core business remains highly profitable and firing at all cylinders.

Alphabet represents a collection of significant call options on industries that remain largely undefined given that both the nature of these sectors and their long term economics are currently beyond our foresight. As investors, we do not need to possess the expertise to navigate these emerging fields, and we can instead rely on management to allocate capital effectively given their superior circle of competence. This strategy allows shareholders to benefit from the massive cash flows of the core advertising business, which provides the necessary fuel for Alphabet to fund potential breakthroughs in technology and beyond. Provided that leadership maintains a disciplined approach to capital expenditure and continues to prune underperforming projects, the company remains a powerful vehicle for capturing value in unknown future markets. By investing in Alphabet, I am able to harness the judgment of Sundar and team, who are uniquely positioned to identify and invest in significant opportunities.

Business Overview

Cellnex is Europe’s largest telecommunications towers and infrastructures operator, enabling operators to access a wide network of telecommunications infrastructures on a shared-use basis, and thus helping to reduce access barriers and to improve services in the most remote areas.

The Company manages a portfolio of more than 130,000 sites, including forecast roll-outs up to 2030, in 10 European countries, with a significant footprint in Spain, France, the United Kingdom, Italy and Poland. Cellnex holds a dominant market share in the countries that they operate in and is the market leader in 7/10 countries that they operate in and ranks the top 3 in market share for all 10 countries.

The company operates multiple business segments including the following:

Towers (81% of Revenue): Passive infrastructure services for MNOs(mobile network operators) and other wireless operators.

DAS, Small Cells & RAN-as-a-Service (7% of Revenue): Solutions for high-density urban areas and indoor coverage (stadiums, tunnels), including active equipment management in Poland.

Fiber, Connectivity & Housing (6% of Revenue): Data transport (FTTT), backhaul, and edge data center hosting to support 5G capacity.

Broadcast (7% of Revenue): Digital TV and radio broadcasting services, primarily in Spain.

Understanding Tower Operators and why the business exists?

Tower operators are responsible for building as well as maintaining telecom towers that are used everyday by millions of users. The main customers of tower operators are the telco companies.

In the markets that cellnex operators, their main customers are as follows:

France: Bouygues, Iliad, Orange, and SFR.

Italy: Iliad, Fastweb, Vodafone, Linkem, TIM, and WindTre.

United Kingdom: Virgin Media O2 (VMO2) and Vodafone.

Spain: Telefónica

Poland: Polsat

Netherlands: Odido

Portugal: Digi

Due to the nature of the industry (slow growth rate, high initial capital required to enter the industry), the competition of the telco industry is limited and what is occurring currently in the EU is the consolidation of the telco companies into a single company.

To explain why cellnex’s business model exists, we have to first understand more about how the telco companies used to operate in the past.

Telco in the 2000s was a booming industry as we were at the beginning of an internet revolution. Mobile penetration was low and the world was largely analog. With the advent of the internet and the PC, the telco industry became a fast growing industry that attracted many competitors that laid out cables and towers in order to “meet” the forecasted demand. As we all know, that demand did not materialise. The high capital expenditure nature of their business models lead to numerous telco companies going bust when revenues did not materialize. Over the last 10 years, a new trend has emerged. Instead of the usual capital expenditure heavy business models, telco companies seek to revamp their business models into capital asset light models and offloaded the building of the towers and cable to towercos such as cellnex. Building and maintaining physical towers is extremely capital-intensive (CapEx). Telcos are increasingly under pressure to invest billions in spectrum licenses and active network equipment (5G hardware). By selling their tower portfolios to a TowerCo, Telcos receive a massive cash infusion that can be used to pay down debt or fund 5G rollouts. Infrastructure generally yields a lower, stable return. By moving these assets off their balance sheets, Telcos can improve their financial ratios and focus their capital on high-margin services like data plans, fiber, and customer experience. In exchange for offloading the towers to towercos, telco companies will sign long term contracts (20-30 years) with the towerco companies. This enables them to treat the towers as an operating expense instead of capex and at the same time reduces the need for the telco companies to take on large amounts of debt.

Why are towercos such good businesses?

The main reason why the towerco industry caught my attention was because of the extremely high barriers to entry to the industry. In an industry where physical presence is everything, the barriers to entry are not just financial, but are deeply rooted in regulatory, geographic, and social constraints. These factors ensure that once a TowerCo establishes its footprint, it becomes nearly impossible for a new competitor to disrupt its dominance.

1. Infrastructure Monopoly

The primary barrier to entry is the extreme difficulty of obtaining permission to build towers. Unlike digital businesses that can scale easily, a TowerCo is tethered to the physical world of Zoning and Regulation. Local governments and municipalities treat tower permits with intense scrutiny, often requiring months or years of bureaucratic navigation.

This barrier is further reinforced by “Not In My Backyard” (NIMBY) sentiment. Residents generally desire high-speed connectivity but are often vocal opponents of new industrial structures in their neighborhoods. These social and legal hurdles act as a natural cap on supply. For an incumbent like Cellnex, every existing tower is a pre-approved asset that cannot easily be replicated, making the existing portfolio increasingly valuable as the demand for connectivity grows and the supply of new permits tightens.

Beyond regulation, the TowerCo business is governed by the laws of physics and geography. A tower’s value is derived entirely from its strategic location. To provide effective coverage, a tower must occupy the high ground which can be a literal hilltop (Bukit Timah hill & Sentosa for Singapore) or a prime urban rooftop.

In most high-demand areas, there is a finite amount of space that offers the necessary line-of-sight for radio signals. Once a TowerCo secures the optimal site within a specific radius, there is often no economical reason or legal room for a second competitor to build a rival structure nearby. This creates a “natural monopoly” at the micro-level. A competitor cannot simply build a tower 50 meters away; even if they could find the land, the regulatory authorities would likely deny the permit on the grounds of visual pollution and redundancy, forcing the newcomer to instead rent space from the incumbent. Once the minimum efficient scale for a particular country is reached, the company will achieve a natural monopoly.

2. Inflation adjusted revenues with clear backlog visibility

The other trait of the towercos that attracted me to the industry is the unique contractual framework that governs its revenue. By providing mission-critical infrastructure, companies like Cellnex command pricing power and a degree of stability rarely seen in other sectors. This stability is built on three pillars: high customer stickiness, multi-decade contract durations, and built-in protections against macroeconomic volatility.

In the telecommunications world, a tower is not merely a piece of real estate but it is a vital node in a live network. For a mobile operator, the prospect of moving equipment from one towerco to another is a logistical and financial nightmare as they would have to 1) employ a tower co to build new towers which requires high upfront costs and/or 2) Pay a breakup fee equal to the remainder of the revenue of the contract. Such a move is highly illogical especially considering the risk of localized service outages, and the potential for coverage holes that frustrate subscribers.

Consequently, the churn rate, the percentage of customers who leave, is incredibly low, typically staying from 1% to 2% (cellnex has a churn rate of 1.2%). Once an operator installs its antennas on a tower, it effectively becomes a tenant for life and is likely to renew its contracts 30-40 years down the road. This stickiness transforms the TowerCo’s revenue from a series of individual transactions into a highly reliable, recurring annuity.

Unlike typical commercial real estate where leases might be renewed every three to five years, Cellnex operates on Master Service Agreements (MSAs) that are 20-30 years long.

This long-term horizon provides the business with unparalleled revenue visibility. The business model of cellnex enables the company to essentially secure a guaranteed cash flow stream that will last through multiple economic cycles. In my opinion, this makes the business model behave more like a high-yield infrastructure bond than a traditional stock. The predictability of these 20-year contracts allows TowerCos to plan long-term capital deployments and dividend policies with extreme confidence. The certainty of the revenue streams also enable the towercos to take up large amounts of debt to fund their capital expenditures to build towers. In the case of Cellnex, the company has a revenue visibility via their backlog of over 30 years (110 Billion ex escalators on a 4B ARR). This backlog is the longest and the largest in the industry.

Lastly, Cellnex’s business model also offers a rare natural hedge against rising costs. The MSAs for cellnex include escalator clauses, which are contractual mechanisms that automatically increase the rent every year. These increases are typically tied to the Consumer Price Index (CPI) or a fixed annual percentage (e.g. 2%). Cellnex’s revenue contracts are split where 65% of the contracts are CPI linked and the other 35% are annual escalators of 1-2%. Cellnex’s contracts are capped on the downside (i.e. if there is deflation, the rent of the towers will not decline). These inflation-indexed contracts ensure that the business’s real margins are protected, regardless of the macroeconomic environment. While traditional businesses may struggle to pass on costs to consumers during inflationary periods, the TowerCo’s price increases are baked into the contract by design. What makes the business model of cellnex even more attractive is the fact that the company passes through electricity and utility costs, making the company immune to oil shocks and energy volatility. The only costs that the company is responsible for are land lease, labour and maintenance costs for the towers.

Understanding the shifts in the telco industry and why cellnex will benefit significantly from it.

1. Transition from 4G to 5G

Unlike America, Europe has yet to adopt 5G to a large extent. By early 2026, all three major US carriers (T-Mobile, Verizon, and AT&T) have officially launched nationwide 5G SA cores, enabling the united states to have 90% of the networks to have 5G capabilities as compared to Europe’s 40%. The strategic importance of 5G capabilities lies in their role as the connective tissue for the next generation of global economic productivity. Beyond simply increasing mobile download speeds, 5G Standalone architecture introduces ultra-low latency and massive machine-type communications, which are essential for the integration of Artificial Intelligence into the physical world as well as the utilization of AVs. The transition to 5G is expected to cause the average consumption to grow from roughly 15 GB/month in 2022 to 80-90 GB/month by 2030, marking a 28% CAGR.

So how will this benefit cellnex?

The most immediate benefit of 5G for Cellnex is the mandatory requirement for network densification. Because 5G operates on higher frequency bands that have a shorter propagation range and struggle to penetrate physical obstacles, mobile operators can no longer rely on a sparse network of large macro towers. To provide consistent coverage, they must deploy thousands of additional “points of presence” (PoPs). This forces operators into two paths, both of which favor cellnex:

1. They must either rent additional space on existing Cellnex towers (Colocation)

2. Commission Towercos like Cellnex to build new sites through Build-to-Suit (BTS) programs.

In 2025 alone, Cellnex saw its organic PoP growth rise by 4.5%, a direct result of tenants adding 5G equipment to existing sites. The 4.5% growth in revenue might seem minimal but in reality, it creates significant profit growth for the company especially if the clients choses option (1)-colocation.

Before I move on to explain what colocation means, I would also like to point out the second trend that is emerging.

(2) Increasing willingness for telco companies to share network infrastructure aka colocate

In the early days of mobile (2G and 3G), the product was the signal itself. If a Telco had a tower in a rural area where its competitor didn’t, it could charge a premium and steal customers. Coverage was a unique selling point. However, network coverage has now reached parity in Europe. Every major carrier has a signal in almost every city. Since you can no longer win customers just by having a tower, there is no longer a competitive advantage in owning the physical steel in the ground. Telcos now compete on brand, content, AI services, and customer experience instead.

To understand why colocation is hugely beneficial for the business below is an image to make it easier to visualise:

Collocation is a situation where there are more than 1 client sharing the cell network. To build a tower, the main cost associated with it is the building of the stem (the core structure). In order to house more tenants, the tower companies only have to add an additional antenna on the cell towers to bring in more revenue. To add the second client onto the tower, the cost is minimal considering that the land leases and the infrastructure costs and maintenance costs of the towers need to be paid for regardless of whether there are 1,2 or 3 clients. This means that almost all of the additional revenue gained from the new client (about 95%) flows directly to the bottom line of the company.

Cellnex currently has a tenancy ratio of 1.6 (which means that there are approximately 1.6 tenants per tower that they own across the 113K towers). We can estimate that the ROI of the capex that they deploy for the towers is likely to be around 15%.

Colocation will be one of the key drivers of the growth for the company in the long term.

The Origins Of Cellnex

The foundations of Cellnex were laid within Abertis Infraestructuras, a Spanish conglomerate primarily focused on toll roads and parking management. For years, the telecommunications arm operated as “Abertis Telecom,” managing broadcast towers and terrestrial television signals in Spain. However, by 2014, the strategic decision was made to spin off this division to capitalize on the burgeoning demand for mobile data. In May 2015, the company rebranded as Cellnex Telecom and launched its Initial Public Offering (IPO) on the Spanish stock exchange at a valuation of approximately €3.2 billion. This independence was the crucial catalyst; as a neutral player no longer owned by a specific phone company, Cellnex could market its tower space to all competing mobile operators without conflict of interest.

Era of Hyper-Expansion (2015–2022)

Following its IPO, Cellnex embarked on one of the most aggressive acquisition sprees in European corporate history. Led by CEO Tobias Martinez, the company recognized that the European tower market was highly fragmented compared to the United States. Cellnex utilized a Build-to-Suit and acquisition model, buying thousands of towers from operators like Wind Tre in Italy, Bouygues Telecom in France, and Sunrise in Switzerland.

The defining moment of this era occurred in 2020, when Cellnex reached a landmark €10 billion deal with CK Hutchison to acquire approximately 24,600 towers across six European countries, including the UK, Ireland, and Austria. This move solidified Cellnex’s position as a pan-European titan. By the end of 2022, the company had grown its portfolio from roughly 7,000 towers at its birth to over 130,000 sites across 12 countries, supported by a series of massive rights issues that demonstrated deep investor confidence in the TowerCo model. This aggressive acquisition period was emboldened by cheap capital with interest rates close to 0. This aggressive expansion plan caused the company to be laden with debt and the company also diluted its shareholders significantly.

Pivot From Growth to Investment Grade

By late 2022 and early 2023, the global macroeconomic environment shifted. Rising interest rates made the debt-fueled acquisition model more expensive. In response, Cellnex announced a fundamental shift in its strategy and chose to move from inorganic growth (buying more towers) to organic value creation and debt reduction. The company committed to achieving an Investment Grade credit rating from S&P, focusing on increasing the tenancy ratio of its existing towers rather than adding new ones (via BTS) that are beneficial to revenue but do not increase profitability by much. This period also saw a leadership transition, with Marco Patuano taking over as CEO to oversee this more disciplined financial phase.

The Pivot to Shareholder Yield

Having achieved its goal of becoming Europe’s largest independent tower operator, Cellnex has pivoted its entire corporate strategy toward maximizing shareholder yield and capital returns. Having successfully consolidated a pan-European portfolio of over 130,000 sites, Cellnex has moved into what management defines as Chapter 2 of its corporate history. As of 2026, the company has pivoted and is focusing instead toward the maximization of shareholder yield and the aggressive generation of Free Cash Flow (FCF).

In fiscal year 2025, Cellnex demonstrated its commitment to returning capital ahead of its original 2026 schedule, delivering a combined total shareholder yield of approximately 6.5%. This was achieved through a dual-track strategy of dividends and buybacks. The centerpiece was an €800 million share buyback program that saw the company repurchase approximately 24 million shares. By reducing the total share count, Cellnex achieved a 16.7% growth in Recurring Levered Free Cash Flow (RLFCF) per share.

This transition to a yield-focused model is further supported by a rigorous commitment to an Investment Grade (BBB-) credit rating. By the first quarter of 2026, Cellnex successfully brought its Net Debt to EBITDA ratio down toward a 6x target, a significant reduction from the high-leverage era of 2021. This de-leveraging was accelerated by the strategic divestments of non-core assets, including the 2025 sale of French data centers and other peripheral infrastructure. The company also initiated a €500 million dividend distribution from the share premium reserve, split between January and July 2026.This transition represents a shift in the stock’s DNA: from a volatile, high-beta growth play into a defensive, bond-proxy asset that offers both the stability of 30-year contracts and a growing yield profile.

Looking forward through 2026 and beyond, Cellnex has established a permanent “floor” for shareholder remuneration. The company has committed to a minimum of €1 billion in annual returns, split evenly between dividends and opportunistic buybacks. This dividend is projected to grow at a minimum of 7.5% annually through 2030, a rate secured by the inflation-linked escalator clauses and 5G densification mandates built into its Master Service Agreements.

Investment Thesis

1. Rapid FCF Expansion As BTS Scales Down

Cellnex reports its annual report using special terminology to better reflect the company’s financial situation. Seeing the company using the GAAP metrics will not give a proper understanding of the company and how it operates. I will be defining each of the terms that cellnex uses and give you a brief explanation of why the metrics are used instead.

1. Revenues (ex Pass-through)

This is the “clean” top-line number. Cellnex often pays for the electricity and ground rent at their sites and then bills that exact amount back to the telcos. The cost of electricity and energy will be passed through to the customers.

To calculate revenue (ex pass through) it is = Total Operating Income minus Pass-throughs (utilities, energy, and business rates) and Advances to Customers.

Since pass-throughs have zero margin and can be volatile (like energy prices), Cellnex strips them out to show the actual “rental income” they are earning from their tenants.

2. Adjusted EBITDA

This is their primary measure of Operating Profit before the heavy costs of their massive debt and tower leases.

Operating Profit before Depreciation and Amortization (D&A), adding back non-recurring items (like redundancy provisions or M&A costs) and non-cash expenses (like share-based compensation).

Adjusted EBITDA shows the cash-generating power of the towers themselves, independent of how they were financed.

3. EVITDAAL(EBITDA after Leases)

This is arguably the most important operational metric for a TowerCo.

Adjusted EBITDA minus Lease Payments (the rent Cellnex pays to the landlords who own the land under the towers).

Unlike a typical company where a lease is just an expense, Cellnex’s business is all about the spread between what they collect from telcos and what they pay to landowners. EBITDAaL shows the profit left over after paying their own landlords.

4. RLFCF (Recurring Levered Free Cash Flow)

RLFCF represents the cash available for shareholders in a normal year.

To calculate RLFCF it is = EBITDAaL minus Net Interest Paid (the cost of their debt) minus Maintenance CAPEX (money spent just to keep the towers from falling down) minus Income Tax Paid plus/minus Changes in Working Capital

It ignores Growth CAPEX (buying new towers) and focuses only on the cash generated by the existing portfolio.

5. FCF (Free Cash Flow)

This is the Bottom Line of the cash flow statement.

RLFCF minus Growth CAPEX (Expansion CAPEX and Build-to-Suit programs).

In the Hyper-Growth phase (2015–2022), this was often negative because Cellnex was spending billions to buy towers. In the 2025–2026 Value Phase Cellnex has pushed this to be slightly positive, as they are spending less on acquisitions and more on returning cash to shareholders.

Understanding the BTS model

Recall earlier that I stated that in order to meet the customer’s demand, cellnex has 2 choices: BTS and colocation. In the Cellnex model, BTS represents a contractual commitment where a mobile operator (like Vodafone or Orange) asks Cellnex to build a brand-new tower in a specific location where they currently have a dead zone or need 5G densification. Instead of building it themselves (which would sit on their own balance sheet and require high CAPEX), they ask Cellnex to do it. Cellnex will then find the land, get the permits, and build the steel structure after the mobile operator signs a 20-30 year lease with cellnex. As I mentioned earlier, colocation is a much more profitable venture for cellnex as the ROI from such an investment is higher. However the current BTS obligations are actually obligations that have been collected during the heavy acquisition era of cellnex. As of early 2026, Cellnex has a massive BTS backlog of approximately 15,000 to 20,000 committed sites to be built through 2030. Cellnex BTS backlog is expected to taper off to 0 from the current 1.2B euros after 2030 with the bulk of which being completed in 2028 which will be highly accretive to FCF.

FY 2025 annual report

The current FCF of the company is understated due to the fact that Cellnex has to still fulfil these BTS obligations. BTS obligations will have a diluting effect on cellnex’s tenancy ratios as the new BTS sites will only have a tenant ratio of 1. However, over the next 3 years, cellnex will be able to onboard other mobile operators onto the BTS towers that they have built and increase the tenancy ratios back towards their targeted goal. If we actually normalise the FCF of the company, the company is now generating approximately 1.55B in FCF on a market cap of 18B representing a FCF yield of 8.6%. Considering the aforementioned traits of the company, the company is trading at an absurdly cheap valuation especially considering the fact that other similar TowerCos such as AMT and INWIT are trading at 4.5% and 6.5% fcf yield respectively and Cellnex’s own towers were sold at a 4% ULFCF yield to private equity.

I’d like to take a moment to talk about INWIT which is a company that is very well run and has margins that I expect Cellnex to be able to achieve over the next few decades. INWIT consistently maintains one of the highest tenancy ratios in the world. As of Q3 2025, their ratio stands at 2.37x and the company’s long-term business plan aims to reach a tenancy ratio of 2.60x by 2030. Unlike Cellnex, which often buys single-tenant towers and then works to fill them, INWIT was born from the merger of the tower portfolios of TIM and Vodafone Italy. This gave them an immediate anchor of two massive tenants on nearly every tower from day one. Unfortunately, this causes the company to have high exposure to 2 companies that contribute 90% of its revenues. Furthermore due to its customer base, the growth path of the company is limited considering that competing companies’ mobile operators will not be willing to utilities their towers as they are run by the competitors. Cellnex on the other hand is a neutral operator and provides its service to every single major operator. Furthermore, Cellnex also has a larger TAM considering that it serves the pan European market whereas INWIT only serves the Italian market. Another key issue that Inwit faces is the concentration of the revenue from the 2 customers TIM and Vodafone who represent 90% of their revenues. Should the customers decide to not renew the contracts, INWIT will be hit severely (Update: Vodafone and Tim are threatening to break the MSA contracts). Cellnex on the other hand has numerous customers across 10 countries, thus the concentration risk for the company in terms of geography and customers are lower.

2. Valuation mismatch between Private and Public Markets

Cellnex over the past 2 years have divested the cell towers that they owned in countries where they are not the market leader. A tower, once constructed, demands a largely invariant cost base. A site technician must periodically inspect and maintain it regardless of whether one or three operators have equipment installed. A field engineer must respond to faults and perform equipment swap-outs whether the site generates €20,000 or €60,000 in annual revenue. A lease manager must administer the ground lease with the underlying landlord irrespective of tenancy. Regional operations managers, network operations centres, and the full apparatus of central functions such as finance, legal, regulatory, HR all exist as overhead that the portfolio must absorb. None of these costs scale meaningfully with incremental tenants.

This logic, however, only functions efficiently when the portfolio is dense enough within a given geography. A regional operations manager overseeing 1,500 sites in a compact market commands a meaningfully lower cost per tower than one overseeing 150 sites in the same country. Ground lease negotiations carry far greater counterparty leverage when the operator controls hundreds of sites and represents a meaningful share of a landlord’s portfolio than when it controls a token presence. The minimum viable cost of operating in a country is largely fixed regardless of portfolio size. This is the country’s overhead problem, and it is the mechanism through which thin market presence destroys unit economics. A 200-tower footprint in a given country bears a country cost structure that is not radically lower than a 2,000-tower one, yet it generates a fraction of the revenue.

It is precisely this logic that underpinned Cellnex’s strategic pivot and its associated divestitures. Having spent 2015-2022 acquiring towers across Europe in an era of cheap debt and accumulating positions in over a dozen markets. Cellnex found itself, by the early 2020s, holding a portfolio that was deep and competitive in some geographies and thin and subscale in others.

Markets such as Ireland and Austria, where Cellnex’s footprint was relatively modest compared to incumbent players, were identified as candidates for disposal. The capital repatriated from these exits were redeployed to reduce leverage, share buybacks as well as reinvested in markets such as Spain, Italy and France where Cellnex commands scale economics, dense portfolios, and the operational leverage to convert incremental tenancy into near-pure margin.

How Much Are Private Equity/Competitors Paying For Cellnex Towers?

1. France — Phoenix Tower International (PTI) + Bouygues Telecom JV

Towers: 3,226 sites

Valuation: €620 million priced at 24X EBITDaaL

Due to the aforementioned benefits of scale, competitors in the country are willing to pay high prices for extra towers in the geography where they have scale.

2. Sweden & Denmark (Cellnex Nordics) — Stonepeak (49% stake)

Towers: 4,557 operational sites + ~2,500 pipeline sites

Proceeds: €730 million (for the 49% stake, implying ~€1.49 billion enterprise value for 100%) EV/EBITDAaL: 24x 2024E EBITDAaL (acquisition was done in 2023)

3. Ireland — Phoenix Tower International (100%)

Towers: ~1,900 sites (~1,700 towers + ~200 land locations)

Proceeds: €971 million EV/EBITDAaL: ~24x EBITDAaL

Towers: ~4,600 sites

Proceeds: €803 million (including €272 million deferred to December 2028)

EV/EBITDAaL: Just over 20x, based on 2023 EBITDA

5. Switzerland — Process Ongoing

Towers: ~6,000 radio sites (Cellnex holds 72%; Swiss Life Asset Managers holds 28%)

Expected Proceeds: Up to ~€2 billion for Cellnex’s 72% stake

Implied EV/EBITDAaL: Cellnex refusing to sell stake unless “private markets overpay” (based on Marcos’s statements) Expected EV/EBTDAaL is 26.

As we can see from the past divestitures to companies and private equity, they are valuing the company’s towers at the Mid 20s EV/EBITDAaL multiples. However, the market is currently valuing cellnex as a whole at just 13.5 EV/EBITDAaL. This clearly shows that there is a disparity between the public markets from the private markets that signals potential undervaluation and potential value unlock as the company continues to divest its towers where it does not have operational efficiency and focus on the regions where economies of scale work towards their favour.

3. Shareholder oriented management and activist investor on the board.

TCI Fund Management, led by Sir Christopher Hohn, accumulated a position of just over 9% in Cellnex by early 2023, making it the company’s single largest shareholder ahead of both the Benetton family’s Edizione and Singapore’s GIC. The stake was built deliberately across direct shares and derivatives, giving TCI substantial economic exposure before it became a visible force on the shareholder register.

The immediate catalyst for TCI’s public intervention was Cellnex’s botched CEO succession following the departure of founding chief executive Tobías Martínez. Hohn wrote directly to the board in March 2023, declaring that the search had been mishandled and demanding the removal of three directors, including chairman Bertrand Kan. Within weeks, both Kan and director Peter Shore had resigned.

TCI’s representative Jonathan Amouyal was appointed as a proprietary director in April 2023 and, critically, placed on the Capital Allocation Committee. Amouyal spent over a decade at TCI specifically covering digital infrastructure and tower assets, meaning he arrived with deep knowledge of the precise levers that drive per-share value in this industry.

The impact on capital allocation has been direct and measurable. Under TCI’s influence, Cellnex formalised a €1 billion shareholder return programme comprising €500 million in dividends and €500 million in buybacks through end-2026, followed by a commitment to a minimum €500 million annual dividend growing at 7.5% per year from 2026 onwards. The buyback is being executed at roughly 14x EBITDAaL, against a private market clearing price of 20–24x, making each share retired mechanically accretive to intrinsic value per share. Simultaneously, the divestiture programme across Ireland, Austria, Sweden, Denmark, and now Switzerland which TCI’s capital allocation oversight helped accelerate. TCI recycled proceeds from subscale, low-return positions into debt repayment and buybacks, compressing leverage and expanding FCF simultaneously.

Compensation structure for Marco Patuno (CEO of Cellnex)

1. The most direct link is the explicit inclusion of cash shareholder remuneration as a standalone metric within the annual bonus. The 20% individual objective weighting tied specifically to dividends and buybacks means that Patuano benefits personally when capital is returned to shareholders. This compensation structure is absolutely unique and the first that I have seen in a company as most executive remuneration frameworks treat capital return as an output of financial performance rather than a performance metric in its own right.

2. The second mechanism is the LTIP’s dual total shareholder return structure. By measuring both absolute TSR and relative TSR against a defined peer group simultaneously, the plan ensures that Patuano’s long-term equity award only fully pays out if Cellnex shareholders have actually made money in absolute terms and have outperformed the alternatives they could have held instead. The compensation structure is as stated in their CEO remuneration investors package: LTIP carries a target opportunity of 183% of fixed remuneration, with a maximum of 275% of that target. The maximum opportunity would only be paid if the maximum achievement scenario was attained in the financial and sustainability metrics, and if Cellnex’s annualised and compounded absolute TSR was at least 20%, with Cellnex ranked first versus peers.

Patuno’s compensation structure is dependent on the following(from cellnex investors presentation):

For the annual bonus, each of the five metrics revenue growth, EBITDAaL, recurring leveraged free cash flow, adjusted net debt/EBITDA, and cash shareholder remuneration carries an equal 20% weighting, with threshold performance defined as follows: revenue growth must reach 95% of the budgeted target; EBITDAaL must reach 95% of target; RLFCF must reach 95% of target; adjusted net debt/EBITDA must be no worse than 103.13% of the target ratio (i.e. leverage can be marginally higher than planned but not materially so); and cash shareholder remuneration must reach 80% of the committed target. In each case, hitting threshold unlocks a payment level of 75% of target for that metric, while hitting maximum defined as 107.5% of budget for the financial metrics and 108% for shareholder cash remuneration unlocks 150% of target. Intermediate outcomes are calculated by linear interpolation.

In euro terms, that translates to the following payout schedule across scenarios:

At below-threshold performance, no LTIP is paid whatsoever. At threshold, the payout is 50% of target incentive (approximately €1.19 million). At target, the full 183% of fixed remuneration pays out approximately (€2.38 million). At maximum, the payout is 275% of that target incentive (approximately €6.54 million), deliverable entirely in Cellnex shares.

3. The third mechanism is the mandatory shareholding requirement, now set at three times fixed remuneration following the 2025 policy revision. At a fixed salary of €1.3 million, this requires Patuano to hold approximately €3.9 million of Cellnex stock on an ongoing basis. Combined with the LTIP being granted entirely in shares, his personal net worth is substantively exposed to the same price movements as any other shareholder. He cannot hedge through option asymmetry or extract value through a rising market while being insulated from the downside.

With current RLFCF yield sitting at close to 11% and normalised FCF yield (ex BTS capex) of 8.6%, we can be sure that management will be returning that FCF to its shareholders instead of doing dumb acquisitions considering that management incentives are aligned well with its shareholders.

To boost the RLFCF of the company, cellnex started celland. Its purpose is to convert the ground leases sitting beneath Cellnex’s towers from recurring annual costs into one-time capital investments, progressively shifting the company from tenant to owner of the land its towers sit on. Ground lease payments to landowners represent 15–25% of tower revenue which is a structurally large and growing cost line. These leases are typically denominated with annual escalation clauses, meaning that in an inflationary environment, the cost of sitting on that land compounds upward year after year with no corresponding benefit to the tower operator. Cellnex currently owns the land beneath fewer than 15% of its towers which is already the highest ownership proportion of any European towerco, but dramatically below the 40–70% ownership rates typical of US tower companies like American Tower and Crown Castle. Cellnex targets to buy back land from farmers/landowners at approximately equivalent to ten years of rental payments, a significant outlay at face value. But Cellnex’s ground leases are predominantly structured as 25–30-year commitments with escalation clauses, meaning that over the full life of the lease, the cumulative rental payments would far exceed the purchase price. By converting the leases into an appreciating asset and also having the ability to raise prices at CPI +escalators will enable the company to have significant operating leverage in the long run as the costs that cellnex is in charge of will not increase as much the revenue.

4) Continuous improvement in tenancy ratios as Colocation rates increase

Over the past 3 years, the tenancy ratio of cellnex has increased from 1.39x in 2023 to the current 1.6 in 2025. As the BTS orders are completed through 2030 (with the bulk of it completed by 2028), cellnex will add an additional 10K sites which will bring their total tower portfolio to 122K from 113K currently. However these new BTS towers will have a dilutive effect on the tenancy ratios as the new BTS towers only have a tenancy ratio of 1. Despite the dilutive effect of the new towers that are coming online, management is still guiding a tenancy ratio of over 1.7 by 2030.

To put into perspective on whether this is feasible, we can compare it to the American tower counterparts who also have a dominant market share in their respective markets. Tower companies in America such as AMT and crown castle are most likely the most representative when comparing to cellnex given that it is what Europe will likely achieve in the future as it. The tenancy ratios are 2.6X and 2.5X respectively for each company. Considering that the markets that cellnex operates in such as Italy, France and Spain have similar market structure for the telco companies (3-4 major operators) it is not unfeasible for cellnex to obtain such ratios in the distant future.

In my DCF model, I will assume that cellnex will hit a 2.5X in tenancy ratios by 2050 and maintain at those tenancy ratios for the years to come.

Common Worries By Investors

1. High Debt Levels

Some investors might point to the high debt levels and say that it is a cause of worry, whilst I tend to avoid high debt companies, I make an exception for cellnex due to the following reasons.

1. Cellnex has a stable recurring revenue that is not cyclical in nature. The visibility of the revenue and underlying profits and FCF are clear due to the contractual nature of the company. In situations like this where the company is able to invest in upgrades and build more towers to generate a leveraged ROI of 15-20% and borrow at an interest cost of 2-3%, it is a no brainer to saddle the company in debt. Furthermore, to alleviate investors concerns of the high debt levels, the management has set a target Debt/EBITDA of 5-6X to ensure that the company remains investment grade that will enable the company to refinance their debt at a more affordable rate.

2. The company also has strong liquidity and does not face refinance risk till 2027. Furthermore, based on management’s guided RLFCF in 2027, the company is expected to maintain 2.1B in RLFCF despite the divestiture of some of its towers. If push comes to shove, cellnex will be able to cease all growth capex to cover the upcoming debt obligations required. The 2.1% average cost of the debt is also an asset to the company considering that current refinance rates for the company are at 3.2%. Considering that the company also has 2.7B in cash and is able to pay down 2.1B in debt a year if they cease all growth and expansionary capex, the likelihood of the company going bust is near 0 and the current high debt levels is maintained due to the management’s decisions instead of them being unable to pay the debt off (as seen from management using the 1B capital from the disposal of their assets to buyback shares instead of paying down debt).

Another key risk that investors are worried about is the recent trend where MNO (mobile network operator) consolidation in Europe is also another factor that is worrying the investors. The wave of European MNO consolidation MasOrange in Spain, VodafoneThree in the UK, and Fastweb/Vodafone Italia in Italy is the single most discussed risk factor among investors in Cellnex and European towercos more broadly. The concern is intuitive: if two mobile operators merge, they previously occupied two sets of tower slots, and post-merger they theoretically need only one. The worry is structural churn, the permanent loss of a co-location tenant who is no longer a separate entity. But the reality is considerably more nuanced, and the investor debate spans several distinct dimensions. The first and most immediate concern is network rationalisation churn. Towercos face real challenges from the increasing consolidation of European mobile markets, such as MasOrange in Spain and VodafoneThree in the UK. When two operators merge, the combined entity will almost always find sites where both networks have overlapping coverage, and on those sites it will eventually decommission one of the two tenancies.

However, in reality what we do see is that there is limited churn due to the special clause in cellnex’s contracts. When a mobile operator merges with a competitor, the merged entity inherits two sets of tower contracts one with each legacy operator’s towerco arrangements. Left unconstrained, the merged operator could selectively retain the towers it deems strategically valuable from each portfolio, cherry-picking the best sites from Cellnex while exiting the less commercially attractive ones. This selective churn would be the worst possible outcome for a towerco as it would lose its highest-density urban sites while retaining only marginal rural towers with weak economics. The merged operator would effectively extract maximum value from the towerco’s portfolio while minimising its own contracted obligations.

The all-or-nothing clause under cellnex’s contracts eliminates this optionality entirely. Under its terms, a merged operator cannot selectively exit individual towers from a Cellnex contract; it must either maintain the entirety of its contracted tower portfolio or exit the entire portfolio in one single transaction. The clause converts what would otherwise be a gradual, unpredictable erosion of tenancy into a binary, negotiated event. Cellnex either keeps all of an operator’s towers or loses all of them, and losing all of them triggers a contractually defined termination payment that compensates Cellnex for the full present value of the remaining contracted cash flows on every surrendered site.

The practical consequence is that a merged MNO faces an enormous financial barrier to any churn. Walking away from a Cellnex contract under an all-or-nothing structure requires paying termination fees across potentially thousands of sites simultaneously which is a capital outlay that in most realistic scenarios exceeds the cost saving the merged operator would achieve by rationalising duplicate coverage. The clause therefore functions not just as compensation protection but as a strong economic deterrent to churn in the first place.

The MasOrange situation in Spain is the clearest live illustration of how this operates in practice. When Orange Spain and MásMóvil (both are mobile network operators) merged to create MasOrange, they became Cellnex’s single largest Spanish customer, sitting across contracts originally negotiated separately with each legacy operator. Rather than facing a wave of unilateral site withdrawals, Cellnex negotiated a comprehensive settlement: the 1,027 site withdrawals that appeared in the FY2025 results were not unilateral exits but negotiated reductions agreed as part of a broader contract restructuring that simultaneously extended the MasOrange relationship through 2048 with an all-or-nothing clause on the remaining portfolio and unlocked thousands of additional contracted PoPs (points of presence) through expanded RAN sharing arrangements with Digi. The churn was compensated, the duration was extended, and the remaining portfolio was ring-fenced with even stronger contractual protection than existed before the merger.

As other MNOs start to merge together, it is unlikely that this dynamic will be impacted and it is highly likely that cellnex will still be able to maintain the low churn rate that it is currently enjoying of 1.2%.

Valuation

Cellnex’s 2026 guidance puts RLFCF between €1.9 and €2.0 billion, giving a midpoint of €1.95 billion as the base year.

Our valuation is anchored to Recurring Levered Free Cash Flow (RLFCF), which we treat as the most appropriate metric for a business of Cellnex’s profile. Unlike net income, which is heavily distorted by depreciation on a vast asset base, or EBITDAaL, which abstracts away the cost of capital, RLFCF captures the actual cash the business generates for equity holders after servicing its debt, paying its taxes, and funding the maintenance and efficiency capital necessary to sustain the portfolio. It is the number that funds dividends, buybacks, and deleveraging, and therefore the number that should anchor any honest assessment of intrinsic value.

We use management’s 2026 guidance midpoint of €1.95 billion as the base year RLFCF. From this base, we apply 11% annual RLFCF growth through to 2028. This rate reflects two concurrent tailwinds that are already visible in the operating numbers: first, the continued clearing of the Build-to-Suit (BTS) backlog, which systematically converts committed capital expenditure into revenue-generating sites; and second, the operational efficiency programme that management has been executing since 2024, including the voluntary redundancy plan in Spain and the industrialisation of site management processes. These are already flowing through results, with EBITDAaL margins expanding from 60.6% in 2024 to 62.2% in 2025.

From 2029 to 2030, we step the RLFCF growth rate down to 7%, reflecting the natural deceleration as the BTS pipeline matures, offset by the compounding benefit of rising colocation ratios. As the tenancy ratio progresses beyond its current 1.6x, each incremental co-location tenant adds revenue at approximately a 95% margin, creating a structural tailwind for cash flow conversion that persists even as top-line growth moderates. Applying these assumptions, we arrive at a 2030 RLFCF of approximately €2.75 billion.

On revenues, we assume 5% annual growth through 2028, consistent with the contractual escalator mechanisms embedded in Cellnex’s MSAs and the organic volume growth from BTS completions and new co-locations both of which are already in evidence as seen from FY2025 revenues growing 5.8% organically. Beyond 2028, we moderate the revenue growth assumption to 3%, reflecting a more mature portfolio where incremental site construction decelerates but the colocation-driven mix shift continues to improve cash flow conversion at the margin level.

To arrive at a terminal equity value, we apply an 8% RLFCF yield on the 2030 RLFCF figure, implying a terminal value of approximately €34.4 billion. An 8% yield is a deliberately conservative exit assumption. The 10-year Bund, the long-run risk-free rate anchor for European infrastructure, sits materially below this level, and comparable tower assets have traded in private markets at implied yields of 4%. Using an 8% yield therefore builds in a substantial margin of safety against rate risk and mean-reversion uncertainty, making it a more defensible anchor for intrinsic value than a private-market-style multiple.

To this terminal equity value, we add the present value of the cumulative capital returns that accrue to shareholders over the holding period. Under Cellnex’s committed capital allocation policy, shareholders receive a €500 million dividend in 2026, growing at a minimum of 7.5% per year thereafter through 2030. In addition to dividends, we model buybacks of €300 million in 2026, €500 million in 2027, and €1 billion per year from 2028 onwards to 2030.

The mechanical significance of the buyback programme cannot be overstated. Aggregating dividends and buybacks over the five-year period yields cumulative capital returns of approximately €6.7 billion to shareholders, which represents over 30% of the current market cap. Aggregating the present value of the €34.4 billion terminal equity value and these cumulative capital returns and dividing it by today’s current share count, we cleanly capture the value per share.

Discounting the terminal equity value back at an appropriate rate and adding the present value of capital returns, we arrive at an intrinsic value estimate in the range of €42 to €47 per share on a fully diluted basis, against a current market price in the mid to high €20s. This implies an upside of approximately 45–60% to intrinsic value before accounting for any significant multiple re-rating toward private market levels. The key sensitivity in the model is the exit yield assumption; if the multiple rerates to a 6% RLFCF yield, which is still above where comparable private market transactions have cleared, the intrinsic value rises above €55 per share. The range of outcomes under realistic assumptions is therefore consistently above the current price, with downside scenarios requiring both a permanent derating of infrastructure assets and an extended period of rate pressure to threaten the thesis materially.

Potential near term catalyst

Cellnex is currently in talks to sell its 72% stake in its Swiss tower unit to Manulife Investment Management, the global asset management arm of Manulife Financial Corporation. The Swiss unit manages approximately 6,000 sites across the country, and Manulife has been working with financial advisors on a possible transaction.

JPMorgan analysts have estimated the entire Swiss unit’s value at approximately €2 billion, with the remaining 28% held by Swiss Life Asset Managers. Cellnex’s 72% stake would therefore imply a transaction value in the region of €1.44 billion for its portion alone. Cellnex entered the Swiss market in 2017 when it acquired Sunrise subsidiary Swiss Towers for €430 million meaning the implied exit price represents more than a 3x return on the original entry cost across the full unit, and a meaningful premium to book value.

This follows a previous attempt to sell the Swiss operations last year to EQT, which CEO Marco Patuano confirmed were halted due to bids not meeting the company’s expectations which clearly shows that Cellnex has pricing discipline and is not a forced seller, which strengthens its negotiating position. Third, the proceeds, if a deal is reached, would be recycled into further debt reduction and shareholder returns, consistent with TCI’s capital allocation framework and directly accretive to RLFCF per share as leverage compresses.

Should the sale of Cellnex’s Swiss towers be successful, we expect there to be meaningful share buybacks similar to the buybacks that the company had done after the sale of Austria towers as well as its data center business in France. Any news of the sale of the swiss division will likely cause the company to rerate upwards.

Let me preface this by saying that SpaceX is genuinely one of the most impressive businesses I have ever come across, what is more sexy than a rocket shooting company that is run by the richest man on earth. The technology, the moat as well as the ambition to achieve the goal as an interplanetary species is genuinely unlike anything else that exists. But as I have said before, our job as investors is not to admire a business. It is to buy it at the right price and ensure that the risks that we are taking are as minimal as possible. And right now, I am not sure the price on offer makes any sense at all.

Before I get into the valuation, I should be transparent about something. I already have exposure to SpaceX. Not directly, but through my position in Alphabet. I remember back when I was valuing the company in mid 2025 when the company shares were collapsing to 160 dollars a share, I was conversing with guan rui on the potential of Space X shares that alphabet owned as a downside protection to the future valuation of the company. Naturally, we did not believe that a mere 6% stake in Space X would shift the valuation much and thus ignored it. However, forward a year and half later, the shares that alphabet has in Space X has the potential to be worth close to 200B which would be over 10% of the market cap that Alphabet was trading at back then. Alphabet is being run by exceptional capital allocators who has shown a knack and a consistent track record in investing in small niche companies that have lead to tremendous economic value to the company. The deepmind acquisition and onboarding Demis Hassabis onto the google team is arguably one of the most successful investments that google has made since the company’s inception. Turning back to Space X, here is the history of Alphabet’s investment in the company. Google invested in SpaceX back in 2015 as part of a $1 billion funding round that valued the company at just $10 billion, and as of 2025 held approximately 6.11% of the company. At a $2 trillion IPO valuation, Alphabet’s original $900 million investment could be worth more than $150 billion which is a return that would make it one of the most lucrative venture bets in history. This windfall flows through to me as an Alphabet shareholder, and it is one more reason why I have been watching this IPO so closely. I am already a beneficiary of SpaceX’s success. The question is whether I should be paying up for more at the IPO price and actually owning a direct stake in the company.

AI Rockets & Social Media-What do they have in common?

Strip everything away and SpaceX today is really three businesses rolled into one. Starlink, the rocket launch segment, and xAI.

Starlink is the crown jewel. It contributed $11.4 billion in revenue and $4.42 billion in operating income in 2025, has crossed 10 million subscribers across 160 countries, and is the only profitable segment in the entire company. It is a near-monopolistic infrastructure business in low earth orbit with real switching costs, a massive and growing addressable market, and zero credible competition at scale. If you were to buy Starlink standalone, the investment case would practically write itself and to be honest, it is probably what will drive the bulk of the intrinsic value of the company. One of the more interesting industries that I looked into before is the VSAT industry which stands for Very Small Aperture Terminal. It has been the industry standard for decades, operated primarily by Inmarsat (now owned by Viasat), and it is one of the most egregiously priced services in the world. Hardware alone runs anywhere from $20,000 to over $100,000 for the antenna equipment, and monthly service fees have historically ranged from $500 to $25,000 depending on bandwidth. For a commercial vessel wanting unlimited data, operators were routinely paying $25,000 a month. Not a typo. Per month. For internet.

The reason shipping companies paid it is simple- they have no choice. When you are in the middle of the Atlantic Ocean, Inmarsat is the only game in town. That is the definition of a captive customer base, and Inmarsat exploited it accordingly for the better part of four decades. The global maritime satellite communication market sits at roughly $4-5 billion today and is projected to nearly double by 2034. It is not a small pond and ripe for disruption.

Then Starlink arrived. In September 2025, SpaceX cut its unlimited maritime data plan from $25,000 a month to $2,500. Hardware costs $2,000. That is a 90% price reduction on monthly fees delivered almost overnight. And critically, Starlink’s LEO constellation offers latency of 20-50 milliseconds versus VSAT’s 600-800 milliseconds, starlink is not just cheaper, it is a categorically better product. This is the displacement story that does not get talked about enough when people discuss Starlink’s TAM. It is not just residential internet in rural areas which elon loves to tout about in his “save humanity campaign”. It is systematically dismantling legacy monopolies like VSAT that have been price gouging captive customers for decades and doing it with a product that is technically superior in every measurable way. It is highly likely that space X will obliterate the competition in that industry given that its product is better, cheaper and faster.

The launch business is similarly entrenched. SpaceX completed over 160 orbital launches in 2025 and is so deeply integrated into U.S. national security and NASA programmes that it has effectively become critical government infrastructure. Governments do not abandon critical infrastructure. The competitive moat here is as close to indestructible as it gets in any industry. Plus you have the potential of data centers being built on the moon in the future (though I would not hold my breath)

Then there is xAI. And this is where I start to lose sleep.

The xAI problem is not small

xAI contributed $3.2 billion in revenue in 2025 whilst burning approximately $14 billion in cash. To put that more bluntly, xAI alone consumed more cash than every other SpaceX operation generates combined. In Q1 2026, the AI segment posted a further $2.5 billion loss on its own. The combined company swung from a $791 million profit in 2024 to a $4.94 billion net loss in 2025. That is not a speed bump. That is a structural problem that investors are being asked to look through and pay a premium for.

The bulls will say this is the cost of competing with OpenAI, Google and Anthropic. And they are not wrong AI infrastructure is brutally capital intensive. But the question I keep coming back to is: what gives xAI the right to win? It is entering a market with deeply entrenched incumbents who have structural advantages in compute, talent, distribution, and data that are genuinely difficult to overcome and added to it there are valid arguments to be made that LLMs will simply become a commodity. Some may cite that XAI having elon musk in the helm is the moat of the company, however being Elon Musk’s AI company is a narrative, not a moat.

The valuation makes no sense to me at this stage

SpaceX is targeting a $2 trillion IPO valuation, which works out to roughly 94 times 2025 revenue. At that price you are not paying for what SpaceX is today. You are paying for what it might become 20 years down the road: Starship achieving commercial viability on schedule, xAI winning the AI race, and space-based data centres becoming a real business, all at the same time. Each of those is a possibility. None of them are certainties, and you are being asked to pay as though they already are.

Too many investors at IPOs look purely at the reward. They get swept up in the story, the brand, the ambition of the founder. They forget to look at the downside which is the key to investing and maintaining your wealth. In this situation, the downside here is real. Musk himself has warned of genuine bankruptcy risk if Starship fails to achieve adequate flight rates and none of the business except for starlink has a proven business model that is able to generate economic profits.

There is also the key person risk that I do not think gets discussed enough. You are buying into a business whose controlling shareholder is simultaneously running Tesla, X, DOGE and a tunnel-digging company. That is a concentration of distraction that has no comparable precedent in the public markets. Furthermore, those buying into space X due to the musk premium might in the near future see it vanish if 1) Musk goes on a political tirade yet again (space x revenue is heavily tied to government contracts) 2) One of Musk’s business fails/stock tanks (Tesla might tank significantly as Musk shareholders sell tesla to buy space X) or 3) He gets assassinated (lets hope not but given how much controversy and hate he has stirred up- man better ups his security as the chance is not non-zero).

Where I actually stand

Starlink on its own justifies a significant valuation. It is recurring revenue, high margin, near-monopolistic, and growing explosively. A fair standalone multiple on that segment alone puts the floor somewhere in the $200-300 billion range which is a real, defensible value that does not require you to model out space-based AI infrastructure to justify it.

But that is not what is on offer at $2 trillion. At that price you are paying for Starlink plus a cash-burning AI venture plus an unproven super-heavy rocket plus governance risk all bundled together and dressed up as the investment opportunity of a generation.

I already have my SpaceX exposure through Alphabet, and I am perfectly happy with that. Alphabet paid $900 million for a stake that may soon be worth north of $200 billion and I bought into alphabet at a valuation where the downside was genuinely capped and the upside was enormous. That is exactly the kind of asymmetric bet I look for. The IPO at $2 trillion is the opposite of that. I do not doubt the quality of some parts of this business but what I definitely doubt is the price and as the old adage goes- price is what you pay and value is what you get. At $2 trillion, there is very little protecting you on the downside, and an enormous amount that has to go right for the upside to materialise.

I will be watching closely as an indirect beneficiary from Space X valuations but I will not be rushing in. I have no doubt that the space X IPO will be successful given the current casino-like characteristic of the markets and the mania gripping investors currently but I will be on the sidelines and gaining indirectly from the IPO via my stake in Alphabet. “Investors” who buy into space X should look into the mirror and recognise themselves as speculators and should not delude themselves into thinking that they are investors- a simple question they should ask themselves is if the shares of the company fell 50% the next day (assuming nothing about the business fundamentals have changed), would they hold on or even buy the company? If they won’t, they should not be buying the company’s stock in the first place.

There will undoubtedly come a point in time where Space X trades at a reasonable valuation that will incentivise me to do a full DCF and really think about the company and its prospects. However, now is definitely not the time as time can be much better spent on a few companies that I am looking to in France and Spain (more about that on a future writeup).

TCI sold Microsoft for two main reasons: (1) TCI sees risk in Microsoft 365 suite of products and (2) TCI is wary of Microsoft’s Azure business, probably tied to the increased concentration in OpenAI and lack of Backlog Growth. TCI reduced their stake in the company from 10% to 1%, allocating the capital into CRAs, Google, and probably Visa.

Addressing the Azure business, they have concerns with concentration and the rate of growth. In the last earnings, Microsoft’s Azure backlog increased by a mere 2%, despite strong backlog growth numbers for some of its peers - Google, Amazon. Compounding this “risk” is the concentration of OpenAI in their backlog numbers, with currently about 40% of backlog tied to OpenAI.

Microsoft’s management directly addressed this backlog number during their earnings call, citing a pivot in the business model, from booking of future revenue to a “pay as you go” model. Meaning, the company is getting enterprises to pay for subscriptions, and layering on top of that a usage based model. Amazon are the kings of this model. Management also mentioned constraints within the data centre equipment market, and that demand is outpacing supply. As such, because of certain shortages in chips, the company has been unable to tie enterprises to long duration contracts that span over years. In the case of OpenAI concentration, stripping it out results in a 26% increase in backlog numbers YoY. While not as impressive as peers, it is still a respectable number, with revenue growth in Azure being up 40% YoY.

In the case of Microsoft 365 suite of products, this is how I would view the situation, and Chris Hohn’s thesis. His thesis for Microsoft has been built around the bundling moat of the company, and in his view, the bundling moat is weakening with the emergence of AI.

On the contrary, my thought process is as such: There are two camps of Microsoft 365 usage, the necessary and the discretionary. In the case of the necessary products, these include Excel, Powerpoint (partly), Entra ID (Identification), Outlook, Microsoft Teams, Sharepoint etc. The risk now that Chris Hohn is pricing in is in this discretionary bucket. His fear, is that AI will increasingly take up tasks within Microsoft’s ecosystem, such as writing emails, creating simple slide decks on Powerpoint, summarising meeting notes etc. He believes that AI will weaken Microsoft’s bundling moat, which is why he sold out of his position.

Microsoft is combating this by incorporating Copilot into their ecosystem, turning Copilot into a mandatory upgrade and a productivity boost, rather than a discretionary one. Microsoft is actively incorporating Copilot to increase the value proposition of the entire bundle of Microsoft products.

So in the case of reduced usage of Microsoft products, there are two views that an individual can take: (1) Enterprises will cancel their Microsoft suite and get AI agents to do it instead, externally. (2) Enterprises will use the AI products that are already within the Microsoft Ecosystem, because they need the other products within the bundle as well.

I believe, the latter is the clearer version of the two. Adding more nuance to this, assuming this is an argument on seat dilution, or ARPU weakness for Microsoft. My view is that Microsoft’s necessary products will strengthen enterprise renewal, providing Microsoft with an active base to upsell Copilot into. In a world where AI becomes increasingly commoditised, the power lies in the distribution of products.

AI Agents are only as useful as the data allows it, due to the highly predictive nature of AI. Without data in Outlook, Sharepoint, OneNote, there will be no use cases for the AI Agent. For an external agent to work, an enterprise would have to "export" its data to a third party, creating a massive security and compliance hurdle. Microsoft, on the other hand, already has permission to these data.

So, while AI may potentially reduce the need for individual applications, the key is whether or not Microsoft can own the orchestration layer sitting above workflows. Even with AI, enterprises will still require systems for identity, permissions, compliance, collaboration, and data storage. Given’s Microsoft’s position in Teams, Outlook, Entra ID, Excel, etc, the company is well positioned to incorporate AI directly into their products, with Copilot sitting as an orchestrator, through a multi-model strategy. As inference costs decline over time through improvements in model efficiency, Microsoft may eventually benefit from operating leverage if Copilot achieves widespread enterprise adoption.

Falling Stocks & 13Fs

1. Constellation Software/Topicus

Constellation Software and Topicus are interesting companies, I highlighted Topicus a few weeks back. Currently, they are trading at around 17x P/FCF2AS, with around 15-20% growth.

Constellation Software reported a very strong quarter, with normalised FCF growing at about 10% (the headline is 44% but it is due to a valuation change), while cash deployed stood at 110% of FCF. This is positive for the company, showing that the software sell off has provided opportunities, although they mentioned that the decline in valuations is slight.

The way to think about Constellation Software is as such: They own hundreds of niche software businesses that bring in cash. They redeploy that cash into more such businesses at great prices. The compounding works as long as (1) the majority of the portfolio stays undisrupted, (2) the deal pipeline remains large enough, and (3) returns on incremental capital stay attractive. Currently, the key risk lies in the deployment rates, that means, can the company continue to deploy these cash at good returns.

Investing in the company is essentially a bet on the company’s entire system that has been built over the years. Treat it like a Berkshire style investment, where you can’t expect much edge in insights, but can stand to gain when sentiments and valuations turn sour.

2. SPGI

SPGI is another company that has been hit with the entire AI narrative. Claude recently introduced new legal plug ins to be incorporated into platforms like Thomson Reuters etc, and the stock fell 5% on the day of the news release.

For SPGI, my thought process is as such. The company has a few different businesses that are disruptable (Market Intelligence), but there are other very strong businesses that are not prone to any form of disruption (Commodity Insights, Credit Rating, Indices). Commodity Insights is a benchmark, Credit Rating is a standard, none of which can be displaced by AI. Each segment commands relatively strong pricing, and has secular tailwinds behind it.

A way to think about this investment is to wait for the markets to completely discount away the entire Market Intelligence business. From then on, the upside is more or less asymmetrical.

3. MSFT

4. MA/V

Mastercard and Visa have been hammered over the recent months, with both stocks down about 10% YTD. In our view, the markets are underpricing (1) the durability of these franchises and (2) their respective VAS businesses.

While the markets continue chasing semiconductor stocks, they have left behind all these quality businesses that are trading at their lowest valuations in the past 10 years. Again, in large caps, while it may be tough to have actual insights to generate alpha, sometimes playing on valuations and sentiments is sufficient.

Safran Analysis

Introduction

Safran is a company that was founded in 2005 following the merger of SNECMA and SAGEM. Its roots trace back to 1945, where SNECMA was formed as a state driven project for manufacturing of engines, rebuilding the aerospace industry as a pillar for national independence.

Explaining Safran’s position in the Commercial Aerospace Industry

In civil aviation, Safran owns CFM international, a 50/50 joint venture with GE that produces the LEAP engine, the most popular jet engine program in the world. CFM international holds a dominant share in the narrowbody segment of commercial aircraft, with a monopoly in B737 aircraft, a 70% share in A320 aircraft and a monopoly in COMAC’s commercial aircraft.

The company provides these engines to aircraft OEMs, primarily Airbus and Boeing in the commercial aviation segment. Engines are typically sold at a loss, with significant incentives given to OEMs to incorporate the engines on the aircraft. This is intentional, as it allows them to collect strong aftermarket revenue from the fleet owner for 25-30 years following the initial sale of the plane.

Safran’s Business Segments

Safran operates primarily 3 different segments: Aerospace Propulsion, Equipment & Defense, as well as Aircraft Interiors.

Aerospace Propulsion

In Aerospace Propulsion, the company derives most of its revenue from CFM international, the joint venture with GE. There are currently two different generations of engines that the company provides - CFM56 and LEAP engines. CFM56 engines are currently being phased out, with the company now prioritising LEAP engine deliveries.

Generally, there are two ways for CFM international to derive revenue: (1) OE sale (2) Services & Spare Parts. Services & Spare Parts are much more margin accretive than the initial OE sale, because these are directly charged to the air lessors/airlines, and proprietary parts are provided which allow for strong pricing power at 6-7% annually. Safran controls the FAA/EASA Part Design Approval (PDA), so Airlines/MROs cannot switch to non-OEM parts without a PMA. PMAs are typically not allowed for critical engine equipment.

Within Services & Spare Parts revenues, there are generally two ways for the company to receive revenue: (1) Performative Restorations & Overhauls (Services) (2) Kits (Spare Parts). The Spare Parts business is higher margin than the Services business due to the need for labour within the services business.

For CFM engines, currently most of the revenues are derived from the spare parts business, because 70% of services are currently being outsourced to 3rd party MROs (Maintenance, Repair, Overhaul). 3rd party MROs, however, will still have to pay for the spare parts, along with a licensing fee to CFM. Within the CFM56 spare parts business, GE receives revenue for High-Pressure Turbine (HPT) parts, which are more expensive and require more replacement than Safran’s Low-Pressure Turbine (LPT) parts. However, following the new LEAP engine generation, the split will be more equal between the two companies.

Within the spare parts (kits) that CFM provides, there are typically 3 different types of kits that are provided:

• Performative Restoration Kits: These are spare parts required during every shop visit. (SV1 and SV3)

• LLP Kits: These are parts that are mandated to be replaced after a certain flight limit, and is the most expensive out of the kits that CFM provides (SV1 and SV2)

• Durability Kits: These are kits to prevent early wear and tear issues, allowing aircraft to extend their wing time. This is currently only offered to LEAP engines, with the provision of these kits extending the time-on-wing, which is accretive to Safran.

For LEAP engines however, CFM operates on a Revenue-per-Flight-Hour (RPFH) model, which is a fixed rate per hour that the aircraft flies, rather than lumpy revenue that CFM collects after each shop visit, with price increases tied to labour/material indices. As a result, the company is now charging for both services and revenue. However, with accounting, the company currently only realises the revenue after each shop visit, though cash flows remain strong with this subscription-like model. Under this LEAP engine segment, GE and Safran operate on a 50/50 basis, allowing them to be official “equals”. This represents material margin upgrade for Safran, once this programme reaches a mature state. For LEAP engines, deliveries have been strong, with about 10,000 units still currently in the backlog through the end of the decade. This implies c. 5 years of OE revenue for the firm. Adding on, currently, the LEAP engine programme is unprofitable due to (1) OE sales at a loss (2) Higher temperature of engines leading to repairs. However, CFM has reduced this through their durability kits, and has started recognising profits in early 2026.

There are some risks associated with the RPFH model, such as (1) mispricing of long-term maintenance costs (2) inflation eating into contract pricing and (3) airlines flying the aircraft harder than usual. However, Safran has mitigated this, through recognising “performance warranties” and index-linked adjustments that are tied to rising labour and material costs. Lastly, Safran offsets the risk of “hard flying” by tracking real time data via EngineLife Connect, adjusting cost provisions.

For an aircraft’s entire lifetime, it typically requires 3 shop visits, 6-9 years, 12-15 years and 17-19 years following the initial aircraft sale (our regression model supports this). Typically, the 1st and the 2nd shop visits are the most costly for an aircraft owner, due to heavy need for LLPs. Because these revenues are only realised during the shop visits, this sort of leads to lumpy revenues for CFM, tied to

deliveries of the past. With the current CFM average age of around 12-13 years, most of them are entering the 2nd shop visit phase. Currently, however, because LEAP engines are still relatively young (4-5 years), the company has yet to be able to realise high margin spare parts and services revenue.

Currently, about ⅔ of aircrafts operate on CFM engines, with ⅓ operating on the relatively newer LEAP engines. This, along with the lack of profitability for the LEAP engine programme, has resulted in most of Safran’s profits being derived from the CFM56 programme.

Within the CFM engine business, revenue is a function of (1) shop visit count (2) shop visit mix (3) pricing and (4) spare parts content. Shop visit count is driven by a few factors - past deliveries, retirement rate of existing aircraft and flight hours. Higher utilisation, lower retirement will further push up shop count. Low retirement rates support fleet size and increase the proportion of ageing engines, which require more frequent and higher value shop visits. With average CFM engine age at ~12–13 years, a growing number of engines are entering their second shop visit, which typically involves high value LLP replacement. Pricing has also been relatively strong, signalled by revenue consistently outpacing volume growth. This is due to their proprietary parts, regulations and high switching costs. Spare parts content is driven by engine age and shop visit mix, with heavy replacement and wear and tear supporting spare parts content increase.

For the LEAP engine business, revenue is a function of (1) flight hours and (2) installed base. Revenue passenger kilometres (RPK) is projected to grow at 3.6% (per Airbus), implying ~3% flight hour growth. Installed base in the narrowbody segment that CFM operates in, is also projected to grow at 3.1% (per Boeing). With ~3-4% pricing, this suggests a steady ~10% growth in this segment of the business. Pricing will likely remain, given the benefit relative to the cost of losing future engine orders. OEMs like CFM control the entire chain, from pricing to maintain a floor for MROs, to licensing and manuals for each equipment.

Equipment & Defense

This segment can be broken up into two parts: Aircraft Equipment and Defense Electronics.

In Aircraft equipment, Safran is dominant in landing systems, wheels and brakes. Safran is current #1 in both landing systems and wheels and brakes, with its equipment embedded in next-gen aircraft such as B787, A320 and A350, and a 54% market share. In Landing Gear, the company operates in a duopoly with Collins Aerospace (RTX), and in brakes, Safran has benefited from a structural shift from steel brakes to carbon brakes.

Wheels and brakes benefit from high replacement frequency and are safety-critical components, resulting in recurring aftermarket revenue with strong visibility. Pricing is supported by certification barriers and the critical nature of the products, as well as availability. Similar to engines, profitability leans toward the aftermarket, where margins are structurally higher than OE sales. Landing gear requires full overhaul every 10-12 years, while Carbon brakes require replacement every 6-18 months. Growth in this portion is primarily driven by flight hours which directly impact replacement rates, as well as underlying installed base expansion. Increased flight usage associated with narrowbody aircraft share gains will be accretive to Safran.

Safran is building a new carbon brake plant in France, scheduled to open in 2030, to support steel to carbon brake transition. They are targeting a 25% increase in overall production volume by 2037, supported by existing sites in France, the US, and Malaysia. Carbon brakes are lighter, allowing for lower fuel usage, reducing carbon emissions, incentivising airlines to switch. Currently, a meaningful portion of the global narrowbody fleet still operates steel brakes (particularly older 737NGs, A320ceos in emerging markets), and upgrading is an untapped revenue opportunity for Safran at every overhaul cycle.

In Defense Electronics, Safran is a tier-1 French defense supplier but a tier-2 global one. Safran provides Optronics & Sights, Navigation Systems, and Tactical Weapons (Missiles). This portion is often driven by long term government contracts often between 15-30 years, supported by government defense budgets. The aftermarket revenue can be 60-70% of total programme lifetime value, but it comes later and is less visible at initial contract award. Margins on production contracts are typically in the 8-12% operating range; aftermarket margins can reach 15-20%.

With structurally increasing defense budgets, tied to heightened geopolitical tensions particularly in EMEA, this supports continued growth in this portion. NATO, for instance, is targeting defense spending to reach 2% of GDP. Compared to commercial aerospace, the defense portion is less cyclical and provides greater revenue stability, but it has generally lower margins than primarily aftermarket driven segments.

Aircraft Interiors

Under Aircraft Interiors, Safran provides Seats and Cabins. They have recently divested their Inflight Entertainment business after years of unprofitability. Currently, this segment provides <1% of total company EBIT, with historically lower margins as compared to the other segments. It is currently undergoing its turnaround, with margins expected to meaningfully inflect from here, and management targeting for a high-single digit operating margin profile, compared to the 3.2% currently. This segment removes a drag on profitability, but does not drive the core thesis.

Safran’s Moat

The moat of Safran is driven by certification lock-in, incumbency, and high switching costs, which create decades of annuity-like revenue from the aftermarket. Within Narrowbody aircraft, CFM operates in a duopoly with Pratt & Whitney, with CFM holding a dominant 60% market share. Once embedded into an aircraft platform, Safran benefits from structurally higher margin AM revenue over ~30 years, turning initial OE contract wins into long term profit for the firm. Certification, for instance, is a multi-year process, often about 5 years. This, along with engine-aircraft integration, increases the switching cost and limits switching after initial integration.

Engines are the most critical part of the entire aircraft. As such, PMAs cannot be used for critical LLPs. To add on, USMs are currently limited given the low retirement of aircraft. This allows for Safran to exert strong pricing power on the proprietary and patented parts that they provide. As shown below, the cost of materials alone often exceeds the cost of labour.

To add on, Safran (CFM) has the ability to control engine production, and hence they have indirect control over retirement rates. For instance, despite Airbus’ attempts to push for market share increases from CFM, CFM has maintained their position with the refusal to increase market share above 60%. By controlling levels of production, CFM has the indirect ability to reduce aircraft retirements, “forcing” airlines to fly their planes longer, and hence undergo more shop visits.

Competition

Pratt & Whitney (P&W): Pratt & Whitney (a subsidiary owned by RTX) serves as CFM’s main competitor within the narrowbody engine space. When it comes to the choice of an aircraft engine, fleet owners typically consider the following: (1) Reliability (2) Fuel efficiency (3) Maintenance cost and (4) Time on Wing. For reliability, CFM’s LEAP and CFM56 engines have been much more reliable than P&W’s GTF engines over the recent years, with GTF engines’ powder metal defect resulting in grounding of fleets since July 2023. This has pushed aircraft OEMs such as Airbus to increasingly adopt CFM’s engines, as opposed to GTFs. While GTF engines are statistically more fuel efficient, the grounding of their engines in recent years has suppressed this advantage. Maintenance costs for GTF engines have also been materially higher following the powder defect incident. Lastly, Time on Wing ties into reliability, and is arguably the most important factor of the four. Delays and aircraft grounding will cost an airline millions in lost revenue. With CFM’s introduction of durability kits, compounded by P&W’s GTF Engine defects, CFM has achieved significantly higher TOW than P&W, with meaningfully lower grounding rates.

Rolls Royce: Rolls Royce currently operates primarily in the Widebody aircraft segment, with the company looking for an entry into the narrowbody business in early 2030s, with the company seeking for UK funding. Rolls Royce is looking for a JV with another engine OE to increase the scale of production, and has claimed that their engines are 20-30% more fuel efficient than CFM’s. However, CFM’s advantage here is in their predictability and reliability of engines. Following the Trent1000 engine groundings, aircraft OEMs will likely have to weigh between reliability and potential revolutionary cost savings provided by Rolls Royce. Besides, Rolls Royce will only be a credible competitor in the 2030s. The disruption by Rolls Royce may introduce further competition which limits pricing power on the OEMs, but the key distinction to note here is that Engine OEs already have limited pricing power on the OEMs front, with most of its earnings derived from the aftermarket. As such, the impact of disruption will likely be minimal.

Independent MROs: Independent MROs serve as a potential competitor to CFM, through the provision of labour services on the engines. Independent MROs may also source for USMs or PMAs to replace original CFM parts to reduce costs. However, there are a few limiting factors against independent MROs. (1) While the 2026 IATA-CFM agreement till 2033 provides independent MROs access to standard manuals, it does not grant them proprietary manuals for advanced repairs. Critical components, such as the Ceramic Matrix Composites (CMCs) and fan blades in engines still require processes that CFM does not share with non-licensed shops. As such, independent MROs cannot work on high margin hot-sections that CFM and licensed MROs provide. (2) With PMAs, aircraft owners are typically against the use of PMAs, as it reduces the resale value of the aircraft, with most lease contracts restricting PMA usage. (3) USMs are typically more favourable to aircraft owners, because these are OEM parts. However, with retirement rates at its current historical low levels, there are limited engines to provide USM parts. (4) CFM’s introduction of Rate-per-Flight-Hour (RPFH) contracts essentially obligates airlines to use CFM’s services, because switching away to independent MROs will result in additional costs, given that they have already paid for the maintenance.

Under Safran’s Equipment & Defense business, the company faces meaningful competition from PMA providers such as HEICO and Trasdigm. For instance, these companies provide brakes, wheels and landing gears that are significantly cheaper than the original list price. Lower costs incentivises low-cost carriers (LCCs) to switch to PMAs, although Safran has also been pushing back with similar “Service-by-the-landing” contracts, which works the same way as the RPFH contracts, to remove the threats of PMAs, and to ensure companies stick with the original OEMs. Air Lessors also remain a significant deterrent against PMAs to preserve the resale value of the aircraft.

Thesis

Our thesis surrounds the fact that we believe CFM56 engine shop visits guided by management are currently conservative, despite management’s increase in guidance.

Management has recently, in the Q4 Earnings Call for FY25, raised their guidance for CFM shop visits, with an additional 750 shop visits over 3 years, leading to an additional ~250 shop visits per year with an average of about 2350 shop visits each year, above their Capital Markets 2024 guidance. However, looking back at management’s forecast over the past 5 years, they have always been conservative on shop visit numbers. For instance GE raised their guidance multiple times throughout 2025, and Safran’s management only guided up in Feb 2026. Management has also substantiated this point on conservatism in their Q1 earnings call.

On shop visit numbers, Safran’s management has expressed that CFM shop visit numbers will likely peak in 2028, and decline thereafter. GE’s management team has also guided for about 2,000 shop visits in July 2025.

However, we believe that management may be still underestimating the duration of the CFM56 aftermarket cycle, because of industry wide supply constraints and increased utilisation rates of older narrowbody fleets. Utilisation rate has also remained high with a 3% increase in 2025, while retirement rates have remained low at 1%. Management is projecting for retirement rates to remain at 1-2% for FY26, and to tick up to 3-4% over the next few years. We believe that CFM has the ability to control this retirement rate, as bottlenecks in production primarily due to engine production have resulted in lower retirement rates for aircraft, with the average aircraft age rising from from 11.1 years in 2019 to 15.1 years old in 2026, a 4-year increase. Lower retirement rates along with an ageing fleet means that aircraft will have to fly their plane for longer periods of time, leading to not only more shop visits, but also reduced USMs to provide for maintenance of engines. As a result, airlines are now required to use original parts due to lack of USMs, especially for older engines. This is accretive to revenue per shop visit numbers for CFM. CFM has the unique ability to balance short term profitability with long term delivery and profits for the company.

A good proxy for shop visits can be determined by looking back at delivery numbers for CFM engines from the past, and applying a relevant Shop Visit Rate (SVR). In the case of CFM, there was a large pool of deliveries from 2005 to 2017, with 2013-2017 delivery numbers averaging at about 1,300 deliveries each year. With the second shop visit at about 12-14 years after the initial Entry into Service (EIS), we expect Safran’s CFM shop visits to remain structurally supported till early 2030s, contrary to management’s guidance. We therefore expect a more extended and flatter plateau, rather than a sharp peak and decline.

What could break this thesis: However, while retirements remain subdued due to supply constraints, retirement rates can also increase if these scenarios occur: (1) fuel price increases (2) LEAP delivery increases (3) Airbus/Boeing delivery bottleneck eases. A recovery in aircraft deliveries, increased LEAP deliveries, or fuel price increases could accelerate retirements, reducing the installed base of CFM56 engines.

Oil price surges result in reduction in flying hours due to higher ticket costs for travelling. This results in lower potential shop visits for CFM. However, there is a balancing act to this. If oil prices are expected to remain structurally high, airlines and aircraft owners will make sure that their aircraft is of the highest fuel efficiency, and push forward shop visits to maintain the fuel efficiency to save on fuel cost. The pulling forward of shop visits also leaves more time for engine components to degrade. And so, the earlier shop visit arguably creates potential economic value for a fourth shop visit.

Supply chain easing could hurt CFM’s shop visit counts and aftermarket revenue, as retirement rates will likely start to pick up when delivery bottleneck eases. However, there is a balancing act to this. Easing of shop visits introduces more LEAP engines into service, accelerating the LEAP engine aftermarket ramp up. CFM has a monopoly in B737 aircraft, and a 60% share in A320 generation. This gives them the ability to control the number of engines that they deliver to find a balance between near term profitability and long term aftermarket ramp up for the LEAP engines. For Air Lessors and operators in general, the cost of maintenance is often lower than replacing the entire aircraft, exacerbated by long wait times tied to long backlog for Airbus’ and Boeing’s aircrafts.

Risks

Wars and economic recessions can hurt Safran’s revenue. Firstly, with LEAP engines now tied to RPFH contracts, the decrease in flight activity will lead to a decrease in flight-hour tied revenue for Safran. Besides, reduced flight activity also leads to deferred shop visit counts, which can temporarily hurt revenue. During the covid pandemic for instance, Safran’s Propulsion business fell by 36.4%, with Civil Aftermarket falling 43.2% due to deferred shop visits. Equipment & Defense fell by 25.0% largely due to lowered demand for landing gear and carbon brakes, despite the more resilient government defense business.

Systemic engine issues and design flaws can lead to grounding of aircraft, affecting flight hours and increasing the costs that Safran will have to bear. Competitors in RTX (P&W metal defect) and Rolls Royce (Trent 1000 cracking of turbine blades) have experienced major grounding in recent years. Over the past 30-40 years, CFM has not recorded any major design flaw that led to grounding of fleets, with the only major incident that happened in the 1980s. The company is mitigating this risk by introducing durability kits to increase time-on-wing, with software solutions (Monitoring Health Premium, EngineLife Connect) now embedded in the aircraft that are transmitted to Safran’s servers, allowing them to spot defects before major grounding incidents.

Supply chain disruptions can hamper deliveries for products and affect relationships with existing aircraft OEMs. For instance, in 2024, CFM delivered 1,407 LEAP engines, a drop from 1,570 in 2023. The decline occurred due to HPT blade supply constraints on the LEAP-1A and reduced Boeing 737 MAX production affecting the LEAP-1B. Safran and CFM are mitigating this, with CFM having dual-sourced components for 80% of LEAP parts to reduce supply chain vulnerability.

A potential misalignment between GE and Safran could lead to break up in the joint-venture between the two companies, however, the likelihood of this is relatively low. For one, the joint-venture has been around for 52 years, with the current expiration of the JV set to be in 2050. Because the companies’ supply chains and intellectual property are so deeply intertwined, our belief is that a split between the two companies is unlikely.

LEAP engines have faced durability challenges since entering in 2016, with HPT blade degradation, and other dust ingestion issues requiring engineering interventions. While Safran has responded with durability kits for both LEAP engine models, there are risks for further technical challenges along the way. Each new technical durability issue will add to contingency provisions rather than reducing them, weighing on EBIT and margins. The key here is to examine whether or not the H1 2026 LEAP-1B kit rollout represents a genuine resolution to the issues faced.

Rate Per Flight Hour contracts, priced at inception based on modelled shop visit frequency, parts costs, and time on wing assumptions, expose Safran to significant margin risk if real-world LEAP performance deviates from original pricing assumptions. If shop visits arrive earlier and cost more than the contract revenue covers, Safran absorbs all the cost. More recent contracts have been repriced to reflect real-world experience, but legacy contracts may represent a margin drag for the company.

A consistent challenge in checking our shop visit thesis is determining how much of the current demand is structural versus cyclical. If a strong portion of the elevated shop visit volumes is cyclical, then a recovery in Airbus and Boeing production rates could trigger a faster than expected decline in CFM56 shop visits well before the early 2030s. While it is in our belief that CFM has the ability to control retirement rates, investors should also consider external factors that could affect retirement rates, such as SAF adoption, P&W’s delivery numbers, Rolls Royce’s entry into narrowbody.

After LEAP (RISE program)

RISE Engines will succeed LEAP Engines likely in the mid to end 2030s, with the company pushing for an Open Fan Model (Plan A), and an advanced ducted engine model (Plan B) with nacelles.

Valuation

Our model assumes strong pricing power for proprietary parts at levels above inflation, with bear and bull cases providing sensitivity in both directions. Growth in the other businesses - equipment & defense and aircraft interiors - are modelled through general industry growth trends and extrapolation from historical trends.

Base Case:

Our base case assumes a fair value of $245 with the Gordon Growth model, and $313 with a 20x EV/EBIT multiple applied. The 20x EV/EBIT represents a 10% discount to the lowest EV/EBIT the company has been rated at over its recent history. With working capital changes, Safran’s FCF can be very lumpy, as such, more weight should be placed on the terminal multiple method. With this premise, our fair value for Safran stands at $313, with an implied 11.1% IRR at today’s $280 price.

Bear Case:

Bull Case:

Google

One word that I have for this earnings report by google: WOWW

Alphabet delivered an exceptional Q1 FY2026, with consolidated revenues of $109.9 billion which is a 22% year-over-year increase and the company’s 11th consecutive quarter of double-digit growth. The standout performer was Google Cloud, which reported revenues of $20.0 billion, surpassing the $20 billion threshold for the first time, representing 63% growth year-over-year and a meaningful acceleration from prior quarters. Net income surged 81% to $62.6 billion (inclusive of a $28.7B one-time equity gain likely from anthropic and space x), while operating income grew a clean 30% to $39.7 billion, with operating margin expanding 210 basis points to 36.1%.

The headline investment thesis is centred on Google Cloud’s $462 billion backlog ,which nearly doubled quarter-on-quarter, with management confirming that just over 50% is expected to convert to revenue within 24 months. I will analyse the Q1 results in detail, synthesises key themes from the earnings call and build a forward revenue forecast for Google Cloud through 2026 and 2027 using management’s own conversion guidance.

KPIs

Total Revenue: $109.9B (+22% YoY) vs. consensus $107.1B — beat by ~$2.8B

Diluted EPS: $5.11 (+82% YoY) vs. consensus $2.64 — significant beat (includes equity gain)

Google Cloud Revenue: $20.0B (+63% YoY) — first time exceeding $20B

Google Cloud Operating Margin: 32.9% (vs. 17.8% in Q1 2025) — remarkable expansion

Cloud Backlog: ~$462B, nearly doubling sequentially; 50%+ to convert in 24 months

Google Search & Other: $60.4B (+19% YoY) — AI Overviews now at 1.5B monthly users

Gemini Enterprise: Paid MAUs +40% QoQ; Enterprise AI now primary Cloud growth driver

Share Buyback: $70B authorised; Dividend raised 5% to $0.22/quarter

2026 CapEx Guidance Updated: $180–190B; 2027 CapEx to ‘significantly increase’

Segment Analysis

Google Search & Other (+19% YoY)

Search revenue of $60.4 billion came in above expectations, with Pichai citing queries at an all-time high. The key driver is the successful integration of AI Overviews — now reaching 1.5 billion monthly users — and the rapid global rollout of AI Mode. Critically, Alphabet has reduced the cost of core AI responses by more than 30% since migrating to Gemini 3, improving the unit economics of AI-enhanced search substantially. My initial worry about Search has been allayed. I initially was worried that higher search costs would lead to a margin compression in the business as it would be tough for Google to raise prices proportionate to the higher costs that google has to inccurr. However, google raises the CPC by 12%, increased Average Click-Through Rate (CTR) by 3.5% and volumes by 7%. An impressive growth in both pricing and volume

Google Cloud (+63% YoY)

Google Cloud delivered its strongest quarter ever. For the first time in the company’s history, enterprise AI solutions overtook traditional cloud infrastructure as the primary growth driver. Revenue from products built on Gemini models grew nearly 800% year-over-year. New customer acquisition doubled compared to the same period last year, and existing customers outpaced their initial commitments by 45%. The number of $100M–$1B deals doubled, and multiple billion-dollar-plus contracts were signed. Google Cloud is now positioned as a full-stack AI provider the only hyperscaler offering first-party solutions across the entire enterprise AI stack. Google now has the Largest cloud backlog excluding Open-AI of $467.6 billion which management is expected to fulfil 50% of it during the next 2 years.

A few notable questions that were asked in the earnings call that I penciled down:

Q: How is constrained compute affecting Search, and which AI experiences will benefit most from next-gen compute in the next 12 months?

A: Pichai acknowledged compute constraints but noted search queries are at all-time highs. He highlighted agentic search workflows and AI Mode as the key next beneficiaries of added compute capacity. AI Overviews, now at 1.5B monthly users, and AI Mode are driving incremental usage.

Q: Can management provide more detail on how the $462B backlog will convert to revenue, specifically the split between GCP contracts and TPU hardware agreements?

A: CFO Anat Ashkenazi confirmed just over 50% of the $462B backlog is expected to convert to revenue over the next 24 months. TPU hardware agreements are included in the backlog, though the majority of backlog remains traditional GCP contracts. A small percentage of TPU revenues will hit in 2026, with the majority in 2027.

Q: Management guided 2027 CapEx to significantly increase vs 2026’s $180–190B range — can you quantify, and how does ROIC evolve given this spend trajectory?

A: Ashkenazi declined to quantify the 2027 figure but confirmed the direction and linked it directly to the demand signal from the $462B backlog. She pointed to Google Cloud's operating margin reaching 32.9% as proof of capital efficiency improving alongside spend.

My thoughts: As mentioned in an earlier writeup on google, the best investments in the world occur when companies are able to deliver stellar ROIC with high reinvestment rates. Google under Sundar has clearly showcased the ability to generate growth from the capital expenditures that the company is deploying. At this point given his track record across multiple acquisitions as well as the numbers that he manages to deliver, I am not worried at all about the amount of capital expenditures that he is deploying as Alphabet is clearly seeing a ROI from the investments.

Cloud is becoming a larger % of the profits of google and will be the key driver of value of google for the next few years. I will try to guestimate the cloud revenue for FY 26 and FY27.

Here are a few key points that management has pointed out in the call:

• Total Cloud Backlog (end Q1 2026): ~$462 billion

• Management-guided conversion rate: just over 50% in 24 months

• Implied 24-month convertible backlog: ~$231–240 billion

• This implies an average of ~$115–120 billion per year from backlog conversion alone over FY2026–FY2027

Google Cloud entered 2026 at a $80B annualised revenue run-rate (based on Q1’s $20.0B). For full-year 2026, we must account for: (i) ongoing organic growth from GCP infrastructure and AI solutions, (ii) a modest contribution from TPU hardware sales beginning in H2 2026, and (iii) the Wiz acquisition integration drag (low single-digit margin headwind, limited revenue contribution near-term).

The Bull Case will be approximately 65–70% YoY growth which represents $73–78B in revenue assuming continued acceleration, capacity additions to meet demand and TPU hardware deliveries begin meaningfully in Q3/Q4 2026.

Base Case (55% YoY growth): ~$68B assuming some moderation in growth rate as lapping becomes harder; capacity partially constrained (management stated cloud revenue would have been higher if compute weren’t constrained).

Bear Case (40% YoY growth): $58–60B assuming material compute bottlenecks persist through 2026 and macro uncertainty delays enterprise deal closures (unlikely given google’s full stack approach to AI)

Now moving on to 2027 Cloud revenue forecasts. Management has explicitly guided that: (a) the vast majority of TPU hardware revenues will be recognised in 2027, and (b) 2027 CapEx will ‘significantly increase’ relative to 2026’s $180–190B signalling management’s own confidence in demand conversion. The 24-month backlog window covers through Q1 2028, meaning virtually all of the $231B+ convertible backlog accretes into 2026 and 2027.

Bull Case: $115–125B assumes full backlog conversion on schedule, strong new deal activity in 2026 adds to 2027 backlog, and TPU hardware revenues are substantial (e.g., $10–15B).

Base Case: $100–110B assumes ~50% of the convertible backlog accretes into 2027, moderate new contract wins, and TPU hardware contributes $5–8B.

Bear Case: $88–98B which assumes execution risk on hardware deliveries, capacity constraints delay backlog conversion, or enterprise customers renegotiate contract terms.

Even with bullish projections for GCP, the strength of google’s earnings will still be highly dependant on the core search engine business. I will still be keeping an eye out on any potential developments that could impact the core ads business and also keep an eye out on advertisers ROAS on google ads. As long as the ROAS for google ads remain high, google will be able to raise prices and attract ad volume. Overall this earnings report shows that google is successfully monetising on AI and that the capital expenditures up to this point are justified. Nonetheless, the valuation of the company is now undoubtedly extended and I personally would not be comfortable entering a position in google. Google has gone back to being 18% of my overall portfolio after I trimmed 15% a quarter ago and I might trim it yet again to reduce exposure. Nonetheless, this was a terrific quarter by google and has surpassed all expectations.

Mastercard

Mastercard delivered a strong Q1 FY2026, reporting net revenue of $8.4 billion, a 16% reported increase year-over-year (12% on a currency-neutral basis), driven by continued global payment volume expansion and accelerating demand for its Value-Added Services and Solutions (VAS) segment. Net income rose 18% to $3.9 billion, and diluted EPS of $4.35 increased 21%, aided by $4.0 billion in share repurchases during the quarter. On an adjusted basis, EPS of $4.60 grew 23%.

The quarter reinforces Mastercard’s durable, high-margin business model while surfacing two key themes for investors: (1) a geopolitical headwind from the Middle East conflict that is weighing on cross-border travel volumes into Q2, and (2) accelerating strategic momentum in agentic commerce, stablecoin infrastructure, and AI-powered security solutions positioning Mastercard for the next wave of digital payments. The planned acquisition of BVNK ($1.5–1.8B) is the most significant M&A signal of the company’s digital asset ambitions.

I own both Visa and Mastercard in my portfolio and I have mentioned in previous writeups that I never once understood why Mastercard was trading at a premium compared to Visa. Granted, nastercard has had a history of higher fundemental growth compared to Visa, however, this was driven mainly by the fact that MA has a higher proportion of their revenue coming from VAS. V has a smaller VAS division compared to mastercard simly due to the fact that management has been prioritising its focus on the core payments processing business. However this shifted the last few quarters where management signalled to shareholders that they were going to tap into the VAS business. Visa’s VAS business has undoubtedly a higher TAM compared to Mastercards. We have to first understand what VAS is and why it is so valuable. Vas primarily serves the fuction of fraud prevention via Advanced Authorization, tokenization, and biometric authentication;Consulting for banks (e.g., how to launch a crypto card) and data analytics for merchants as well as help Open Banking & New Flows. While Mastercard has historically grown its services segment faster as a percentage of total revenue, several structural factors suggest that Visa has a larger Total Addressable Market (TAM) for these services in the long run due to the following factors:

1) Visa is simply bigger. It handles roughly 60% of global card volume compared to Mastercard’s ~30%. VAS products like fraud detection and tokenization are proportionate to the size of the network. Because Visa has nearly double the transactions and cards in circulation, its "surface area" for selling security and data services is inherently twice as large.

2) Visa has been more aggressive in expanding the “endpoints” where its services can reach. By Q1 2026, Visa Direct reached 18 billion endpoints (bank accounts, digital wallets, and cards). Mastercard however only reaches ~7 billion endpoints. Coupled with the fact that most of Visa’s endpoints come from firms located in developed countries such as the US, UK, EU where banks and merchants operate in hyper-competitive, saturated environments where a 0.1% improvement in conversion or a small reduction in churn is worth millions thereby increasing the companies willingness to pay for VAS solutions. Furthermore there is regulation and compliance in these developed countries such as PCI-DSS and PSD3 make advanced fraud scoring (VAS) a mandatory line item for institutions which enables Visa to bake its VAS into the standard infrastructure of the bank. It is much easier to sell an expensive analytics suite to a Tier-1 US bank that already processes billions through your network than to a fragmented set of smaller players which is the issue that Mastercard is facing.

3) Developed market endpoints generate richer data (higher average transaction values, more diverse merchant categories). Visa’s Gemini-integrated AI models (which are driving their current growth) learn faster on the high-velocity, high-complexity data of the US consumer. This creates a "flywheel": Better data from developed endpoints will lead to better VAS products thereby leading to higher prices and a higher TAM.

Well it is highly likely that over the next couple of years in the short term that Mastercard will be able to grow faster than Visa, I believe that it is foolish to allocate a higher multiple on mastercard simply due to this fact. As a long term investor what I am purchasing is a business. I could care less of the growth rate of the business in the short term and I am more interested in the TAM of the business that I am purchasing as well as the sustainability of the growth rates decades into the future. Purchasing Mastercard at a higher valuation simply does not make sense to me as the key driver of future returns in the business comes from the VAS segment of the business and not the payment processing side of the business which has arguably reached close to the TAM and is unlikely to grow faster than GDP growth+Inflation+1-2% cash to credit conversion in the long run. The VAS business is estimated to only be 2% of the TAM and Visa is seeking to grow its VAS business significantly which can be seen by a 25&27% growth in VAS rev in Q1&2FY26 respectively as compared to Mastercards 23&22% revenue growth. This represents the first 2 quarters where Visa actually overtook Mastercard in VAS growth. Over the next 10 years, I believe that V will eventuall overtake Mastercard in the VAS market and become the market leader. That being said I own both companies in a ratio of 75% V and 25% mastercard. In the situation where the market assigns an unreasonable premium to mastercard, I will be swapping my Mastercard position for Visa.

Mastercard’s earnings this quarter saw its payment network revenue grew 8% on a currency-neutral basis, driven by healthy domestic volume (+6% CN), strong cross-border volume (+13% local currency), and pricing improvements. The 13% cross-border growth is particularly meaningful given it generates approximately 3–4x the yield of domestic transactions. However, the Capital One debit migration created a modest drag in U.S. debit metrics but the underlying network health is strong with contactless penetration at 78%, up 5 percentage points year-over-year, demonstrating continued digitalisation of in-person payments. Cross-border volume grew 13% on a local currency basis in Q1. Non-US GDV grew 9%, with international credit and debit each growing 9% and 8% respectively. Q1 results were largely robust, but management called out that the Middle East conflict began impacting cross-border travel metrics in March 2026 a trend that has meaningfully accelerated into April and is expected to weigh most heavily on Q2.

VAS is now the fastest-growing and strategically most important segment of Mastercard’s business. At $2.8 billion in Q1 and growing at nearly twice the rate of the core network, VAS is becoming a larger portion of total revenue. Key drivers included security solutions (Recorded Future reached 500+ enterprise customers within 12 months of launch), Ethoca fraud intelligence (+25% YoY), digital and authentication solutions, business and market insights, and consumer acquisition and engagement services. This mix shift toward VAS is structurally positive for margins and revenue quality.

On buybacks: Mastercard repurchased $4.0 billion of stock in Q1 2026 and an additional $1.7 billion through April 27, bringing total buybacks to $5.7 billion in just four months. With $11.7 billion remaining under authorisation, the company is deploying capital with conviction. Buybacks contributed approximately $0.10 to adjusted EPS in the quarter. The dividend payout was $777 million in Q1.

Important Q&A I penned down:

Q: How significant is the impact of the Middle East conflict on cross-border travel volumes, and how has April trended vs. Q1?

A: CEO Miebach acknowledged cross-border travel volumes began to soften in March. CFO Mehra quantified April cross-border travel at +2% YoY (vs. +8% in Q1), while overall cross-border payments decelerated to +9% in April from +13% in Q1. Management guided Q2 revenue to the low end of low double-digit CN growth — but stated that absent the conflict, Q2 would be 'generally in line with Q1.' They expect the headwinds to be 'largest in Q2 and then progressively recover.'

Q: What is the quantified impact of the Capital One debit portfolio migration on U.S. debit GDV and switched transaction growth?

A: Mehra confirmed that excluding the Capital One debit migration, U.S. debit GDV grew 7% (vs. 1% reported) and switched transactions grew 10% (vs. 9% reported). This is a temporary, structural drag that reflects the migration of a major portfolio away from Mastercard rather than any fundamental consumer spending weakness.

Q: Is the 18% CN growth in VAS sustainable, given it's growing 2x faster than the payment network — what are the primary drivers?

A: Mehra confirmed VAS growth was entirely organic in Q1 (no acquisition contribution). Key growth drivers were security solutions (including Recorded Future, now at 500+ customers within 1 year of launch), digital and authentication solutions, business and market insights, and consumer acquisition/engagement services. Ethoca products grew 25% YoY. Management expressed confidence in VAS as a durable, higher-growth layer of the business, with a mix shift towards these services over time.

Q: Given the Q2 softness from Middle East headwinds, is the full-year guidance for high-end of low-double-digit CN revenue growth still achievable?

A: Management maintained FY2026 guidance at the 'high end of low double digits' on a CN basis, implying ~12-13% CN growth for the year despite the Q2 softness. They project headwinds to 'progressively recover' in H2. FX is guided as an approximately 1.5% tailwind to reported revenue for the year. No explicit 2027 guidance was given, but management's tone on structural drivers (VAS, agentic, stablecoins) was constructive.

Amazon

Amazon delivered a landmark Q1 FY2026, reporting net sales of $181.5 billion, a 17% year-over-year increase with operating income of $23.9 billion representing Amazon’s highest-ever operating margin at 13.1%. The quarter was defined by one headline above all others: AWS reaccelerated to 28% growth, its fastest pace in 15 quarters, crossing a $150 billion annualised revenue run rate. CEO Andy Jassy described the quarter as a historic moment in the AI infrastructure buildout, underpinned by a $364 billion AWS backlog (excluding a newly announced $100+ billion Anthropic deal), over $225 billion in Trainium revenue commitments, and AWS AI revenue growing triple digits year-over-year.

Segment analysis

AWS (+28% YoY)

AWS is the axis around which the entire AMZN investment thesis rotates. The acceleration from 24% last quarter to 28% this quarter on a base that is now $150 billion annualised is the single most important data point in the report. AWS operating income of $14.2 billion at a 37.7% margin demonstrates that the business generates exceptional returns even as Amazon invests aggressively in AI infrastructure. The $2 billion sequential revenue increase from Q4 to Q1 was the largest Q4-to-Q1 increase in AWS history. AI revenue within AWS is running at over $15 billion annually and growing triple digits meaning AI is already a significant and fast-growing subset of a business that is itself accelerating.

North America Retail (+12% YoY, 9.0% margin)

North America operating margin expanded from 8.0% to 9.0% year-over-year, a significant achievement given ongoing investment in same-day delivery infrastructure and the roll-out of new robotics automation across fulfilment centres. Unit growth of 15%, the highest since COVID-era lockdowns, demonstrates that the Prime flywheel is accelerating. Amazon is now the second-largest grocer in the US by gross sales ($150B+ in 2025), with Whole Foods expanding to 650+ stores. The combination of faster delivery, broader selection, and AI-powered personalisation (Rufus, Creative Agent) is widening Amazon’s moat in retail. I view the retail segment not as a segment that is chasing profitability but one that locks in the customers into the amazon ecosystem that enables amazon to monetise on its users via advertisements as well as serving as a launchpad for amazon to introduce any new business segments.

International (+19% YoY, 3.6% margin)

International swung from an operating loss of -$0.5 billion in Q1 2025 to +$1.4 billion in Q1 2026 — a $1.9 billion operating income improvement year-over-year. Revenue growth of 19% (11% ex-FX) demonstrates that the international expansion strategy is beginning to pay off. Same-day delivery expansion across European hubs and the early-stage international rollout of Amazon Now (30-minute delivery in India growing at 25% month-over-month) are meaningful signals for further margin improvement ahead.

Advertising A Durable High-Margin Engine Tapping On Existing User Base (+24% YoY)

Advertising services revenue of $17.2 billion grew 24% year-over-year, significantly beating the 21.2% consensus estimate. With trailing twelve-month revenues crossing $70 billion, Amazon Advertising is now one of the largest digital advertising platforms in the world. The expansion of Prime Video’s ad-supported tier, new partnerships with Netflix (via Amazon Audiences) and Comcast Advertising, and the global rollout of the Creative Agent tool are driving incremental inventory and yield improvement. At mid-to-high-20s growth rates and structurally high margins, advertising is Amazon’s fastest-growing high-quality earnings contributor and is growing with virtually no capex investments as it is simply tapping on its existing user base and monetising eyeballs.

Key questions asked during earnings call

Q: What is the current AWS backlog figure and how concentrated is it? Does the Anthropic deal change the picture materially?

A: Jassy confirmed the Q1 backlog is $364 billion, explicitly excluding the recently announced Anthropic deal worth over $100 billion implying total committed pipeline is over $464 billion. He stressed the backlog has 'reasonable breadth' and is not concentrated in one or two customers. He also cited Trainium revenue commitments of over $225 billion from Anthropic, OpenAI, Uber, and others as an additional, separately tracked demand signal.

Q: CapEx hit $43.2B in Q1 — nearly $175B annualised. How does Amazon justify this level of investment given trailing free cash flow of only $1.2B?

A: Jassy acknowledged the front-loaded nature of AWS investment where land, power, buildings, and hardware must be committed '6 to 24 months before billing customers.' He expressed 'high confidence' that 2026 AWS CapEx will be 'monetised well,' backed by customer commitments for a substantial portion. CFO Olsavsky echoed that the investment is producing 'compelling operating margins and ROIC,' pointing to AWS's 37.7% operating margin as validation. Management expects FCF to recover as revenue scales into the CapEx already deployed.

Q: How does Trainium’s economics compare to NVIDIA GPUs, and is Amazon planning to sell Trainium racks externally at scale?

A: Jassy stated Trainium2 offers ~30% better price-performance vs. comparable GPUs and is largely sold out. Trainium3 (30-40% more performant than T2) is nearly fully subscribed despite having just started shipping in early 2026. Trainium4 is still ~18 months from broad availability and is already substantially reserved. The chips business has a $20B run rate (growing triple digits YoY), and Jassy estimated a standalone rate of $50B if all chips produced this year were sold externally. On rack sales: ‘I expect over time there’s a good chance we’re gonna sell racks over the next couple years.’ Trainium is expected to save ‘tens of billions’ in annual CapEx and deliver ‘several hundred basis points’ of margin advantage for inference at scale.

Q: When does Leo generate meaningful revenue, and how should investors think about the $1B incremental Q2 Leo cost as a drag?

A: Olsavsky guided for ~$1B of incremental North America cost in Q2 related to Leo manufacturing and satellite launches. Commercial service is 'on track to launch in a few months' (i.e., Q3 2026). Amazon plans over 20 launches in 2026 and over 30 in 2027. Jassy cited meaningful revenue commitments from Delta, JetBlue, AT&T, Vodafone, Apple (for iPhone/Watch satellite services), and NASA. Leo costs will shift to capitalisation in Q4 2026. Jassy positions Leo as a long-term structural opportunity comparable to AWS's early days where there is a large upfront cost with durable return potential.

Q: With advertising crossing $70B TTM at 24% growth, is there a ceiling — and how does agentic AI change the advertising model?

A: Jassy said advertising will 'do well' in an agentic world, noting that multi-turn agent conversations create multiple natural points to surface sponsored products. Rufus monthly active users grew over 115% YoY, with engagement up nearly 400% — establishing the foundation for agentic shopping monetisation. The 24% advertising growth (vs. 21.2% consensus) reflects continued share gains in sponsored products, streaming ad inventory (Prime Video, expanded Netflix partnership via Amazon Audiences), and the Creative Agent tool expanding globally. Management sees no near-term ceiling.

All in all this was an outstanding report by amazon that showcases improved acceleration in the non retail portion of the business as well as a clear monetisation of the capex that the company is spending. I model that advertising revenue will reach over 150B by 2030 and has the potential to contribute over 75B in OCF which represents over half of the current OCF that the business generates. Amazon is a busines that shines and as the revenue mix shifts towards higher margins business, I expect the bottom line of the business to be able to maintain growth of over 25% over the next 5 years where the revenue maintains at LDD growth.

(1) Visa’s earnings:

Visa had a very strong quarter, with revenues accelerating to 16% YoY on a constant currency basis, and adjusted EPS up 20% YoY. VAS accelerated to 27%, showing strong growth, though some of the outperformance was due to timing and adjustments. In the case of Visa, I believe that the markets have not been pricing in VAS numbers, and now that the numbers are showing up, the stock is rallying with it. In management’s view, VAS is still relatively untapped, with the company currently only capturing 2% of TAM.

In terms of management remarks, there were a few critical questions answered:

1. On WERO and Digital Euro being implemented, along with other payment alternatives outside of Debit and Credit, Visa’s management asserted that alternatives have long been around. In my view, it’s down to consumer behaviour, trust, and habits. It is really hard to displace a company like Visa, and payment alternatives have had much trouble doing so.

2. On Agentic Commerce, the company is acknowledging this as a tailwind for the business, because agentic commerce means more tokenisation usage, allowing for safer and higher take rate for Visa. This was also mentioned in our Visa investment thesis.

3. Visa’s positioning: Visa is currently positioning itself increasingly as a fraud deterrent, with management acknowledging that Fraud protection is a key concern for most commerce companies. It is riding on the wave of digitalisation and technology evolution, providing security by leveraging on their brand and trust effect.

Overall, Visa’s earnings were surprisingly great, despite the Middle East tensions. With the FIFA World Cup coming up, we could see further acceleration in transaction volumes for Visa in the coming quarters. I remain bullish on the company.

(2) Microsoft

Microsoft’s earnings is strong evidence of a company that is doing well fundamentally, but falling in stock price due to heightened expectations, or perhaps heightened perception of risk. The company saw great numbers all across the board, with Productivity and Business Processes (M365, LinkedIn, Dynamic 365) up 17%, and Intelligent Cloud up 30% in revenue, with Azure up 40%, yet the stock still fell.

Productivity and Business Processes revenue came in strong, with the company introducing price hikes across its products, evidence of strong pricing despite hypothetical AI risks. My feeling is that the markets are in a way, neglecting this number here. In my view, and in Satya’s view, the markets are too concerned with one singular number - the Azure number.

There are a few plausible reasons for the fall in stock price: (1) Shrinking margins, (2) Capex guide or (3) lack of revenue acceleration. Margins in Intelligent Cloud fell, primarily due to increased depreciation associated with Capex spend.

On margins, Management was asked a few tricky and critical questions on how AI protects and possibly expands margins, despite the high Capex and high depreciation associated with it. Management’s response was that the AI Margins currently are much higher than Cloud margins during the previous Cloud buildout cycle in the 2010s. In other words, Management is saying that (1) AI Margins are accretive compared to previous Cloud buildout margins (2) “We have seen this before, and we are doing better than previously.” Management expects operating margins in FY2026 to be up slightly, signalling their believe in margin accretion.

Management guided for $190B in Capex spend, with about $25B due to increased costs of equipments linked to shortages. The markets punishing the stock is clear evidence of heightened expectations with increased spend, despite strong revenue numbers. In the case of Microsoft, the company has yet to show revenue acceleration despite increased Capex numbers, primarily due to a weakening Personal Compute business, making up 20% of total revenue. An analyst flagged this out to management during the call, asking for their thoughts on the disparity between revenue growth and Capex growth. Management responded that Azure Revenue was strong at 40% YoY, with backlog growth being a clear indicator of ROI on Capex.

Another key insight in during the call was their shaping of Microsoft’s business model. Management mentioned that Microsoft’s business model should be seen increasingly as a per-seat + per-consumption (usage-based) business model, rather than purely seat-based. The management believes that there is real opportunity for TAM expansion through a usage-based model. On this point, I view Anthropic’s change in charging methods as a clear indicator of this shift. Anthropic, earlier last month, shifted their subscription-based model to a usage-based model, increasing cost for heavy users. This is indication of strong demand, and that a usage-based model is probably more accretive to the company, and protects margins. For Microsoft, this shift is a good move by management.

Last point, on Azure revenue, management mentioned during the call that they expect modest acceleration in the second half of this year. It is interesting that the markets are not reacting to this, given their obsession with this number. In short, my view is that the markets were expecting too much and making active comparisons with Google (which saw Cloud revenue accelerating to 63%), and overreacting despite strong numbers.

(3) Meta

Similar theme to Microsoft, Meta’s earnings were great, with income from operations up 30%. The stock fell primarily due to increased Capex forecast to $125 billion and $145 billion, compared with its prior range of $115 billion to $135 billion, associated with higher component spend. For Meta’s case, there is clear and actual evidence of revenue acceleration.

“Unlike Alphabet, Microsoft and Amazon, which all have massive cloud infrastructure businesses that enable them to turn their AI investments into revenue, Meta has no such offering, making it harder to prove it can deliver returns.” This is essentially the narrative in the markets. The markets are skeptical about Meta’s need for such high Capex without a tangible cloud business model proving strong numbers, despite strong revenue growth figures. My feeling is that most market participants are unclear about Meta’s goal with their Capex spend, brushing it off as “pointless”.

Meta’s Capex spend is primarily to train its internal models (Deep Learning Recommendation Model) to suggest better content, better ads, and hence improve ROAS for advertisers. Ad Revenue is a function of (1) Ad Load (2) Fill Rate (3) Ad Pricing. Ad Load refers to number of ad placements on a platform (e.g. Google has banner ads, video ads, in banner ads). Ad Pricing, is driven primarily by ROAS (Return on Ad Spend). Higher ROAS drives a feedback loop, where advertisers bid for higher ad prices, and so on. When an advertising company is able to suggest ads, at the right time, to a better target customer, ROAS increases, driving Ad pricing.

Comparing Meta’s ad pricing with Google’s ad pricing, Meta Ads have shown significant comparative strength, with ad pricing up 12%. While one can argue that Meta is basically the economy on leverage (with most of the business tied to ads), it is undeniable that Meta’s Capex is working, and has been working. And so, I find the narrative about Meta not having a cloud business rather perplexing. Until there is a clear indication of Meta’s overbuilding capacity, like selling APIs at discounted prices, Meta’s Capex buildout, in my view, is working.

(4) FICO

Fair Isaac is a company that is on the chopping block now, with sentiments that are terrible, and with the market within the crosshairs of regulations. The stock rallied +11% following the earnings release, only for the stock to fall back down to +3% by market open, and -5% the next day following Steve Eisman’s CNBC interview on his FICO short position. Funny enough, he actually disclosed the short position a few days ago, before the CNBC interview.

Revenue accelerated, though in my view, it was expected. The management hiked prices for FICO Classic scores last year in the mortgage division, effectively raising prices by +100%.

There were a few insights that I would like to highlight from the earnings call:

(1) Score pricing changes: Management mentioned that they will be changing FICO 10T Scores pricing when implemented, from $4.95 + $33 to $0.99 + $65 once implemented. FICO is offering free usage for FICO 10T after each purchase of FICO Classic until it is approved by the FHFA.

(2) VantageScore’s market share: Lansing mentioned that VantageScore’s market share is 2%, with a 0% market share in non-conforming markets. VantageScore hasn’t been used in an actual cycle, and lacks credibility.

(3) LLPA Grids and FICO 10T Approval: These are in regulatory control, meaning only the GSEs and FHFA have complete control over these.

Listening to the call, my feeling is that Lansing became a little annoyed towards the end of the call. In his call, he implied: (1) We don’t know what is going to happen (2) VantageScore has no market share. While the stock is priced a lot more appropriately at today’s valuation, the increased regulatory scrutiny and focus on management’s actions are weighing down the stock price. Adding on, management’s latest call, despite the claims of VantageScore’s market share, is not giving investors a lot more confidence.

FICO’s change in pricing structure for their FICO 10T Score is a positive, though the company has probably shed off a lot of goodwill with increased public scrutiny. I remain cautious of the company, despite strong earnings.

By Guan Rui

Introduction

Honeywell’s story began in 1885 when Albert Butz patented a furnace regulator to control indoor temperature. He formed the Butz Thermo-Electric Regulator Co. In 1906, Mark Honeywell founded Honeywell Heating Specialty Co, competing directly with Butz Thermo-Electric Regulator. The two companies merged in 1927, and following WW2, the company shifted from home heating into defense, formally establishing Honeywell’s aerospace business. In 1963, the company officially shortened its name to Honeywell Inc, as the company expanded beyond thermo regulators. In 1999, AlliedSignal acquired Honeywell to consolidate their aerospace dominance, but chose to keep the name “Honeywell” because of its immense global brand recognition. Fast forward to 2025, the company announced its intention to split up the company into two key segments, Honeywell Automation and Honeywell Aerospace.

This writeup will be specific to the Honeywell Aerospace (HONA) division of the business.

Explaining the Aerospace industry and how it works

On the demand side, there are two groups of customers seeking commercial airplanes - Air leasing companies (SMBC, Aercap, Air Lease Corp) that make up half of the global fleet, and larger direct airline companies (Direct carriers) making up the other half. Aircraft lessors purchase aircraft from OEMs and lease them to airlines under long-term operating leases. Demand is driven by the strong air travel demand. In Defense, the sector operates as a “monopsony”, with governments serving as the sole customers.

On the supply side, in commercial air travel, the market is an Airbus-Boeing duopoly. These two Original Equipment Manufacturers (OEM) sit at the top of the value chain, and they source for parts from Original Equipment (OE) providers. OEs often provide steep discounts to the OEMs, to incentivise the OEMs to incorporate their parts on the aircraft. Adding on, OEs and OEMs often collaborate as Risk-Sharing Partners (RSPs), with OE providers spending on Research and Development (R&D) with the OEMs to test that specific equipment on the plane. The OEs are, in a way, bankers or insurance providers for the OEMs. The decision to choose an OEs often lies in performance, reliability, commonality as well as incumbency. OEs are often grouped into tier 1 and tier 2 suppliers. Tier 1 suppliers are the OEs that directly supply their parts to OEMs, whereas tier 2 suppliers provide parts for the Tier 1 OEs.

Tier 1 OEs operate on a ‘Razor and Blade’ economic model, where OEs capture long-term value through the aftermarket. The “Aftermarket” (AM) refers to everything that happens after the plane is delivered. Aftermarket fees are often paid by the airline companies, given their weaker position, and with the cost of maintenance only making up a small portion of all their cost. Airline companies often absorb the majority of price hikes from OEs for maintenance, due to regulation and their generally weaker bargaining power. In AM, there are 4 sub-categories:

MRO (Maintenance, repair, overhaul): These are direct maintenance provided by the OE for their specific part on the aircraft, including spare parts and service maintenance. Service maintenance contracts are regular contracts that include fixed intervals for maintenance, which is regulated by the FAA. Spare parts and replacements are tied to flight hours and air travels, with replacements required after a certain number of flights.

In terms of MRO revenue, they can be split into Airframe MRO and component MRO. Airframe MRO is tied to the inspection of the fuselage, wings and landing gear of a plane. This is often labour-intensive, competitive, and lacks barriers to entry, resulting in lower margins. In component MRO, repairing an APU or Avionics equipment requires proprietary data and tests, allowing for higher margins compared to Airframe MRO. With significant component MRO exposure as compared to Airframe MRO, Honeywell’s Aftermarket revenue boasts high margins.

RMU (Retrofits, Modifications, and Upgrades): These are upgrades that tier 1 OEs provide for the existing installed base that they have.

PMA (Parts Manufacturer Approval): These are providers for non-OEM parts, otherwise known as generics or 3rd party parts. These companies typically provide equipment such as brakes, wheels, safety belts, etc.

USM (Used Serviceable Material): These parts are typically harvested from retired aircraft to save on cost, following testing and regulatory approvals. USM is often practised for Engines to save on high costs, allowing an airline to buy a used engine with only 2-3 years of life left to avoid paying $10M+ for a full overhaul on an old plane.

Why barriers to entry are so high in the aerospace industry

In aerospace, regulation serves as a significant barrier to entry for new entrants and competitors. The reasons for this stringent regulation include (1) passenger safety (2) certification process (3) system reliability (4) environmental sustainability.

In order for a plane to be licensed for commercial use, the OEs and OEMs have to be certified by the FAA based on safety regulations. This involves a multi-year process of stress testing and Flight Test campaigns, with the entire process that can span 8 to 10 years for complex systems like engines. with the FAA issuing a “Type Certification” for a new plane as a whole. Because the aircraft is certified as a whole, any change to a component requires an “Amended Type Certificate”. The delay in deliveries result in OEMs incurring high production costs while being unable to recognize revenue. This means that OEMs often lock in incumbent OEs, with minimal switching away from existing OEs in an aircraft. This implies that the incumbent OE already on the aircraft will be able to continually generate AM revenue for decades (~30 years) after the initial production of an aircraft. Products have decade-long cycles, resulting in minimal switching away from incumbents. To further expand on this, when Boeing or Airbus creates a derivative aircraft (like the 737Max vs 737NG), they try to certify it by similarity, allowing incumbents to keep their legacy parts on new planes extending aftermarket revenue.

Industry growth projections

In commercial air travel, the key driver for growth is tied to air travel demand. Revenue Passenger Kilometres (RPK) can be broken up into two key components - volume of passengers and distance flown, where RPK = Number of Passengers x Distance Flown. Based on Airbus GMF forecast, RPK is projected to be at 3.6% CAGR over the next 20 years, while Boeing CMO forecasts net fleet growth of around 3.1% over the next 20 years, suggesting stronger growth for OEs and OEMs with increased pricing and recurring revenue over time. Net fleet growth is lower than RPK due to increasing larger seat planes. With retirement rates at ~1%, total annual fleets built will be closer to 4.1%, providing a baseline for growth in the aerospace industry.

In Defence, the company has strong tailwinds behind. Defence as a percentage of GDP has grown to over 2.5%, following a period of about 2% before the Russia-Ukraine War. With heightened uncertainty and increased wars and conflicts, defence spending will likely remain elevated.

Honeywell’s Business Segments

Honeywell Aerospace has been sorted into 3 key business segments:

Electronic Solutions (ES): ES is $6.8 billion of 2025 net sales. To summarise ES: Avionics, navigation and sensors are the brain of the aircraft, often in the cockpit of the plane, controlling all components of the plane tied to the software, with software like recurring streams of revenue. Honeywell is currently #2 in this segment, behind Collins (RTX). In Avionics, the three major suppliers - Thales, Honeywell, and Collins - are each specialized in distinct avionics areas, controlling about 60-70% of the commercial segment. Even though Honeywell competes across many avionics segments, its barriers in this segment are relatively high with only a few players like Collins and Thales that can even compete at that level. In ES, earnings are driven by proprietary software-like revenue, with revenue growing with pricing and installed base, driving long term recurring revenue.

Engines & Power Systems (E&PS): E&PS is $5.4 billion of 2025 net sales. This segment supplies propulsion systems, auxiliary power units and electric power solutions. In E&PS, Honeywell is dominant in two main segments, APUs and propulsion. In APUs, Honeywell holds a dominant ~70% market share, in a duopoly with Pratt & Whitney at 23% (RTX), with the company as a sole-provider for many of its existing fleets, serving as a significant barrier against competitors. Currently, the company has about 36,000 units in service globally. In propulsion, while GE, Pratt & Whitney and Rolls-Royce dominate the engines for commercial planes, Honeywell has a dominant position in business aviation (business jets and small-mid sized jets). They are a key provider for aircraft from Bombardier and Gulfstream. E&PS operates on a ‘razor and blade’ model, where installed base growth and pricing power drive earnings, especially because the proprietary parts have highly constrained sourcing due to certification.

Control Systems (CS): CS is $5.2 billion of 2025 net sales. CS is broken up into two key segments - Air & Thermal Management (Environmental Control Systems) as well as Motion Control Systems (brakes, landing gear). In CS, Honeywell’s standing is weaker than the other two segments, due to this segment’s less differentiated and more competitive nature. This segment is largely driven by volume and frequency, due to higher usage and wear and tear.

Value is primarily driven by ES and E&PS, which combine high switching costs with long-duration aftermarket revenue, while CS is lower-margin and less differentiated. In terms of its dominance in next-gen airplanes, such as (A320neo, A350, B737 Max except B787), as well as regional jets, a meaningful share of them contain Honeywell’s APU. While avionics systems are typically multi-sourced, Honeywell maintains strong positions within the segment. With dominance in next-gen aircrafts, revenue visibility is strong for the next 10-15 years, given the long duration product cycles.

For Honeywell Aerospace, ~57% of Aerospace revenue is driven by AM revenue, with the other 43% made up by OE revenue. Honeywell Aerospace revenue is currently split into 60% Commercial Business and 40% Defense. In the commercial business, Honeywell Aerospace’s current OE revenue is split into 50% Business Aviation and 50% Commercial Air Travel, due to higher content cost in Business Aviation aircraft, despite lower fleet count. In AM, Business Aviation makes up 33% of the total AM revenue, with Commercial Air Travel still driving significant portions of revenue.

Aftermarket revenue is primarily a function of flight hours, total installed base, pricing and maintenance intensity. Aftermarket service contracts typically contain inflation-linked or labour-index linked escalation clauses to keep up with inflation. Aftermarket gross margins are much higher than OE margins, driving 65-70% segment EBIT for Honeywell. Besides, Aftermarket Revenue is typically 2-3x the original OE revenue. This is primarily due to downtime cost for the airlines, the need for identical parts for critical equipment of the plane due to Type certifications, and intellectual property. Because planes cannot afford any downtime, they prioritise reliability and speed, allowing Honeywell to leverage on this by offering Maintenance Service Agreements (MSA) that are typically Power-by-the-Hour contracts. Honeywell’s primary goal in the aftermarket is to move customers away from “transactional” purchasing towards MSAs. Transactional purchases are when airlines buy a part only when it breaks, which is normally unpredictable for the airline. On the other hand, MSAs are charged on a flat fee per flight hour basis. This creates recurring, high margin revenue and locks in the customer for years.

For MSAs, once signed, it is nearly impossible for a third-party shop or a PMA competitor to come into competition, as the airlines are already “pre-paying” Honeywell for the service. Also, if Honeywell can make a part more durable, MSAs allows Honeywell to convert engineering improvements into direct profits. This is a win-win scenario for all.

Within Aftermarket revenue, another key driver for Honeywell is their RMU business, representing ~10% of total revenues today, growing at low double digit. RMUs are independent of production rates and flight hours. They are upgrades that they perform on existing products on those aircraft platforms that address obsolescence, with introduction of new features that align with safety mandates. When aircraft production delays occur, Honeywell will be able to push for more upgrades as older fleets become obsolete, providing another meaningful revenue stream for the company.

In the case of Honeywell, with RPK projected to grow at 3.6%, which typically translates to similar growth in aircraft utilization and flight hours over time, and with pricing at 1-2 points above inflation (~4% pricing growth in FY25), base line revenue growth for aftermarket revenue is at ~7%. Maintenance frequency will likely continue to increase given production delay and ageing fleets, further pushing aftermarket growth numbers.

Another key arm in Honeywell’s entire business is their Defence business, currently about 40% of total HONA revenue. Similar to its Commercial Business, Honeywell aims for platform wins by selling OEs at low margins, deriving long term aftermarket revenue thereafter. Honeywell’s Defence Business has remained resilient throughout the years, as shown during the Covid-19 pandemic, where Honeywell’s Defence Business grew 9%. In Defence Aftermarket, most of Honeywell’s profits come from Performance-Based Logistics (PBL) contracts. In these deals, the military pays Honeywell a flat fee annually to ensure a fleet is ready to fly. Similarly, if Honeywell is able to make these parts more durable, their margin increases because they platform fewer repairs for the same fixed cost. Even when new aircraft deliveries slow down, the military must maintain existing fleets, ensuring that Honeywell’s high-margin aftermarket remains stable even if the OE side fluctuates.

Honeywell’s Moat

The moat of Honeywell Aerospace is driven by certification lock-in, incumbency, and high switching costs, which create decades of annuity-like revenue from the aftermarket. Once embedded into an aircraft platform, Honeywell benefits from structurally higher margin AM revenue over ~30 years, turning initial OE contract wins into long term profit for the firm.

However, there are risks to this model - OEMs such as Boeing and Airbus may pursue dual-sourcing to reduce supplier dependency, which could weaken switching costs over time. However, despite this, there are still sufficiently high barriers against new entrants, signally a resilient oligopolistic industry.

Honeywell mitigates platform concentration risk through diversification, with top 20 platforms representing only 30% of total revenue, and with no single platform contributing more than 5% of total revenue.

Honeywell’s global logistics network also serves as a moat for the company, particularly in time-critical Aircraft on Ground (AOG) situations. The company supports aircraft owners through a network of over 600 authorized service and sales centers worldwide, on a much larger scale than competitors in Collins Aerospace or Dassault Aviation, which operate about ~60 core service locations. Greater network density allows for shorter time required to deliver components, allowing the company to also capture a greater share in AOG situations.

Honeywell’s ability to leverage on pricing power should not be understated. While Honeywell often sells OE at low margins (sometimes at a loss) due to selection credits given to aircraft manufacturers, the company has been able to enact strong pricing power in the aftermarket due to the mission critical systems that they provide. Because Honeywell provides proprietary parts where no PMA alternative exists, Honeywell has been able to implement annual price increases that far exceed inflation (often within 5-10% ranges). Currently, with industry wide OEM share at around 77%, we expect Honeywell to be able to exercise strong pricing power for the majority of its products.

Competition & Risks

Honeywell has demonstrated superior margins and aftermarket revenue as a percentage of total revenue relative to competitors, with steady and high ROIC. In the past 2 years, due to litigation, one-time settlement and spin off expenses, Honeywell’s segment margin has been pushed lower. Normalised segment margins should be closer to 27.5% in 2023. However, while switching risk on current certified platforms is low, competitive risk exists in new platform bids and in the aftermarket. There are also risks associated with PMAs, STC certifications that allow for upgrades from other companies as well as Airlines MROs.

PMAs (Parts Manufacturer Approval) refer to parts supplied that are non-OEM. PMAs provide a viable alternative to airlines, due to lower costs as compared to the original OE. Transdigm and Heico are the dominant companies in this segment, with strong focus on the parts that are not safety critical, allowing airlines to switch away from OEMs easily. However, for PMAs, there are key stumbling blocks along the way. Using a non-OE part on the plane may significantly lower the resale value of the aircraft. Leasing companies (like AerCap) may also include “OEM-only” clauses in their contracts, preventing usage of PMAs. OEs may also void the warranty if PMAs are used as an alternative. Therefore, PMA risk is structurally limited to non-critical components, meaning Honeywell’s high value avionics and APU segments remain largely protected. Honeywell stays ahead of PMAs by making parts so complex that they are difficult to reverse engineer. Furthermore, with Honeywell’s expansion of PMA offerings today, this segment could turn into a tangible revenue stream for the company.

While certification lock-in secures long-term aftermarket revenue for incumbents like Honeywell, STCs weaken this advantage by allowing airlines to replace components. STCs (Supplemental Type Certificates) are FAA-approved modifications to an existing aircraft. STCs may curtail an existing OEs access to recurring revenue if the airline/fleet owner chooses to upgrade the fleet’s equipment to a competitors’ equipment. However, the economic incentive to switch is normally driven by cost savings or regulatory requirements, and adoption is slowed by installation costs, operational disruption, and airline preference for certain systems. In the case of Honeywell, STCs are a double-edged sword that Honeywell can also leverage to expand beyond its installed base.

Airline in-house MROs represent a potential risk to OE service revenue, as airlines attempt to reduce costs by performing repair work internally. However, the threat to OEM aftermarket is likely overstated. Even when maintenance is done internally, airlines will still have to source proprietary spare parts from the OEs. As a result, while airline MRO may capture service labour, OEMs such as Honeywell retain significant revenue through parts provision and proprietary systems. However, the risk could still increase PMAs gain broader acceptance, eroding OEs pricing power over time. But nonetheless, estimates above suggesting that airlines could displace a large portion of OEM aftermarket revenue may overstate the impact.

Aircraft production delays may represent a potential headwind to Honeywell’s OE revenue, as aircraft delivery rates remain below historical peaks for both narrowbody and widebody platforms. Airbus and Boeing have failed to catch up with delivery projections. However, delays in aircraft deliveries from Boeing and Airbus extend the lifespan of existing fleets, increasing maintenance intensity and spare parts requirements, driving incremental aftermarket revenue for Honeywell.

Another risk is that if Honeywell loses share on next-gen platforms, installed base growth slows, breaking long-term aftermarket compounding. However, strong incumbency on current next-gen platforms provides 10-15 years of insulation. Their RMU strategy also means they actively upgrade existing platforms, deepening lock-in even on fleets they already serve.

RTX owns both Collins Aerospace (Avionics) and Pratt & Whitney (Engines, APUs), allowing it to bundle their products in new platform bids. RTX could bundle engines + avionics + APUs, structurally reducing Honeywell’s win rates on new platforms. For instance, RTX recently signed three agreements with JetZero, with Pratt & Whitney providing the engine and APU while Collins supplies the nacelle and propulsion structure. If Honeywell loses share, this could break aftermarket compounding for Honeywell. However, our view is that the risk is real but slow, and is isolated only to new platform bids. On existing platforms, TCs make RTX displacement effectively impossible without a long re-certification process that OEMs avoid pursuing. That protects a decade of Honeywell’s aftermarket revenue regardless of what RTX does. OEMs also have an incentive to avoid single-supplier concentration, which weakens RTX ability to bundle their products.

Investment Theses

(1) Honeywell’s has structurally high quality earnings power that will likely expand over the years, with this catalyst serving as a way to correct it.

The markets currently underestimate Honeywell Aerospace’s long term earnings power, and are still pricing its Aerospace segment as a cyclical OEM-driven business, rather than as a high margin, recurring aftermarket business. The planned spin-off will create $30-40B of unrealised equity value by forcing the market to price Honeywell Aerospace as a standalone aftermarket compounder rather than a diversified conglomerate.

The market has failed to account for Honeywell’s structurally (1) pricing power in select proprietary components, particularly in Avionics and APU, (2) increasing maintenance intensity from aging fleets and a (3) large pool of deferred aftermarket revenue. This has resulted in a conglomerate multiple that fails to reflect the segment’s high quality earnings. The planned separation acts as a catalyst to connect the two views, allowing Aerospace to be valued as a standalone business more comparable to premium aftermarket peers such as TransDigm, RTX and HEICO.

While Honeywell’s management is guiding for mid single digit revenue growth in the Commercial Aftermarket, we believe that this outlook remains intentionally conservative when compared against the maintenance multi-year cycle currently impacting the industry. Compared to pre-2019, average fleet age is significantly higher today, up from 11.1 years to 15.1 years old - a 4-year increase that, at approximately 3.5% incremental maintenance cost per flight hour per year of age, implies a structural ~14% increase in maintenance intensity above pre-2019 levels. This drives two distinct revenue tailwinds for Honeywell, higher parts consumption per flight hour on the current installed base, and a wave of shop visits as aircraft cross the 12-15 year mark. Management’s guide does not appear to reflect this fully, creating meaningful upside to consensus aftermarket estimates.

Furthermore, global flight hours have grown past 2019 levels to reach a projected 5.2 billion passengers in 2026. However, while MRO capacity still remains bottlenecked, turnaround times (TAT) are ~60 days higher than pre-covid years. This means that maintenance durations are longer, and wait times for maintenance are also longer. This divergence between high aircraft utilization and slow servicing creates a large pool of deferred demand.

To back this up, assume a global commercial fleet size of 25,000 aircraft, and assuming the average shop visit frequency for an APU is about 8,000 to 10,000 flight hours. If TAT is extended by 2 months industry wide, this means that 1,400 shop visits are currently being deferred. This represents $440 million of deferred revenue for APU shop visits alone, about 5% of total commercial aftermarket revenue. This represents latent guaranteed aftermarket revenue that will eventually flow to the company.

An ageing fleet will require more spare parts, forcing higher parts consumption per flight hour, while maintenance delays result in a strong deferred pool of shop visits. To further develop this idea, maintenance delays also increase the demand for spare parts due to longer usage of aircraft. Based on a simple model, installed base growth (~3%), pricing (~4-5%), and incremental maintenance intensity, Honeywell’s long term aftermarket uplift will likely exceed management’s forecast. And so, with a significant portion of revenue derived from the Aftermarket, Honeywell stands to benefit from this supercycle, with our projections at high single digit Aftermarket revenue growth.

We believe the appropriate comparables are RTX, HEICO, and TDG, due to strong aftermarket revenue and OE mix. The planned spin-off serves as a catalyst by separating Aerospace from lower quality automation businesses, allowing the market to more clearly value its recurring, high-margin earnings profile. This, along with a structural aftermarket upcycle, will support both earnings growth and multiple expansion. While high-quality aftermarket peers trade at a premium, we conservatively assigned our base case at 18.2x, in line with broader industry median. At our base case multiple, the investment is driven largely by earnings quality and earnings growth rather than just a re-rating, with multiple expansion providing additional upside.

Consensus aftermarket growth of mid single digit reflects management’s own guide. However, it does not include the structural maintenance intensity increase, the deferred revenue backlog, or Honeywell’s strong pricing power in proprietary components. These three factors will push aftermarket growth into high single digits.

Assuming a 2027 EBITDA of USD 5.8B based on our model estimates, and 15B net debt based on Form 10 filings

Honeywell Automation is the combination of Industrial Automation (IA) and Building Automation (BA). BA includes CCTV systems, lighting controls, air systems, access controls for commercial buildings, airports, hospitals, facilities and data centers. IA provides sensor solutions, process solutions, etc for oil and gas, logistics and utilities companies.

We apply a discounted multiple of ~15x EV/EBITDA to the Automation business (~10% discount to peer group median), due to its structurally lower growth profile and greater exposure to short cycle industrial demand. Unlike Aerospace, which benefits from long-duration revenue visibility and embedded aftermarket economics, Automation revenues are more sensitive to capex cycles and lack recurring, high-margin service streams. As such, a valuation discount to pure-play automation leaders is fair.

However, we have to acknowledge that we may be underestimating their remainco’s true value. The Remainco has divested its Personal Protective Equipment (PPE) business, which was creating a margin lag due to unprofitability. They are also planning to spin off Warehouse and Workflow Solutions (WWS). Removing those two segments will push for margin expansion of 3-4pts, putting its margin profile at the high end of industry peers.

Assuming 2027 EBITDA of USD 5.8B, the implied enterprise value is:

Adding on, Honeywell’s Quantinuum stake is set to IPO in 2027, which could add a further $5–$10B in EV, as of its latest private Series E valuation at about $10 billion, with IPO targets at $15 billion. Combining the estimated values of the all three segments, the base case combined EV is at 203B. With current EV of 158B, this implies a 28% discount to its SOTP fair value. A SOTP also neglects the quality of a standalone business, and we believe that HONA’s value is probably underappreciated today.

(2) Near-term margin recovery from escalator resets, combined with a business model that benefits in both high and low OEM production environments, creates a scenario where earnings growth is supported regardless of OEM production schedules.

Honeywell Aerospace’s current reported margins of 24.5% are artificially suppressed. Normalised margins in 2024 were at 27.1%. The compression was driven by one-time Flexjet litigation charges and impairment on held-for-sale businesses. A recovery to 2024 levels of 27.1% represents 260bps margin improvement, implying ~$450M of EBIT not in current numbers and about $8B of equity value at 18x EV/EBITDA, adding an additional 5% upside. The short term catalyst for this recovery is the reset of long-term service agreement price escalators in 2H 2026, as price increases will now reflect the increase in costs over the past years.

Adding on, with regards to production targets, there are generally two stances: (1) Airbus and Boeing will increase production over time back to pre-covid levels and (2) the OEMs will continue to disappoint production targets. Honeywell’s aftermarket focused business reduces sensitivity to OEM production delays. When productions are high, increased aircraft deliveries expand the installed base, ensuring long term aftermarket growth. When production levels fall, aging fleets drive longer utilisation and maintenance intensity, benefiting short term aftermarket demand. Honeywell’s revenues are structurally more balanced than traditional aerospace suppliers. This creates a rare business where near term earnings are supported by installed base resilience, while long-term growth is driven by structural expansion of the global fleet.

Total deliveries for commercial aircraft from Boeing and Airbus stand at 1,393 today, which is still below pre-COVID peak of 1,658 aircraft, despite strong recovery in global air travel demand. Persistent supply-chain constraints have limited manufacturers’ ability to ramp production. Backlog for aircraft production is at a record of ~15,000, providing more than 10 years of production visibility. This suggests that OEM production ramp is structurally supported, implying that installed base will continue to grow at 3%+ for a decade.

Bear case for the thesis: The bear case is a demand shock that causes discretionary maintenance deferral, delaying the aftermarket recovery. If escalators resets are negotiated lower than expected, margin expansion could be less than expected. However, FAA-mandated maintenance is non-discretionary, limiting the downside to pricing.

Valuation

Our base case assumption is based on projections by Boeing and Airbus, with sensitivity for each figure reflecting our bull case and bear case. Flight growth along with ARR growth were used to model our aftermarket revenue growth, while net additions in fleets each year were used to model for yearly OE revenue.

Assuming an Equity value for Honeywell Aerospace at about USD 91B today, the base case equity value for Honeywell Aerospace stands at USD 128B, signalling a 40% current discount to fair value, implying a 14.8% CAGR.

Bear Case:

Bull case:

Appendix

The 3.5% terminal growth rate is supported by long term GDP growth of ~2% and Honeywell’s ability to price 1-2pts above inflation

Plane parts and descriptions

The Engine (Powerplant): GE, Safran (CFM Joint Venture), Rolls Royce, Pratt & Whitney are the major OEs, with airlines charged based on Power-by-the-Hour (PBH) contracts and miles driven. The Body (Airframe): This includes the tail, wings and fuselage of the aircraft. Structural checks are required every few years. The brain (Avionics): This includes the flight computers, navigation, and cockpit displays. Companies like Honeywell obtain recurring revenue primarily through spare parts, service contracts, software and updates. Extra Auxiliary power unit (APU): It is a small engine in the tail that provides power and AC while the plane is parked at the gate. APUs are charged based on PBH contracts, which is a flat fee per hour.

On 28 February 2026, the United States and Israel launched coordinated military strikes against Iran, killing Supreme Leader Ali Khamenei and triggering a series of retaliatory missile attacks across American bases in Jordan, the UAE, and Qatar. Oil prices soared past $100 per barrel for the first time since 2022, and stocks retreated sharply as markets began processing what J.P. Morgan’s co-head of economic research described it as generating “greater macroeconomic risk than recent military conflicts.” Three weeks later, the conflict shows no sign of resolution. The Strait of Hormuz, through which approximately 20% of the world’s seaborne oil and a significant share of global LNG transits daily,remains effectively shut. The International Energy Agency has called this the largest supply disruption in the history of the global oil market.

The question every investor must now answer is not whether this matters, but precisely how it matters to their specific portfolio and what, if anything, they should do about it.

The most direct route through which this conflict reaches portfolios is energy pricing. Brent crude rose from around $70 to over $110 per barrel within days of the conflict beginning, disrupting approximately 20% of global oil supplies that transit the Strait of Hormuz. For investors, this is not simply a commodity price story. It is a broad cost-push shock that runs through the income statements of almost every industrial, consumer, and logistics business on Earth.

Airlines face immediate margin compression, jet fuel is their single largest variable cost, and the forward hedging positions of most major carriers extend only six to twelve months. Logistics and shipping companies, road hauliers, and any business running a large vehicle fleet see operating costs rise sharply. Consumer staples companies face pressure on packaging, transportation, and raw material costs simultaneously. Emerging market economies particularly those in Asia with thin foreign exchange reserves and heavy dependence on Middle Eastern crude face currency weakness and inflation simultaneously. The ADB has flagged that smaller energy-importing economies including the Philippines, Pakistan, and Sri Lanka are likely to experience the strongest macroeconomic effects, as higher oil prices transmit rapidly into inflation and exchange rate pressures through widening current account deficits. The key distinction investors must make is between businesses that can pass through higher energy costs and those that cannot. Regulated utilities with cost pass-through mechanisms, businesses with pricing power, and natural resource companies are insulated or benefit. Fixed-price contract manufacturers, asset-light services businesses with no pricing power, and highly indebted consumer-facing companies face genuine margin risk if elevated energy prices persist for two or more quarters.

The second and more structurally dangerous transmission mechanism is the monetary policy response. Central banks in both the US and Europe entered this conflict already navigating a delicate pivot. They had paused easing cycles but had not yet declared inflation conclusively defeated. The Federal Reserve now faces a situation in which any resumption of rate cuts has been significantly complicated by an oil-driven inflationary impulse, while labour market softness raises the spectre of stagflation, a situation with a particularly difficult combination of high inflation and weak growth.

A more severe scenario in which the conflict persists for several months could see oil prices rise to around $130 per barrel before declining. Under this scenario, the eurozone economy would probably contract in Q2 and then flatline over the second half of the year, while the US would experience a slowdown but avoid contraction outright. The ECB, which held rates at 2.15% at its meeting on 19 March, is already navigating revised inflation projections of 2.6% for 2026 with GDP growth cut to 0.9%. A sustained oil shock complicates that picture materially. The worst-case scenario, as articulated by economists at the Middle East Council on Global Affairs, is an economic slump combined with an interest rate hike to curb inflation which is a combination that could trigger the bursting of asset bubbles and potentially lead to another debt crisis similar to 2008.

Long-duration assets such as growth equities, infrastructure, real estate, and long-dated bonds are mechanically sensitive to higher discount rates. The equity risk premium, already compressed in many markets, expands in a stagflationary environment. An investor who had been extending duration to capture the tailwind of falling rates must now consider whether that trade has been reversed.

Not every portfolio position suffers in a prolonged conflict. History is clear that geopolitical shocks of this magnitude create durable sectoral winners alongside the losers.

Energy producers and commodity companies such as warrior met coal which I possess in my portfolio are the most obvious beneficiary. LNG supply from the United States now has added long-term appeal as global buyers reassess concentration risk from the Gulf, and the structural case for US LNG export infrastructure has strengthened considerably as buyers seek to diversify away from Qatar’s dominant market position. With LNG setting new record highs, it is likely that we will see a similar situation in the met coal markets as we saw back in 2022 where energy producers were burning met coal and thermal coal instead of LNG due to the LNG shortage and energy crisis.

Defence and aerospace represent a second clear winner category. Morgan Stanley explicitly recommends increasing exposure to defence, security, aerospace, and industrial resilience themes, where government spending can drive multiyear demand. European governments already accelerating defence spending following the Ukraine war have further political justification to increase procurement budgets. Missile systems, radar infrastructure, intelligence technology, and satellite communications all fall within the category of government expenditure that is directly counter-cyclical to the broader economic slowdown. A recent addition into my portfolio was Airbus which has a defense segment as well as supplies helicopters to the military in Europe. I am seeking to increase my exposure to other companies such as Safran which produces turbines for airbus helicopters and having a clean business model with long term revenue visibility. I will be doing a separate writeup for both companies.

A third and less obvious winner is domestic demand infrastructure in non-exposed economies. Tower companies, utilities, toll roads, and regulated infrastructure assets in politically stable Western economies exhibit genuine recession and conflict resilience as their revenues are contractual, domestically denominated, and largely indifferent to oil prices. A towerco’s ground lease costs do not rise because Brent is at $110. A utility’s contracted revenues do not compress because Hormuz is closed. For investors seeking shelter from energy and macro volatility without exiting equities, infrastructure assets offer a genuine alternative return stream.

Consumer discretionary is perhaps the most vulnerable broad sector. Rising oil and gasoline prices reduce the purchasing power of consumers directly, with US consumers spending more on fuel and less on discretionary goods as the average petrol price reached $3.41 per gallon in early March, up $0.43 in a single week. Luxury goods, leisure travel, hospitality, and any business dependent on consumer confidence face a particularly difficult combination: higher input costs and weaker demand simultaneously.

Aviation faces a structural challenge. The closure of Middle Eastern airspace has already disrupted routing and added materially to journey times and fuel burn on long-haul routes. The closure of the airspace above the Middle East has caused widespread disruptions in aviation and tourism globally, with carriers forced onto longer transoceanic routes that increase costs while consumer appetite for leisure travel softens. Though this may have serious repercussions on air travel in the short term, it hardly impacts Airbus considering that Airbus has 10 years of backlog most of which is for narrowbody, single aisle planes that do not pass by Dubai (considering that Dubai is the transport hub for long haul flights from europe to asia pacific)

Investors focused solely on energy pricing risk missing a secondary set of supply chain dislocations that could prove equally persistent. Qatar accounts for roughly a third of global helium supply, a by-product of natural gas extraction that is essential for semiconductor manufacturing and medical imaging equipment. Disruptions to production or shipping could affect technology and healthcare industries far beyond the Middle East. ASML is potentially affected by this and I will be keeping a close eye on any further developments. GIven that the valuation of the company is stretched, it is highly likely that any bad news will lead to a severe decline in the short term. Fertiliser supply chains are equally vulnerable as the Persian Gulf is a critical hub for global fertiliser production, with natural gas a key input for urea, the world’s most widely used nitrogen fertiliser. Agricultural commodity prices are already responding, with rice and wheat moving higher as Asian export routes through the Gulf are disrupted.

Preparing a portfolio for a prolonged Iran war does not require dramatic repositioning it requires an honest stress test of existing exposures against three distinct scenarios: a short, contained conflict (six to eight weeks); a prolonged conflict of three to six months; and an escalation scenario involving broader regional destabilisation. Each scenario has materially different implications for energy prices, interest rates, credit spreads, and corporate earnings.

In the short-conflict scenario, the portfolio changes required are modest. Oil producers benefit temporarily, duration assets recover quickly, and consumer stocks rebound. The market already prices something close to this. The risk premium that justifies defensive repositioning is precisely the asymmetry between the short scenario which is already partially priced into the markets and the prolonged or escalating scenarios, which are not.

In the prolonged conflict scenario, a balanced portfolio would hold meaningful weights in energy production, defence, domestic infrastructure, and commodity companies. It would reduce exposure to consumer discretionary in oil-importing economies, long-duration growth equities trading on stretched multiples as well as speculative small cap positions that are highly interest rate sensitive.

Introduction

The story of FICO started in 1956, when two businessmen - Earl Isaac and Bill Fair formed a joint venture together, Fair Isaac. Back in the 1950s, lending decisions were mostly subjective, based on judgement and knowledge of the environment. Earl and Bill saw a need for standardisation, and saw an opportunity where statistical models can be used to make credit decisions. Between 1956 and 1972, FICO focused on creating custom scores for retailers, shops, banks, without a standardisation. In 1972, they created an Automated Application Processing System specifically for Wells Fargo, one of the bigger banks in the US at that time. This helped with the scalability of their FICO scores.

In 1989, that was the year when the company saw a significant breakthrough. They created their first credit bureau risk score, the FICO classic, sold to Equifax. In 1995, the scale of the company was further strengthened with requirements in mortgages under Government Sponsored Enterprises (GSEs) Fannie Mae and Freddie Mac. Adoption accelerated after this, with agreements with Transunion in 2000, and expansion of usage across credit bureaus and consumers. In 1996, FICO entered the software platform space, with the acquisition of Blaze Advisor. This marked the start of their dominance in both the software and credit score space.

Business Summary

Fair Isaac has three key business segments, scores, software and professional services. Scores refer to the FICO scores that are issued to assess a consumer’s credit health. Software is broken up into two parts, platform ARR and non-platform ARR. Platform ARR refers to software sold on FICO’s cloud native platform, whereas non-platform ARR are legacy or on-premise software licenses and maintenance..Professional services refer to consulting based services, typically to help a customer start using FICO’s software platforms.

FICO’s scores business is the crown jewel of the company, with ~90% operating margins, accounting for about 55% of total revenues. FICO’s consolidated software business accounts for the remaining 45% of total revenues, with operating margins of about ~30%.

The Software Business

FICO’s software business is one that is often overlooked, due to weaker margins as compared to the scores business. FICO’s software business is split into two arms - Platform ARR and non-platform ARR. Non-platform ARR includes FICO Blaze Advisor (Rules Engine), FICO Falcon Fraud Manager, FICO TRIAD Customer Manager and FICO Xpress Optimization. Currently, Non-Platform ARR is declining, with FICO pushing for a switch towards platform ARR.

Under Platform ARR, FICO offers similar SaaS replacements for those on-premise, with additional SaaS focused on scoring models and originations. Currently, Platform ARR is split between Originations, Customer Management, and Fraud protection. Originations involve using APIs to train internal scoring models (Build-your-own-model), allowing for greater credit loan decisions.

Currently, the on-premise software business is growing strong, with increased demand for FICO’s APIs driving a significant portion of this. Net Retention Dollar has been strong at about 122% in FY25, along with strong Annual Contract Value Bookings (ACV), signalling strong growth for the years ahead.

Types of Scores and Revenue of Each Score

FICO 8: Industry standard, base scoring for most use cases.

FICO 9: Incorporate changes like ignoring paid accounts and giving less weight to medical conditions.

FICO 10: FICO 10 uses the same single-point-in-time credit report.
FICO 10T: FICO 10T is used with trended data from the past 24 months.

Within the FICO base scores lie the industry specific scores: FICO autoscore 8, 9,10 (used for auto loans), FICO Bankcard score 8, 9, 10 (used for credit cards). For mortgages, FICO Score 2 (Experian) FICO Score 5 (Equifax) FICO Score 4 (TransUnion) are typically used. FICO 1, 3, 6, 7 are earlier versions that were commercialized, but eventually replaced.

FICO mortgage scores are priced at $4.95 per score. FICO Bankcard and Auto score are priced closer to $1-2 per score, with only a single pull required, without any regulations for tri-merge scores. Autoscores are typically used by automakers when giving out credit, Bankcard scores are used by banks when assessing consumer credit worthiness. FICO’s consumer segment is subscription based, at a pricing of $19.95/month for one specific score, or $40/month for all your scores.

FICO’s B2B scores represented 82% of total scores revenue, with Consumer scores at 18% of total scores revenue, growing at mid-single digit. Within B2B Scores, Mortgage scores accounted for 54% of total B2B scores revenue and 44% of total scores revenue, with Auto at 24% of B2B revenue, and 19.7% of total scores revenue and consumer credit card scores at around 25% of B2B revenue, and 20.4% of total scores revenue.

Understanding the Lending Market

Generally, a lending sequence begins with origination, where a consumer goes to a lender in order to ask for a loan. In order to determine the appropriate interest rate to charge the consumer, and whether or not to approve the loan, the lender has to assess the credit worthiness of the consumer, through a score. Lenders typically obtain their credit reports and scores through Loan Origination Systems (LOS), such as Encompass, MeridianLink and LendingPad, who connect credit bureaus via integrations. Currently, for mortgage loans, the lenders have to pull a “tri-merge” report, which is a credit report from all three bureaus. For auto or credit loans, the lenders will only have to pull the data from 1 bureau. After successfully extending the loan to the consumer, there are three possible courses of action that a lender can take: (1) securitization (2) hold it on their balance sheet or (3) sell the loan. Securitization refers to pooling many loans together and then packaging it up as a bond through a special-purpose vehicle (SPV), with investors buying this bond. Investors include (1) major banks in the US, (2) international sovereign wealth funds and (3) the Federal Reserve, who want cash flow backed by loan payments.

For auto loans and credit card loans, lenders typically hold the loans on their balance sheet and collect interest payments, or securitize the loans by pooling them together and issuing asset-backed securities that are sold to investors in the capital markets.

For mortgage loans, lenders typically will choose to sell the loan to Government Sponsored Enterprises (GSEs) in the Agency market or the To-Be-Announced (TBA) Market (or the conforming market), provided they meet the GSEs’ conforming loan limits of about $835,000 as of 2026. GSEs then securitize conforming mortgages into Agency Mortgage Backed Securities (MBS). GSEs include Fannie Mae and Freddie Mac. For Jumbo loans (loans exceeding conforming limit), and Non-QM loans (loans that fail), lenders will be required (1) hold them or (2) sell the loans to the private market. Private financial institutions then securitize non-agency loans into private-label MBS. The Agency Market currently makes up the majority of the mortgage market, around 70, with the remaining made up by the private market (non-conforming market).

There is a key nuance here when lenders sell their loans to GSEs. Currently, GSEs allow for the traditional FICO Classic score to be used. The Federal Housing Finance Agency (FHFA) has approved a future transition to FICO Score 10T and VantageScore 4.0, though the mortgage system still primarily uses the FICO Classic scores today. When purchasing loans, the GSEs apply Loan-Level Price Adjustments (LLPAs), which adjusts the price of the loan based on the borrower’s credit score and the loan-to-value ratio. Because the mortgage market currently relies on FICO Classic scores, existing LLPA grids are calibrated to those score ranges, though future modifications for VantageScore and FICO 10T scores are pending.

To understand FICO’s moat, a further development of the GSEs earning is important. GSEs are responsible for guaranteeing the loans from lenders, and pooling the loans into Agency MBS. These MBS are typically sold to (1) major banks in the US, (2) international sovereign wealth funds and (3) the Fed. About 75% of MBS are sold to domestic buyers, with 25% sold to international investors.

Currently, Global Banking Mandates (specifically Basel III Endgame) have made it such that rated debts are higher quality in nature, even with government guarantees. For example, to be “Basel III compliant” for a Japanese or European bank, an official credit rating is preferred. And so, a Moody’s rating will need to be issued, and with Moody’s currently mostly using the FICO Classic score distributions as a key input, with FICO’s decades of default data needed to stress-test the score, this implicitly strengthens FICO’s positioning. Domestically, FICO scores are still preferred for standardization.

FICO’s Moat

FICO’s moat lies in its decades of historical data, predictive scoring, as well as regulatory barriers.

Historical Data Moat: FICO has been the language of consumer credit since 1989. Rating agencies and institutional investors rely on credit models that have been proven through multiple economic cycles, including recessions and credit downturns. No other score, apart from FICO scores, have gone through these cycles. FICO 10T score serves to leverage on this, upgrading predictability with algorithms built through economic cycles. FICO scores serve as trust engines, especially in securitization, allowing for lower yield demands.

Predictive scoring and standardization: When deciding which score to use for lending decisions, predictive scoring and standardization are key considerations that a lender has to make. Lenders want to avoid repurchases of mortgages sold to GSEs, and repurchases happen when the loans become delinquent. While competitors like VantageScore often focus on inclusivity and scoring more people, FICO has focused on precision. VantageScore’s strength is in providing scores for those with “Thin-File” (those with limited credit history). FICO 10T, however, is currently focused on lifting default detection. With both companies having similar Gini coefficients, the varying factor lies in the different focuses. VantageScore’s focus of inclusivity allows for more loans to be made, but on the balance, this increases the likelihood of poor lending decisions tied to individuals with lower credit quality. Investors care about risk predictability, not mortgage inclusion. In our view, FICO’s methodology prevails.

Regulatory barriers: Currently, GSEs only allow for VantageScores and FICO scores to be used. Because regulatory barriers have mandated the use of FICO scores over the past 40 years, FICO has become the language for consumer credit risk. If a lender switches to a different score, they risk breaking the liquidity of the market, especially for international investors. Lenders who have used FICO for industry standardization have infrastructure built around FICO. This creates inherent inertia in switching. Because most of the cost for credit scores are passed down to the consumer in mortgage loans, lenders generally have minimal incentives to switch.

How to think about growth drivers

For FICO’s Autoscores and Consumer Credit Scores, pricing power is typically more moderate, with periodic price increases. Product migration, usage growth and upgrading of scores are now the main drivers for growth. Because the Autoscores and Consumer Credit Loans typically rely on Private Label markets for securitization, FICO scores are typically preferred, with greater trust in the private market.

For FICO’s Mortgage scores, FICO has adopted an aggressive pricing model over the years, with pricing growth outweighing volume based growth. Volume growth is more cyclical, with volumes tied to interest rates and overall economic activity in the housing market.

Competition

Following the Consumer Credit Competition Act and the 2025 FHFA approvals, lenders will be able to choose between VantageScore and FICO scores when making lending decisions to consumers, with both scores now accepted at the GSEs. VantageScore has been aggressive with pricing, with the most recent USD $1.00 scores to disrupt the mortgage industry. This is likely a push towards gaining market share, after FICO’s Direct Licensing push in 2025. The key argument here for VantageScore as a competitor is the erosion of FICO’s share in price sensitive scores. VantageScore could see share gains especially in Auto Loans or Credit Card Loans due to significantly lower prices relative to FICO scores. In Mortgage loans, following FHFA approval, VantageScore 4.0 can now be used in mortgages alongside FICO, introducing competition and potentially expanding credit loans, especially for borrowers with thin files.

However, there are a few structural advantages that FICO has over VantageScore. FICO Classic score is currently the only score that has GSEs’ LLPA grids, with extra surcharges typically tied to VantageScores. While there is the possibility of the development of LLPA grids for VantageScore, the time line is still unclear. This risk will be discussed later in the piece. To add on, because of the lack of historical data for VantageScores and VantageScore’s shorter mortgage specific history, investors typically demand a premium yield for mortgage pools backed by VantageScores (~15bps higher, thereby reducing lender economics if investors require higher yields, based on Fannie Mae releases). Our view is that FICO is intentionally letting go of price sensitive scores, in favour of significant price increases and capturing economics.

Other forms of competition include generated scores by companies like Upstart. Our view is that scores from these firms lack the trust to be implemented into lending decisions. FICO’s moat has been built on trust, the company is a stamp of approval on loans.

Risks

There have been suggestions by the FHFA to reduce credit reports from a tri-merge to a bi-merge, which reduces the volumes of scores required by one-third. The FHFA has continually pushed this back despite indicative suggestions a few years ago, but investors argue that Bi-Merge reports are generally less reliable than Tri-Merge reports and would miss the worse scores, therefore requiring a higher yield on loans. On July 8, 2025, Director Pulte announced that Fannie and Freddie would maintain the tri-merge requirement, effectively ending the plan to move to a bi-merge system. This suggests that plans to change to bi-merge have currently been put on hold.

LLPA Grid for VantageScores could significantly reduce the barriers for the usage of VantageScores. However, the creation of an LLPA grid requires historical mortgage performance data, something that VantageScore currently does not have. The likely scenario would be to tie VantageScores to FICO Scoring bands, with continued surcharges to be paid for the usage of VantageScore. If the spend here is lower than the cost savings narrative currently pushed by VantageScore, lenders will naturally stick with FICO scores.

Mortgage market cyclicality is a key risk to FICO’s mortgage scoring business. The mortgage market is tied to the Federal Reserve’s interest rate policy, which is inherently unpredictable. This could temporarily drive down mortgage score volumes for FICO, which could hurt FICO’s earnings in the short term. Our view, however, is that the predictability of the recognition of revenue is not important, so long as the company has recurring revenue. Those who need the scores will eventually have to get it, and during that period of time, as long as FICO’s ability to increase prices remain, it’ll be better economics for the firm.

Credit Bureau competition is a key risk for the firm. The company has stirred up competition with the 3 credit bureaus who created VantageScore, putting FICO’s scores business at risk, especially since the credit bureaus hold an important piece of data - the credit reports. Our view, however, is that the credit bureaus have long provided VantageScores for free along with their credit reports. There is minimal risk of FICO being alienated from the system, because it is (1) regulatory backed and because (2) FICO has been increasingly connecting their scores onto pipeline platforms that link lenders with the credit bureaus, such as MeridianLink and ICE Encompass Loan Origination System. FICO is embedded within the lending infrastructure.

Investment theses

There are two main investment theses that we would like to highlight: (1) FICO’s direct licensing programme serves as a form of apparent price increase, which will dramatically benefit revenues in FY26. (2) FICO’s business is defensible, with the markets increasingly pricing in a company that is facing structural headwinds.

FICO’s direct licensing programme serves as FICO’s way to bypass the credit bureaus, with the company providing scores on the direct to lender basis on loan origination systems.

On a blended weighted price basis, FICO is effectively doubling the scores prices this year, capturing the upcharges that the CBs have been charging. Thereafter, our model assumes moderate price increases, more or less in line with inflation, at ~4-5% annual price hikes. FICO has spearheaded this transition away from the CBs, with partnerships with key loan origination systems. After this price reset, the model assumes a steady state business with stable cash flows annually, barring any economic shocks. Given the high operating margins of the scores segment (~90%), the majority of this increase in revenue flows directly to the bottom line, creating a meaningful earnings rise in FY26.

Besides pricing, there is optionality with the mortgage market. From a macro perspective, with interest rates moderated over time, along with the Trump Administration’s push for the housing market, there should be an increase in mortgage activity over time. Our model assumes a 3.6B in annual revenue by FY28, ahead of the street’s estimate of 3.3B.

FICO has a defensible business, with the markets increasingly pricing the company for more market share loss over time.

FICO’s business has been embedded in the mortgage industry, and serves as a means to ensure continued liquidity in the mortgage industry. With the recent fall in FICO’s stock price, our view is that the streets are overreacting to this. FICO’s FY26 PE has compressed to 26x, below its 10 year median of 34x, despite a more dominant business today. FICO historically trades at a premium multiple due to (1) asset light business model (2) strong pricing power (3) strong recurring revenue. These characteristics are still present in FICO’s business today. Even assuming slower pricing growth after FY26, the company still shows mid-teens EPS growth, suggesting the current 26x multiple is undemanding relative to historical levels.

VantageScore’s charge has been overestimated by market participants today. VantageScore lacks the key ingredients needed to be a strong competitor: (1) Trust (2) Data and history. Even if VantageScore saves lenders and borrowers $9 on a single score, assuming a $400,000 mortgage and a 15bps difference in interest rates, lenders will lose $600 on a $9 savings cost. This is significant and destroys the economics for VantageScore. VantageScore has long been provided to lenders for free, along with credit reports, but has yet to receive any form of meaningful adoption. Hence, our view is that FICO will remain a dominant powerhouse, despite weakened pricing growth over the coming years due to regulatory and industry pressures.

While the scores business remains the main earnings driver, FICO’s software platform also represents a longer-term growth opportunity. As lenders create more lending infrastructure, FICO’s cloud-based decision management platform could drive further recurring SaaS revenue and deepen integration within the lending ecosystem.

Valuation

Our base case model assumes moderate price increases after FY26, 1-2 pts above inflation. We assume a gradual recovery in mortgage originations as interest rates normalize over time. This implies volumes recovering toward levels observed when the interest rates were at 3%. Auto scores growth and bankcard scores are modelled after industry growth trends along with embedded annual price increases.

Base Case:

Bear case: VantageScore adoption increases, putting pressure on pricing power and reducing FICO’s share of new mortgage originations.

Bull case: Mortgage volumes recover more quickly, while FICO remains as a monopoly in mortgage scoring. Direct licensing allows FICO to sustain higher pricing growth than currently expected by the market.

Appendix: Revenue Build

Author’s Notes:

At today’s price of $1150, we view FICO as a hold today, with the markets already pricing in the base case scenario for the company. While FICO remains a dominant business, we await for further multiple compression before the company becomes even more interesting to us.

By Guan Rui

(1) Company Overview:

Netflix started out as a “Kibble”, a company that sold and rented DVDs, it then evolved to DVD subscriptions, till today’s streaming service. The company survived the dot com crisis as well as intense competition with the DVD leader back then - Blockbuster - with its current co-CEO Ted Sarandos being with Netflix since 2000.

The majority of Netflix’s business is tied to a subscription business model, where users sign up for their Netflix streaming platform, along with ad revenue coming from their ad tier. The company sets a content budget every year, and they use this budget to either buy the licenses to content or to create content of their own. Most companies in the SVOD industry operate under the same model.

(2) Competition in the SVOD Industry:

SVOD streaming and traditional Youtube/Twitch are overlapping, with a growing number of people that use Youtube subscribing to Netflix. SVOD has been increasing in terms of total time share, with expectation that SVOD streaming will continue taking time share from other activities in normal daily lives, with a shift towards “Homebody Economy” and with the growth of Smart TV allowing the use of these SVOD services.

Specifically in the SVOD industry, there are a few key players - Netflix, Disney, Amazon, Paramount, HBO Max.

Extract 1: By Nielsen

Market share dynamics have shifted over the years, with new entrants coming into the space. Netflix used to hold a ~50% market share of the SVOD industry, back in 2020 when streaming was booming during Covid. It currently holds a market share closer to around 24%. This can be attributed to a few reasons, the main one being the start of other streaming services. Companies like WBD, Disney, NBC (currently Peacock) used to license out their films and shows to Netflix, but they have changed their model and created their own streaming services. Sony, Lionsgate and A24 are the remaining 3 ‘timeless’ content companies that have not stepped into streaming. Sony and Lionsgate have expressed disapproval of streaming, due to high competition and lower margins than licensing. Any new drastic entry driven market share moves is unlikely to continue.

Streaming is a non-zero sum industry, where users can subscribe to many different platforms. SVOD subscribers per user on average has grown from 3 services in 2020 to about 6 currently. Growth in the number of subscriptions per user has led to a more fragmented market share for the companies, which means that falling market share does not necessarily link to poorer economics.

For streaming companies, the best measure of a quality streaming platform is in its churn rate (the number of users cancelling), and the watch time per user (a measure of level of engagement by users). Netflix’s average watch time per subscriber is 1,077 minutes, which is higher than its closest competitors.

Extract 2: Streaming Watch Time per Subscriber using Nielsen’s Numbers

Netflix’s President of Advertising, Reinhard, has also emphasised that the company’s DAU and WAU metrics are superior to Amazon.

Extract 3: President Reinhard on Netflix’s dominance in engagement metrics

In terms of churn rate, Netflix has structurally lower churns, at a monthly rate of 2.1%, below the premium streaming average of 4%, even during periods of price hikes. Disney’s churn rate is closer to 8%, while Amazon’s Prime Video churn rate is likely low, due to its bundling effect with the Amazon Prime subscription.

Annual churn rate for Netflix is closer to 22%, while Disney+ is closer to 40%. Netflix’s churn rate is significantly lower than the share of “serial churners”, at around 40%, suggesting that the company is probably less susceptible to this behaviour. Win-back of customers for Netflix is also substantially higher, with 60% of users rejoining within 12 months, as compared to Disney+ at 35%.

Netflix reported total watch time of 95 billion+ hours in H1 2025 with a 1% y/y increase from 2024’s 94 billion hours. With the increase in subscribers, this implies that watch time per subscriber is down. The reasoning behind this include (1) Password crackdown leading to a surge in subscribers (2) Ad tier subscriptions driving growth which tend to be lower engagement users (3) Netflix’s weakening content appeal. We suspect the primary reason to be password crackdown, leading to subscriber surge, and hence lower watch time per subscriber since users were previously sharing.

The key difference in Amazon’s model and Disney/Netflix’s model is best explained with two words, utility vs feature. Amazon’s prime video has always been an extension of their existing Amazon Prime subscription. Disney & Netflix, on the other hand, serves high intent customers looking for streaming services, leading to higher engagement and lower churn. Netflix also has a higher monetisation efficiency, measured by revenue per engaged minute. However, a key nuance to note here is that Amazon’s ability to be indifferent towards profitability of its streaming arm limits pricing power for Netflix, which might be a concern. Amazon’s streaming arm is seen as an ecosystem complement, rather than a profit generating machine. It is a funnel for their ads.

Extract 3: Average revenue per engaged minute by streaming players.

In the case of Netflix vs Disney, Netflix’s strategy has been different: focusing on high volume content to reduce consumer’s churn, increasing reality TV show output to serve as fillers to improve retention, and niching down on local markets to drive growth in domestic industries, with increased content spent over time. Disney, in contrast, has reduced content spend, and largely copied Netflix in terms of strategies. Disney have been branching out their existing IP and franchise, while also heavily licensing sports, possibly leading to content fatigue, substantiated by higher churn rates.

On local content spend, Disney, for example, has a forecasted content spend of 24B for 2026, which is down from the 33B the company spent in 2022. Half of the 24B is allocated for sports, with the other half allocated for entertainment, and a subset of that being on local content spend. Netflix’s content spend is closer to 18B in 2025, with more than 50% allocated for local content spend. This allows for specialisation and niche content, driving international subscriber growth, with more than 70% of Netflix subscribers coming from outside North America, compared to Disney at 55%, with also, a greater range of countries.

Extract 4: Netflix’s Subscriber Data

Extract 5: Disney’s Subscriber Numbers

(3) Competition with Youtube

The SVOD industry faces competition with other streaming alternatives, such as Twitch, Youtube, even Instagram and Tiktok. However, the pushback here is that Netflix’s business is enhanced by Youtube. Netflix’s goal is to channel new subscribers on their platform, and what Youtube allows them to do is to provide the initial discovery, with post show clips to ensure retention for future seasons. These two companies are also not necessarily mutually competing, with overlapping shares of subscribers. Stronger arguments can be made on Netflix and Youtube collectively taking away share from other forms of entertainment. The decline in Linear TV, for instance, provides a strong case in point.

Extract 8: Netflix’s Share of Watch Time (FY25, Q3 Report)

Comparing Youtube’s and Netflix’s business model, Netflix’s business model is more hits driven, with an initial upfront investment required, with Youtube being more UGC-linked. This can mean that there is less supposed risk associated with Youtube’s content generation. However, the pushback here is that Youtube has 55% of Ad revenue tied to TAC, to pay content creators. On the other hand, Netflix only spends about 37% of revenue on content. What this implicitly suggests, is that Netflix has much higher room for margin expansion, should the content budget remain unchanged. In contrast, Youtube’s margins will likely grow slower than that in Netflix, with the fixed 55% paid to users, and with most growth coming in from fill rate, ad inventory and CPM improvements.

Adding on, Netflix users are typically more intentional, providing brand safe environments with predictability of content alignment, allowing for greater CPM for advertising, which will be discussed later.

(4) Netflix’s Moat

Netflix’s moat can be best explained with (1) Scale and (2) content IP.

Netflix is significantly larger in terms of subscriber count with 325M subscribers, relative to its closest competitors in Disney and Amazon, with 200M and 240M subscribers respectively. Distribution and scale allows the company to reduce cost of content per subscriber over time, with about $18B content spend amortised over 325M+ subs globally. This is significant, as compared to Disney+ at $24B content spend, and Amazon at $15B content spend. Netflix has also delivered strong returns on incremental content spend, with earnings growth significantly outpacing content spend growth.

Besides Disney and Amazon, this also serves as a barrier for new entrants, because the cost of entry is significant for new entrants. This, along with Netflix’s superior profitability relative to the other firms, allows Netflix to spend more, while also maintaining a lower cost per subscriber and attracting more subscribers as the spend grows, creating a continuous flywheel effect. With greater subscribers, data on content preferences can be more easily gathered.

Content spend as a percentage of revenue has been steadily declining over time, from 60% of revenue in FY21 to 38% of revenue in FY25. Free cash flow (FCF) margins have also been steadily increasing over time, from 5% FCF margins in FY22 to 20% FCF margins in FY25. With amortization expenses similar to content spend, Netflix is transitioning from a capital intensive to a more capital light scaling phase. With content reinvestment stabilising, the company is increasingly turning into a cash-generative business. While FCF margins have expanded meaningfully, there is optionality for growth through pricing power, ad monetization and subscriber growth.

Besides scale advantages, Netflix’s content IP is significant and will grow over time. Although films and shows are amortized over typically a 4 year period, Netflix owns the content IP to the shows even after the 4 year amortization cycle. Over time, as the number of shows grows, the value proposition for customers will grow as well, increasing the incentive for subscriptions. Competitors can spend $100M on content in a year, but they cannot recreate catalog of shows over the past 15 years.

(5) Netflix Thesis

Going over Netflix theses, the markets view Netflix’s ad business as one that will only drive incremental revenue, but we believe that Netflix’s ad-driven business will become a primary engine for both margin expansion and revenue growth, becoming a key Netflix earnings driver. Negative sentiments relating to the WBD acquisition has also driven the stock price down to a more reasonable level, providing an opportunity for long-term investors.

(1) Ad tier subscription is under-appreciated and will drive high margin growth over time.

Ad revenue is largely a function of: Fill rate x ad load x CPM x number of viewers in thousands. CPM (Cost per mile) refers to the amount that Netflix charges per 1,000 viewers for an ad. This is largely driven by the type of content, the company’s pricing power, and demand supply dynamics. Fill rate refers to the amount of ad inventory that actually shows an ad.

About 65% of new users on Netflix last year have signed up for their ad tier, showing strong ad tier growth. In comparison, Disney is closer to 50%. Amazon prime video is closer to 80%, with users automatically signed up for the ad-tier unless they opt out. Amazon actually has an advantage in this, because the company has the ability to funnel viewers on Prime video to their Amazon platform, basically putting Amazon marketplace ads on Amazon Prime video, allowing for greater reach.

Currently, Netflix has a fill rate of about 40%, largely because the company does not have its own ad tech yet to provide optimization for advertisers. Netflix, over the previous few years, has largely relied on Microsoft’s ad tech, and in late 2025, the company announced the development of their own “Netflix Ad Suite”. Besides having greater control over its own data and optimisation for advertisers, Netflix is also opening up their ad inventory to make it more accessible for advertisers. Over the recent years, Netflix has largely relied on sales reps to fill their ads, with large enterprises and advertisers needing to contact a Netflix rep directly to pay for an ad. Netflix has now opened up their ad inventory to many demand-side platforms (DSP), like The Trade Desk, Amazon and Google, allowing for greater fill rates. Netflix is transitioning from a limited liquidity platform for ads (Microsoft exclusive) to a highly liquid automated auction. Based on historical precedents from Hulu, this shift typically leads to a significant increase in fill rates.

Netflix has superior ad inventory as compared to Youtube that command a higher CPM, because shows played are predictable, and of a higher quality, providing for greater general brand alignment. This is in contrast to Youtube, where video contents are usually user-generated, resulting in lower CPM in general. Further adding onto this, the company currently only has about 12 ad supported countries, supporting the point that the ad revenue now that the company has is relatively underpenetrated. In terms of types of ads, there are typically two types of ads - performance ads and brand ads. Performance ads are what Meta and Amazon do best, nudging consumers towards the call to action, to buy a product. Brand advertising, on the other hand, is more for brand awareness, and is typically pushed by larger firms. It’s typically better to have a mixture of both, and this will likely depend on their ability to expand their gaming division.

Netflix currently has 94 million MAU under their ad tier, with an estimated 1.25 hours average watch time per paying customer. (This is assumed to be lower than the average premium tier subscriber at 1.5-2 hours, due to lower monetary commitment) Assuming 5 minutes of ads per hour of watch time, and 30 seconds per ad, this implies 56 billion ads per month played to users, if fill rate is at 100%. Assuming a more reasonable optimal fill rate of about 75% (contrasted to Youtube’s fill rate at 90%), the company will achieve 50.76 billion ads per month played to users. With a CPM of $30, the optimal ad revenue will be $9.5B. For context, Netflix’s current ad revenue is at $1.5B, with management expecting a double in 2026. The current discrepancy is due to lower fill rates, and lowered CPM due to the lack of ability for targeting of ads.

To analogise, Youtube took 3-5 years to significantly increase their fill rate in the early 2010s, at a time where digitalisation was still poor. Our belief is that Netflix will be able to do so in a much shorter time duration with much significant digitalisation today, similar to how Tiktok did it in 2-3 years. Their expansion to DSPs is a significant move by management, it effectively provides for an expansion of bid density, and available advertisers, which in turn drives CPM and fill rate. Reduced friction drives fill rates, while premiumisation of the Netflix brand will help to normalise towards a high CPM number.

Top down approach for confirmation: The TV ad market is currently a $90B TAM in the US. With Netflix’s share of TV watch time at 9% of total TV watch time, the ad dollar they receive should converge to their share of attention, at $8.1B.

The greater ad fill, premium content allowing for greater CPM, along with Netflix’s continued expansion of this ad offering will allow the company to spearhead ad growth over time. Management is planning for a global ads push in late 2026. This is smart by management, testing the demand first in key countries before a full expansion. The company has a “crawl, walk, run” strategy. We are currently in the walk phase for this ad roll out. Currently, the ad tier is providing a lower average revenue per subscriber relative to the premium tier. The key here is to determine whether ads are accretive to ARPM, or dilutive to ARPM.

With the global rollout of the ad tier, we can expect both growing subscriber count, along with the optionality for substantially greater average revenue per user for their ad tier. Long term, ads will be a margin accretive business for Netflix, with every additional ad having minimal additional cost. While ad tier is still dilutive to ARPM globally, we expect the gap to close, with ARPM already higher than Netflix’s standard plan in the US.

Sensitivity Analysis:

Extract 9: Netflix’s Ad Revenue Sensitivity Analysis

In terms of ad dollars, there are a few strong competitors in the space, including Meta and Google. There are typically two forms of ads that a company can provide - brand advertising, as well as performance based ads. In the case of Netflix, given the premiumisation of the brand and with a large share of its view time coming from TV users, we expect Netflix to lean into brand advertising, rather than performance-based advertising, which are heavily dependent on Return on Ad Spend (ROAS). Brand advertising, on the other hand, is more focused on increasing visibility of a large brand, rather than increasing ROAS. So in the case of competitors, we expect Netflix to be competing against traditional media, as well as possibly Google, not so much against Meta.

(2) Negative sentiment in the stock provides for an opportunity to add, with significant margin expansion on the horizon that are not priced in.

Netflix can expand margins in primarily two ways, by reducing costs, or by accelerating ARPM. The company sets the useful life of these content assets at about 4 years, with 90% of amortisation coming within those 4 years. Based on the report released by Netflix for 1H 2025, Netflix originals that are more than 4 years old made up about 15% of total watch time across all shows, including Netflix originals and licensed shows, (14,124,700,000 Hours vs 95,186,500,000 hours) compared to 8% in 2024. Incremental viewing of fully amortised carries minimal cost, which enhances operating leverage. Shows fully amortised may still generate substantial watch hours and engagement for the company, and hence ad revenue with ads rollout, coupled with reduction in churn with sticky retention. For instance, content budget has grown at a 5.5% CAGR over the past 5 years, compared to revenue at 15%. The streets are pricing in a 200bps increase in operating margins yearly, implying a continuation of the 5.5% CAGR in content spend, and about 13% revenue growth. The implied 200bps operating margin expansion is modelled by revenue growth outpacing content amortisation growth, suggesting limited upside priced in from incremental ad monetisation. If ad monetization scales meaningfully, operating margins could exceed current consensus projections.

(6) Catalysts

There are two main catalysts for the firm: (1) Price hikes and (2) optionality with Netflix pivoting its strategy.

(1) Price Hikes

Extract 10: Subscription prices Extract 11: UBS & Nielsen

Surprisingly, despite Netflix’s premium experience, the company charges lower prices for its streaming services compared to its peers. Netflix has always led with pricing, with peers following suit with greater percentage hikes after the initial Netflix hike in price. We expect Netflix to close the gap in pricing, and a $1 hike in subscription prices with 325M subscribers will drive significant incremental revenue and hence income for the company. Assuming 90% incremental margins, an additional $3.5B flows to the bottom line, allowing for 7.8% y/y growth in earnings from FY25 numbers just from price hikes.

(2) Free platform for fully amortized shows

Management has hinted at launching free channels and releasing fully amortized shows, along with FAST channels (Free Ad-Supported Television), to form the top of the funnel. Experiments have been done in Vietnam and Kenya previously, and the conclusion was that the monetisation infrastructure (ad tech), was not mature enough to fully monetise the free channels. With Netflix building out their ad tech, the current stance might be different. Further adding on to this, a free version also partially offsets revenue loss related to piracy, and strengthens Netflix’s position relative to free user-generated alternatives like Youtube or Twitch. As long as shows released for free are those that are not-dilutive to the premium brand, introducing this will bring in more impressions to drive subscriptions.

(7) Netflix’s key risks

Competition: Despite strong engagement metrics, Netflix has been losing some of its share over time. The number of Netflix shows that make the top 10 most viewed shows in the SVOD industry has been declining gradually.

Extract 7: Top 10 Originals Data

My pushback is this: Fewer top 10 shows matters less as long as total hours and retention remain strong. Even without major hits, Netflix’s ability to provide less viral niche shows help to provide a “long tail” effect to reduce churn and maximise profitability. The more concerning metric to watch will be when watch time starts to fall, or market share of the SVOD industry plummets.

AI risks: The key narrative surrounding AI now is that if AI can generate high-quality shows cheaply, content becomes commoditized, resulting in Netflix losing pricing power and hence subscriptions become unnecessary. Content oversupply can ultimately lead to lower pricing power for Netflix.

Pushback: Netflix has not exercised significant pricing power. In fact, based on the price paid per consumption hour, Netflix currently has the lowest cost for subscribers. Content has long been commoditized and we have already seen an oversupply of content, with Youtube’s user-generated content, and Tiktok’s clipping short form content. Despite these, Netflix’s business has been able to continually expand, and will likely continue to do so. The key here lies in distribution and trust. Netflix has the ability to distribute its content globally, and subscribers trust that the company will provide quality content. If AI were to lower the cost of production, this could instead be a margin tailwind for Netflix. In a world where AI floods the content space, the ones with distribution will be the ones with the power, and the ones that prevail.

Illegal streaming risks: Another key risk commonly discussed is the rise of illegal streaming services. The common argument here is that if there is another alternative to the Netflix subscription that provides the same Netflix content that is completely free, then why pay the $15 subscription?

While this is valid, it is more nuanced than the statement suggests. Piracy has long been around for decades, and similar argument to previously, Netflix has been able to grow substantially over the years. Piracy is only a terminal risk to the firm if it is a perfect substitute (Substitution effect). But in reality, piracy is plagued with malware, hard to find, unreliable, inconsistent in quality. Piracy’s competitor is friction and convenience, which Netflix provides.

Netflix’s management has sued pirated streaming companies in the past, and is currently part of Alliance for Creativity and Entertainment (ACE), a coalition of 50 media companies aimed at combating piracy. Besides legal action, Netflix reduces the appeal of piracy by increasing the appeal for their platform, through increasing the offerings on their platform, and to reduce piracy to an inferior product, while simultaneously increasing the appeal for their platform for new subscribers.

Engagement peak risks: Another key and lethal risk to Netflix is engagement peak cycles, or content fatigue risks.

Pushback: Netflix has mitigated this by adopting a structured cycle of releasing shows, with occasional split-season release and live streaming of sports on the platform. Besides shows, games have also been introduced to increase stickiness to the app. Even with these strategies, Netflix cannot fully eliminate content fatigue. The only way for the company to mitigate the risk is to provide users with an experience that is value for dollar, which so far, has been working.

Saturation in broadband households: A key argument is that Netflix has reached saturation in terms of US households.

Pushback: In a saturated market, consumers often rotate subscriptions to watch shows across several streaming platforms. Netflix’s goal is to be the one service you never cancel. They have made deals with broadband services to turn Netflix into an invisible utility. This also ties into the previous argument raised on how Netflix is turning from a subscription driven growth business, to an operating leverage growth business.

Amazon competition: Amazon competes with Netflix on ecosystem optimization. Amazon’s ability to neglect profit optimization on their streaming arm structurally caps pricing power for Netflix. Amazon’s ability to funnel ad traffic to their traditional business will also be significantly stronger than that of Netflix’s advertising business.

Pushback: Amazon’s existence hasn’t structurally capped pricing power, with Netflix being the first mover to increase prices before competitors. Prime Video serves as a retention tool for Prime membership, whereas Netflix’s entire business depends on engagement and pricing discipline. Amazon and Netflix compete for advertising dollars, but they monetize different advertisers. Netflix will likely be leaning more towards brand advertising, compared to Amazon’s commerce driven performance based advertising. However, brand advertising is typically more cyclical than performance advertising, which could make Netflix’s ad business more sensitive in an economic downturn.

(7) Valuation

Base Case Estimates:

Base case assumes ads scale by 2030 driven by $35 blended CPM. This reflects moderate DSP adoption and global rollout by 2026. Base case also assumes moderate pricing growth in each country segment.

Base case assumes weak ad roll out, with low CPM in line with current CPM numbers. Bear case also assumes weak pricing growth in each country segment.

Bull case assumes stronger than expected ad roll out, with high CPM in line with historical CPM numbers, although still significantly lower than traditional media. Bull case also assumes strong pricing growth in each country segment.

Base Case Intrinsic Value:

Bear Case Intrinsic Value:

Bull Case Intrinsic Value:

Summarising Citrini’s points:

AI becomes so intelligent to the point where it substitutes white-collar labour, creating a self-reinforcing loop of wage compression, weaker consumption and financial instability. The end result here is an economic crisis.

There are parallels with the dot com crisis, oversupply of a certain something, which could lead to a bubble and a margin collapse of the financial markets. During dot com, we had an oversupply of capital, and an oversupply of infrastructure. In AI today, Citrini argues that we have an oversupply of both intelligence and infrastructure, causing the price of intelligence, and hence the price of labour, to fall.

Without getting too philosophical, when you mistake a well explained story for reality, you may be interpreting a fallacy. My core argument here, will lie on the very fact that humans have always found a way towards survival. Sounds overconfident, yes we may have an economic shock, but humans have always survived crisis, decades of wars and diseases. An AI crisis, even if it happens, should not be painted as one that is terminal to humans and the economy. History suggests that adaptation is likely, although the extent of impact varies. And I don’t think predicting the next crash, or predicting the next bubble, is particularly intellectually provoking here. While one should always prepare, predicting one without tangible signals and evidence is like throwing darts blindfolded. You might get one someday, someday..

The Research piece also singled out a few companies, namely AXP, Visa, Mastercard, Doordash.

On Visa and Mastercard, here are some thought provoking arguments: One of the points raised was that their entire business model was built with friction, and the friction can be removed once there are blockchain networks and agentic routes bypassing their payment rails. While this sounds plausible, the problem here is accountability; who is going to be accountable to the agents if there's fraudulent behaviour?

We have insurance companies, credit rating companies, auditing companies that have been around for years. They exist because someone is willing to pay a premium to offload the cost of uncertainty.

Similarly, blaming Visa, Mastercard, the banks, or the merchant if something goes wrong is possible, but unless the agents can guarantee zero fraud that humans can trust, then you don't have yet have a commercially viable product. Visa and Mastercard are no longer just commoditised "card networks", they are trust networks, rule makers and an identity layer.

A possible argument here would be compression of margin over time with the presence of viable alternatives. The key here is to weigh whether or not their VAS can make up for their potential margin compression. "Frictionless" alternatives have existed even before the blockchain - A2A transfer networks is an example. The narrative here may be a valid case of margin compression, not complete replacement, which the researchers here has tried to subtly paint.

We should not mistake a story for reality. Steve Eisman, one of the big shorts, is also long on the Mag 7s. This suggests that his positioning may be more nuanced than extreme bearish stories imply.

Note: Do note that this is an opinionated piece, make sure to do your own due diligence.

The American Tailwind is no longer a simple, predictable force. Today, as we peer into the news articles of 2026, we see a world where strategic autonomy has replaced comparative advantage as the governing principle of global trade.

If you want to understand why we can no longer ignore politics when investing in U.S. giants, you needn’t look further than the struggles of our great technology and financial cathedrals as they try to keep their doors open in Europe and beyond.

I. The "Sovereign Cloud" and the Amazon Conundrum

Consider Amazon. For years, AWS was the gold standard, a utility so efficient and scale-advantaged that it seemed untouchable. But in the current political climate, efficiency is being sacrificed at the altar of sovereignty.

In 2025 and 2026, we’ve watched Amazon perform a fascinating, expensive dance to convince the European Union that its data centers are European enough. The EU, wary of U.S. laws like the CLOUD Act, which theoretically allows the U.S. government to subpoena data held by U.S. companies anywhere in the world, has effectively told Amazon: "We love your tech, but we don't trust your passport." 

To stay in the game, Amazon had to launch the AWS European Sovereign Cloud. Look at the concessions they’ve made: 

1) Physical Isolation: Entirely separate infrastructure from the U.S. regions. 

2) Personnel Restrictions: Operated and supported exclusively by EU residents located in the EU. 

3) Governance: A separate legal entity to buffer against U.S. judicial reach. 

When you have to build a separate factory for every country just to prove you aren't a spy, it reduces your scale advantage. It’s a massive duplication of capital that would make a rational businessman wince. For the investor, this is a political tax on earnings. If Amazon has to build bespoke, sovereign silos for every major trading bloc, the margins of the cloud business will inevitably feel the squeeze.

II. The Revolt Against the Card Giants: Visa and Mastercard

This sovereignty movement isn't limited to data; it’s hit the very plumbing of global commerce. For decades, Visa and Mastercard (V/MA) enjoyed a duopoly that has benefited them immensely. They are the toll booths on the global highway. 

But the politicisation of the U.S. dollar, specifically the use of financial sanctions as a primary tool of American foreign policy, has sent a chill through Brussels. The EU realised that if they are 95% dependent on two American companies for their daily bread and coffee transactions, they aren't truly a sovereign power.

In response, we are seeing the aggressive rise of the European Payments Initiative (EPI) and its digital wallet, Wero. Launched in late 2024 and gaining massive steam through 2026, Wero is the EU’s attempt to build its own "toll booth" and kill its reliance on V and Ma. We see major European banks like Deutsche Bank and BNP Paribas urgently migrating customers to these sovereign rails. Whilst Guanrui and I believe that the impacts are likely to be minimal on V/Ma’s business, it is undeniable that regulation support of wero will provide a headwind for V/Ma.

III. The AI Export Trap

The U.S. currently leads the world in AI models and hardware. Naturally, the current administration wants to use this lead as a geopolitical lever. However, given the current political climate, I highly doubt that countries in the EU are willing to embrace American AI as willingly as they did from embracing US tech. We can clearly see the pushback of the EU on big tech companies and the increases polarisation and tension between the EU and Trump further threatens the ability of US AI to diffuse internationally .

By tying AI adoption to U.S. political alignment, we are inadvertently encouraging the rest of the world to build Anything But American (ABA) AI stacks which will be detrimental to the Big tech companies should their 2nd largest market refuse to adopt their new AI tech stack for fear of being held hostage in the future due to their dependency as they are being held hostage now.

If a country in the Global South or even a close ally in Europe fears that their AI "brain" could be switched off by a U.S. export ban during a trade dispute, they will choose a slower, domestic alternative or a Chinese one.

By politicizing AI standards, the U.S. is fragmenting the global market. In business, you want your product to be like air, something everyone needs and no one thinks about where it comes from. The moment you make your product a political statement, you've given your competitors the best sales pitch they’ve ever had.

The U.S. is currently using its corporate champions as the front-line soldiers in a global geopolitical struggle. While this might be good for national security, it is poison for the shareholders.

When Amazon has to allocate extra resources for the EU and when Visa sees its rails being dug up by European bureaucrats, the intrinsic value of these businesses must be discounted. The American Tailwind is still there, but it’s now blowing through a series of political toll gates that didn't exist ten years ago.

We still love great businesses. But we love them even more when they don't require a diplomatic mission to access to European market. In an increasingly polarised world, we have to take into account the political risk in our evaluations instead of turning a blind eye and discounting them. The intersection of politics and the boardroom is the new high-wire act of the markets. Our advice remains the same: Look for the companies that can survive even if a moron (or a particularly misguided politician aka Trump) is running the country. At current valuations it might be wise to diversify and invest in other high quality companies across the globe or invest in companies that will benefit from this de globalisation, nationalistic trend.

Feel free to look at the writeup of Moutai as a potential candidate for a high quality business outside the US trading at a decent valuation.

Introduction

Visa is one of the most dominant global payment networks, with a market capitalization exceeding $600 billion as of January 2026. Visa originated from Bank of America’s BankAmericard program, which launched mass consumer credit card distribution in California during the 1950s. While it was a poor credit event, it created this idea of a self-fulfilling loop, where sufficient merchants and card holders will incentivise more merchants and card holders to join.

In the 1970s, Visa was spun off as a separate entity, and handled its first electronic authorization in 1973. In 2008, Visa went public in an IPO that raised approximately $17.9 billion

Understanding the company’s business model and moat

Visa has several distinct revenue streams, namely: (1) Data processing revenue (2) Services revenue (3) Value added services revenue (VAS).

Data processing revenue: This is primarily linked to the authorization fees, which is a fixed fee of approximately $0.195 per transaction. This is a fixed dollar amount per transaction. Some portions of Visa’s VAS are also embedded in this. This represents approximately 35% of gross revenues.

Service revenue: This is related to the assessment fees, 0.13% or 0.14% of each transaction charged to acquirers and passed on to merchants along with certain VAS solutions. This represents about 30% of gross revenues.

International: This is related to cross border assessment fees and currency conversion fees. This is about 25% of revenues and Visa charges ~1.5%+ on cross-border transactions on aggregate, made up of cross border assessment fees and currency conversion fees.

Value Added Services: Visa consulting, analytics, license fees for using the Visa logo, and marketing fees. This is about 7% of gross revenue. VAS is underreported in the revenue segments because many services are embedded in data processing and service revenues, making up about 24% of total net revenue.

To understand Visa’s moat, we must first understand how the entire payment sequence works. Below are the specific players, along with the steps for the entire payment process.

There are 6 main players in an entire payment process:

Merchant: This is the store, e.g. Walmart, Lululemon etc

Merchant’s bank: This is the store’s bank, otherwise called an acquiring bank/acquirer.

Consumer: The cardholder

Consumer’s bank: This is otherwise called an issuing bank/issuer.

Payment rails/networks: Visa, Mastercard, Discover, Star, NYCE

Payment processors: These companies send the initial data to the merchant’s bank, and issue the point of sale systems (POS).

The steps are as follows:

1. Authorization: When the card is swiped or tapped, transaction data is sent from the POS to the merchant’s bank. The merchant’s bank will ask Visa/Mastercard to find the cardholder’s bank. Visa and Mastercard will then route the request to the consumer’s bank via Visanet (their infrastructure). The consumer bank checks for fraud and credit limits, then puts the funds on hold. This approval will then be transferred back to the network and to the payment processors, with the “approval” signal on the POS system. [This is where Visa charges their authorization fees, swipe fees]

2. Clearing (The receipt): At the end of the day, your merchant account will send a batch of transaction details to Visa and Mastercard, and Visa and Mastercard will sort it and send it to the different issuing banks. Visa and Mastercard will also settle net obligations (refunds and total payable amount), providing the information to both the merchant and issuing bank via Visanet. [This is where they charge their assessment fees]

3. Settlement: Consumer bank transfers money to the network, and the network transfers money to merchant’s bank.

Visa functions as the central routing and rules-based intermediary is highly significant, serving as a router that connects the issuing banks to the acquiring banks.

In terms of the cost structure and how much each player takes in this entire chain here, a summarised view for a credit card transaction shown below:

Consumer’s bank: Almost the entire interchange fee, with interchange fees normally set at between 1.5-2.5%.

Merchant’s bank: Between 0.05% and 0.20% of the total transaction volume.

Visa: About 0.14% of the total transaction amount.

Understanding interchange fees and who pays for them, and how it is beneficial to the consumer, the bank, and Visa.

Interchange fees are charged to merchant banks who pass it on to merchants and received by issuing banks, with fee rates set by the payment networks. Visa and Mastercard set different interchange fees for the different types of cards, different merchants etc. When issuing banks are contemplating which payment network’s cards to issue, this is one of the main factors for them. Because the higher the interchange fees for the cards, the higher the issuing banks get to keep, with their fee revenues tied to a percentage of the total interchange fees charged to the merchants.

Higher interchange economics allow issuers to fund greater consumer rewards, reinforcing card usage and issuer preference, while Visa and Mastercard act as the middleman to continue supporting wide scale merchant acceptance. Because high ticket transactions account for a disproportionate share of payment value, the loss of rewards materially affects consumer behaviour despite representing a small share of transactions.

The point here is not only that interchange fees incentivises the issuer and the consumer, but more importantly, it substantiates the point of Visa being a rule maker and that their rules create a self-fulfilling loop for the continued usage of their infrastructure. For the issuing bank and consumer side at least, it is growing increasingly difficult to switch away from Visa and Mastercard, with a scale and a network effect that is only getting larger by the day. Contracts further reinforce this point, with major banks being locked in to issuing cards from only 1 network (Visa), typically for 7-10 years. That could be the reason for higher client incentives in recent years, with many contracts up for renewal.

To touch on merchants, they are the ones that suffer the most with higher interchange fees, because those fees directly come out from the revenue that the company collects. For merchants, there are typically only a few options. (1) Reject the sale, which means they lose all revenue. (2) They pressure consumers to use a card from different networks that have lower interchange, such as Discover, Star, NYCE, maybe even American Express. (3) They ask for payment in cash or via another method of transfer. [which will be discussed later]

(2) seems the most viable, and this leads on to the second important portion of Visa’s business, apart from the standard authorization fees and assessment fees that they collect (under data processing and services revenue) - which is their Value Added Services, that locks in merchants.

Value Added Services (VAS) has become an important component of Visa’s entire business, with total VAS driven revenue currently at about 24% of total net revenue per Visa’s CEO.

The current VAS services reported under their VAS line item in their consolidated revenue streams relate to consulting and analytics services that the company provides. It is currently about 7% of the total gross revenue. The more significant VAS products that Visa provides are embedded in data processing and services revenue, with the two primary products being fraud detection and tokenisation. The revenue mix for tokenisation is as such: Issuing Solutions (40% of VAS), Acceptance Solutions (28% of VAS), Risk & Security Solutions (17% of VAS) and Advisory (15% of VAS). VAS has seen significant growth in recent years, and it has been a key growth driver for the firm, with about 60% of incremental growth now attributed to VAS and New Flows.

Issuing Solutions (40% of VAS) refer to software solutions that Visa provides to banks, with transaction controls (allowing you to freeze your Visa card), subscription managers (manage your card subscriptions on the bank app), and management of card accounts. Most importantly, Visa makes use of the data that they collect from transactions, and offer them to banks under this segment, allowing banks to gain insights into card-linked offers and provide APIs for notifications related to reward redemption for consumers.

Tokenization is the idea of using tokens for specific purchases at specific merchants for specific devices. Two tokens cannot be swapped and used at two different merchant places, reducing the ability for fraud. Tokenization falls under the Acceptance solutions segment of VAS.

Tokenization is not a payment method, it is a management system similar to managing passwords, that can be used for payments. Tokenization increases authorisation volumes, reduces fraud. The issuing bank pays for generation of tokens and lifecycle of tokens. Merchant banks pay to request for a token, while the merchant pays for usage. Tokenization is also important to agentic commerce.

Because Visa sets the rules for the usage of their networks, tokenization has become an extra monetisable pathway for Visa, with Visa increasingly charges (Secure credential framework at 7.5bps) if tokenisation is not used, while also providing more value by reducing fraud and providing better “life cycle” management of customers. For instance, if a merchant, e.g. Netflix, has a consumer with an expired card, tokenization allows for continued recurring payments.

Tokenization replaces PANs (16 digit numbers). When a card is added to a wallet or merchant, the Visa Token Service generates a token that is linked to the PAN that is stored with Visa. During payment, this token is retrieved and used, and Visa detokenizes it (changes it to a PAN) before routing the transaction for authorisation. Tokenization reduces fraud, improves authorisation rates, and reduces the dependency of physical card payments, with increased usage in digital form.

Visa Advanced Authorization (VAA) is an internal scoring system, another one of the company’s offerings, to help prevent fraud, and reject illegal forms of payment and transactions. VAA falls under the Risk & Security Solutions portion of VAS.

The tokenization TAM is still large, with about 50% of ecommerce transactions still yet to be tokenized. Authorization lift was about 5% in 2024, and fraud rates decreased by 8% during the same period. Greater authorization rates leads to better economics for all parties involved, with increasing authorization leading to greater sales for merchants, and all banks involved, along with greater payment volumes for Visa. Furthermore, the cost of fraud is typically a lot higher than the cost of fees from Visa and Mastercard, with the cost of fraud often higher than the sale of products, making it a top priority for merchants.

With increasing risks of fraud and increased technological advancements, the need to reduce fraud is a secular tailwind that Visa is riding on. 75% of merchants acknowledge that fraud protection is their highest priority. Visa’s unique power dynamic allows it to upsell products outside of its core offering. Expansion of tokenization outside of cards will come in the form of bank account numbers, to indirectly monetise A2A transfers - an alternative to Visa that will be covered in a later part.

In terms of fraud rates and comparisons, the Federal Reserve 2025 report, dual-message transactions (primarily Visa and Mastercard) had fraud levels at 17.9 basis points, compared to single-message transactions (primarily STAR, NYCE) at 6.7 basis points. To put it simply, $17.9 is lost per $10,000 transacted for dual-message transactions, compared to $6.7 for single-message transactions.

This paints a weak image for Visa’s fraud protection, but a key thing to note here is that a large proportion of Visa’s transaction volumes are online, which typically has significantly higher fraud rates. Card-not-present (CNP) transactions for dual message transactions are currently 61% of total transaction value, compared to single-message transactions at 7% of total value. Fraud rates for CNP transactions are typically 7.56x greater than non CNP transactions, at 41.6 basis points and 5.5 basis points respectively.

Assumptions:

• The 7.56× CNP/non-CNP fraud multiple is applied uniformly for illustrative purposes.

• Actual CNP fraud likely differs between Visa and STAR/NYCE due to differences in transaction type, geography, and merchant mix; therefore, derived non-CNP fraud rates should be interpreted as implied rather than observed values.

The above table implies that derived non-CNP fraud rates for Visa are lower than STAR, even despite STAR/NYCE’s greater use of PIN authentications which further reduces fraud rates. However, because CNP transactions differ across the networks, this result should be viewed as an implied set of results and should not exclusively be interpreted as evidence of superior non-CNP fraud performance.

In sum, with increasing risks of fraud due to technological advancements, Visa is offering VAS to benefit from this tailwind.

Understanding regulations imposed on the firm, with potential upcoming regulations

A key risk for Visa comes in the form of regulations. That is why there is a key distinction between credit and debit card rewards.

The Durbin Amendment of 2010:

The Durbin Amendment in 2010 pushed for a cap on debit card fees, which effectively killed debit card rewards. The ceiling imposed was an interchange fee of $0.21 + 0.05% for US banks above $10B in assets, with the EU doing something similar with a 0.2% cap in debit card interchange fees. With interchange income reduced, banks largely lost the economic incentive to fund debit rewards programs, leading to their widespread elimination of rewards.

The Durbin Amendment regulated that all cards issued by banks above 10B of assets must have 2 different rails tied under one card. In theory, this gave merchants the ability to route toward lower-cost networks, though in practice this option was limited by POS defaults and network acceptance constraints. Impact was further mitigated by e-commerce, because only Visa and Mastercard pinless rails were supported for ecommerce. Other rails required pins for purchases, which was deemed as a security problem. This structural limitation that benefited Visa and Mastercard was only curbed in 2023 through Regulation II.

In response to the new regulation, Visa reduced assessment fees for banks to continue the encouragement of Visa cards issuance, while expanding merchant and brand usage fees. They also introduced administrative fees and brand usage fees for the merchants. Despite the regulations, while banks reduced their push for consumer debit cards, Visa’s debit card payment volumes in the following year rose by 14% yoy in 2011, outpacing credit cards at 7% yoy, though credit card growth saw a great jump in growth. The company’s overall revenue grew by 13% yoy, with US revenues up 8% yoy. Net revenue yield fell from 0.26% in 2010 to 0.24% in 2011 following the amendment, recovering to 0.25% in 2012.

The Durbin amendment reduced Visa’s revenue per debit transaction, but Visa mitigated this through fee adjustments and volume growth. This demonstrates Visa’s resilience, allowing it to adapt despite regulatory changes. Regulatory pressures can be interpreted as threatening the margins for the firm, rather than an existential one.

With about half of US payment volumes tied to debit, it is intuitive to assume that the amendment would materially hurt Visa’s economics. Critically, Visa’s fees are not tied to interchange fees, and their economics improve with volume growth rather than issuer interchange economics. While the Durbin Amendment lowered the attractiveness of issuing debit cards for issuing banks, the consumer continued to spend on the card.

This has led to a strong distinction in terms of the margin profile for their debit vs credit business. Credit transactions are currently the more margin accretive one, due to higher allowable interchange fees that indirectly feeds to Visa’s bottom line.

The Credit Card Competition Act (CCCA)

The CCCA has been gaining much more traction lately, with Trump’s “official” endorsement on Truth Social. The CCCA implementation is similar to that of the Durbin Amendment, and mandates two unaffiliated networks in each credit card, so Visa and Mastercard combinations are no longer allowed. This theoretically allows alternative rails such as Discover, STAR, and NYCE to compete on price, though only Discover currently has the infrastructure to support credit transactions. Critically, STAR and NYCE have yet to build out their credit networks, as they only offer single message networks suited for Debit card transactions. Significant investments will need to be made, to support dual message transactions for credit cards.

The CCCA breaks card exclusivity. Currently, Visa has exclusive agreements with issuers, shifting the choice of networks to issuers and consumers. With the inclusion of a second network, the decision shifts to the merchant, leaning them towards lower cost alternatives that provide adequate fraud protection and authorization rates. With the top 10 banks making up more than 80% of the existing credit cards, and with all of them having more than $100B in assets, most consumers will see their issuer funded rewards affected, if the transaction is routed towards a separate network with lower fees.

While CCCA indirectly affects Visa - because the presence of feasible alternatives with lower interchange fees will naturally lead to lower Visa interchange fees - the impact is mitigated by Visa’s unique offerings. In today’s age, Visa’s offerings in the forms of Tokenization and Visa Advanced Authorization will help to maintain merchant’s stickiness, with bundled pricing. Visa’s strategy is less about monetising volumes outside of Visanet directly, and more to make alternative routing economically and operationally inferior, unless merchants are willing to accept higher risk, lower authorization rates, or poorer user experience.

Adding onto this, should merchants decide to push away Visa’s cards, there are mitigating revenue streams that Visa can tap on to continually drive its revenue. The new CCCA act does not lead to reduction in Visa cards introduced. Instead, it recognises two networks in a single card, one from Visa and another from Discover/STAR/NYCE - likely Discover. Discover is the only immediately viable secondary credit rail under CCCA, but its ability to meaningfully compete with Visa depends on its ability to incentivise issuers and for merchants to willingly accept lower consumer usage of Discover cards. Adding on to this, Discover cards are typically directly controlled by Capital One. This control limits Discover’s appeal to external issuers, due to lack of control over rules and data for the cards.

Visa Network; Interchange fee of 1.15% – 3.15%

Discover: Interchange fee of 1.56% – 2.40%

As such, cost savings for merchants can be viewed as potentially incremental, though the economic viability of this is still in question. The practical benefits will be primarily through long term pricing pressures rather than actual cost savings.

Let’s assume a scenario whereby Visa loses 50% of its existing credit card volumes in the US to competitors like Discover, or potentially STAR/NYCE when they build out their infrastructure in the future:

50% choosing an alternative payment network x 48% of total volumes global volumes x 44% credit card volumes in the US x 96% of total domestic volumes = 10% of total volumes get hurt.

This represents an aggressive bear scenario for instant volume erosion rather than a base case scenario, and with the stock falling more about ~10% since Trump’s endorsement, the market might have already reflected the regulation into valuation, even without the regulation playing out.

In short, the CCCA implementation lacks feasible alternatives to Visa and Mastercard, and the market reaction has shown that most of the drop in fundamentals have already been priced in. Merchants will route transactions where risk is lowest, using alternative rails as a way to negotiate with Visa rather than a full replacement, which pressures margins. Although it is currently framed as a regulation that will cut immediate cost savings, its practical benefit is modest.

The EU and their regulations:

The EU has long been critical of Visa and Mastercard’s interchange fees, which are viewed as driving up costs for merchants. Over the years, the EU has imposed caps on both credit and debit cards. In 2015, caps on credit and debit cards imposed were at 0.3% and 0.2% respectively for cards issued and used in the EU. For cards issued outside of the EU and used inside the EU, relevant caps were imposed as well.

The EU have now shifted their stance from limiting Visa to replacing Visa and Mastercard. Network alternatives such as Wero, a government-backed real time payment system that is receiving wide scale adoption.

Risks and alternatives

Alternatives to Visa include (1) A2A transfers, (2) Closed looped alternatives, (3) Mastercard and Discover, (4) BNPL, (5) Stablecoins.

(1) A2A (Short form for account to account), is a transaction method that bypasses the traditional payment networks, and transfers directly from bank account to bank account. There are multiple forms of A2A transfer methods, but the main one discussed will be Government-backed real-time payment systems (RTP).

Government RTP systems such as Pix (Brazil), UPI (India), Weri (EU), FedNow (US), and PayNow have enabled instant, low-cost payments and eliminated traditional interchange fees through flat, minimal pricing structures. RTP networks are instant, allowing immediate transfer of funds from one bank account to another bank account, compared to Visa’s typical merchant settlement speed of about 1-3 business days. However, there are caveats to this as well. RTP’s instant and irreversible nature significantly limits post-transaction recovery, particularly in authorized push payment (APP) scams, which are scams where the victim transfers money to the scammer. Visa transactions, on the other hand, benefit from structured disputes and chargebacks, with an allowable dispute period of up to 120 days after the initial transaction. This increases the perceived reliability of Visa relative to RTP networks, especially for high value transactions, making it attractive to larger merchants and higher-income spenders who prioritize protection over transaction speed. [RTP excel in low-ticket transactions]

To counter this, Visa has created their own A2A transfer network - Visa Direct - which operates on Visanet, and is integrated with Visa Protect. Visa’s VAS is also operating on these government rails, with Visa Risk Manager (VRM) and Visa Advanced Authorization (VAA) being updated to be “network-agnostic”, allowing it to operate on all networks. This can be seen as a way for Visa to further capture the economics in this high growing segment, allowing Visa to monetise transactions that bypass the traditional Visa payment network.

While one can argue that Visa’s A2A offering is cannibalising their core business, one should also consider consumer behaviours and incentives. Besides security, another key consideration for consumers when it comes to deciding their payment method are the rewards that they will receive from transactions. Credit is still the preferred mode for high ticket transaction. Losing the rebates and cashback for high ticket items will often be seen as a loss for them. High value merchants and transactions are low in volumes, but contribute disproportionately to transaction value. An extrapolation of this trend can be seen in Australia, where merchant-level data from the Reserve Bank of Australia indicate that merchants processing more than $10 million in card transactions, accounted for roughly 40% of total card transaction value. This is similar to the concept of the wealthy making up the majority of transaction value, and the concept of a k-shaped economy. The true appeal of A2A transfer lies in low ticket transfers, due to higher speeds and convenience.

(2) Closed looped alternatives include merchant specific payment cards and American Express in the US. Merchant specific payment cards allow large scale merchants to save on cost, by issuing cards and giving rebates on them. Target, Starbucks and other big merchants have hopped on this to save cost. Closed loop cards typically restrict refunds to store credit instead of cash, and consumers give up flexibility of a general purpose card with similar rewards tied to them. Closed looped merchant cards can be seen as complementary, rather than a flat out alternative.

American Express is another closed loop alternative worth noting. American Express targets a different set of consumers, those that are much higher income than the typically Visa card user. Visa wants every dollar that is transacted, where American Express’ main focus is on higher ticket transactions. American Express’ brand is often associated with exclusivity and prestige, allowing it to charge merchants even higher interchange fees. While American Express competes with Visa in the premium credit card division, they pursue different objectives. Visa prioritizes large scale acceptance and transaction volumes across all income levels, whereas American Express focuses on higher-income consumers and higher-ticket transactions.

(3) Discover and Mastercard

Discover is a competitor to Visa, with a market share of about 4% in the US. As discussed in the previous segments, Capital One’s acquisition of Discover further disincentivised other competing banks to issue Discover cards, as doing so would indirectly support a competitor’s network. In contrast, issuing Visa cards is more attractive to banks due to Visa’s broader acceptance, higher transaction volume, and stronger economic incentives.

Mastercard is another company that is competing directly with Visa. The key difference between the two payment networks is in the mix and geographies. Visa is more US centric, with 44% of transaction volumes coming from the US, compared to Mastercard at around 33% of total transaction volumes. Visa and Mastercard’s duopoly can be classed as a “Tacit Collusion”, where two firms in a duopoly adopt parallel pricing rather than a “race to the bottom” mentality. Adding on to this, decreasing interchange fees to compete on scale and wider spread merchant adoption is difficult, because both Visa and Mastercard operate in a demand driven payment industry, where issuers decide what cards to issue, and consumers decide what cards to use. Should any of the two reduce their interchange fees, the banks will naturally tend to push for the other card with higher interchange fees for better economic benefit.

Mastercard and Visa compete on issuer adoption, and this “competition” actually leads to higher prices and better economics for the firm, should they decide to embark on one. This is counter-intuitive and margin preserving, rather than the traditional margin dilutive competition.

(4) Buy-Now-Pay-Later (BNPL)

BNPL is still a very small segment of transaction volumes in the US, and is an additional layer on top of traditional credit cards. A common narrative is that BNPL cards hurt credit cards and hence credit volumes, but BNPL virtual cards actually run on Visa’s payment network, allowing Visa to capture transaction volume that way. Visa, in collaboration with banks, started Visa Installments, using credit cards as a means towards BNPL transactions, allowing banks to reclaim volumes that they lost to fintech companies.

(5) Stablecoins: The Genius Act institutionalised Stablecoins, regulating that banks back their cryptocurrencies with liquid assets, while selected issuers were allowed to accept payments via stablecoins. Stablecoins are fast and cheap for transactions, making them particularly attractive to merchants. The bear case is that stablecoin payments will route around the traditional payment networks. This is plausible, but two key constraints are consumer friction and trust. For one, a consumer transacting on a stablecoin wallet will need to do two things when paying for a product. (1) Transfer the funds from fiat to stablecoins - from their bank to the wallet (2) Pay on the wallet. The friction can only be reduced if banks create stablecoin linked cards, and even then, Visa’s stablecoin network still presents itself as a strong and viable option for transaction, with multiple different blockchains allowed. For consumers, the lack of opportunity for chargebacks also serves as a massive hurdle.

(6) Agents: Agentic AI may allow the establishment of autonomous payments, with security being the major requirement. Even with autonomous agents, Visa’s traditional network rails are likely to remain central to transaction routing. Visa’s ability to monetise tokenisation is something that other payment alternatives can’t. Additionally, Visa’s Trusted Agent Protocol (TAP) also provides merchants with the ability to distinguish between AI agents and fraudulent bots.

How to think about Visa’s growth floor

Visa’s future revenue drivers depend on (1) volume growth (2) pricing growth and (3) unlocking of more revenue streams. Pricing growth will be discussed in a later segment.

Volume growth has a floor in terms of growth, made up of average global GDP growth, and inflation rate, with further accelerations fuelled by increasing card adoptions in underpenetrated markets and decreasing cash share of transactions. Assuming global real GDP expansion of between 2.5-3%, with inflation at 2-3%, a rough long-term nominal volume growth baseline would be around 6% in terms of transaction volumes. This provides a reasonable baseline for long‑term nominal volume growth.

Visa generates 49% of global volumes from the US, with 54% Debit and 46% Credit. Debit has seen faster growth over the years, driven by lower barriers for sign ups without access to credit, while its lower and capped interchange fees allow for greater merchant acceptance.

Cards represent roughly 44% of global consumer transaction volume, indicating significant remaining TAM for further penetration. Payment alternatives like cash, legacy automated clearing houses, short-form ACH, run by the Federal Reserve in the US (FedACH) lack the transaction speed that normal payment networks have. Only Fednow is an A2A alternative that might have some push back in Visa volumes. Cash is expensive to transport, and it is a defenseless alternative that will continually lose share.

ACH is another legacy alternative, run by a non profit, that involves transactions in batches, several times a day. ACH charges a low fee per transaction, and is used primarily for B2B transactions, recurring consumer utility bills and for handling employee’s payments.

To capture the economics of the B2B and the P2P market, Visa has moved into Visa Commercial & Money Movement Solutions (CMS), with two sub categories. These are termed “New Flows”. Visa Commercial Solutions (VCS) refer to B2B payment networks, and Visa Direct is Visa’s form of A2A transfer networks, with Visa controlling about 40% of total carded commercial payment volumes. Visa’s CMS also allows VAS to be sold on new flows, further strengthening the ecosystem. Visa is actively monetising the transactions where speed, data, and security matter.

The B2B Market is an underpenetrated one, and dollar amount per transaction is typically a lot higher. The B2B market is currently via SWIFT/Wire payments (direct bank to bank transfers), or ACH (bulk network transfers), with ACH access only allowed within the US. This is another potential pathway to drive continued growth for Visa.

Investment Theses

(1) Visa’s increased focus on VAS provides a significant growth driver that is underappreciated by the markets today. While VAS currently makes up a smaller share of revenue compared to Mastercard, there is meaningful upside potential as VAS scales. With management’s guidance for mid-teens earnings growth, this positions Visa for strong future profitability, assuming volume growth and VAS adoption meet expectations.

VAS is the fastest growing segment in Visa’s entire business portfolio, accounting for a significant portion of incremental revenue. Visa’s current VAS is about 24% of Visa’s total revenue. For comparison, Mastercard’s VAS offering makes up 40% of total revenue, primarily due to an earlier shift to services. Visa does not need to reach 40% VAS mix for the thesis to work, moving to 30% would already be meaningful.

VAS is still a largely untapped market despite growth of over 20% from FY22 to FY24. Based on Visa’s own $520B VAS opportunity, current VAS revenues represent about 2% of that TAM. Current valuation discrepancy between Visa and Mastercard stands at ~11% (28x forward PE v 25x forward PE), largely driven by higher growth rates and higher VAS revenue mix. With continued VAS growth outpacing payment volume based growth, a further market repricing is highly plausible, especially with Visa’s wider issuer base and adoption. Based on Q1 26 numbers, Visa’s VAS services (+28% YoY) are also growing faster than Mastercard’s (+22% YoY).

VAS is a defensible business segment that is not directly correlated with consumer spending. While Visa’s core volume business was hit in 2020, VAS showed resilience and continued to grow, owing to its lower correlation to consumer spending. While traditional payment networks benefit from GDP, increasing card penetration and consumption patterns, VAS benefits from a wave of increased digitalisation risks and fraud complexity, leading to a tailwind in prevention measures. For instance, Tokenized transactions are set to double by 2029, with the fraud detection market projected to grow at 20% over the next few years. VAS services are also mostly recurring transaction or subscription based, bringing in predictable cash flows to the business.

Visa’s VAS lies outside of interchange regulation, allowing it to negate margin pressures, because the regulation of VAS can be easily argued as one that threatens security of transactions. The key reason behind Visa’s ability to monetise VAS lies in widespread trust, data advantages and their ability to monetise their tap on their payment networks to expand their offerings. This has resulted in high VAS utilisation rate per user and high revenue share of 60% derived outside of the US, circumventing US regulatory pressures.

Caveats to this include the lack of disclosure for VAS at a high level, including metrics linked to attach rates and client retention, along with the complexity and the fragmentation of each segment, making it difficult for investors to determine to a high level of certainty the true profitability of this business segment. However, the segment’s large contribution to incremental revenue growth suggests further monetisation tailwind, with Mastercard’s 40% being the upper bound.

(2) The current weak sentiment and regulatory headwinds around Visa serves as an opportunity to add into a company with a defensible business and a strong moat.

Despite regulatory headwinds and constant chatter about interchange fees, and alternatives such as RTP, stablecoins, and BNPL that have been around for years, Visa’s payment underlying business fundamentals have stayed strong. Payment yield, or net revenue yield, otherwise known as take rate, has remained relatively strong, and has accelerated in recent years despite regulatory pressures. This is a testament of Visa’s pricing ability. Visa’s payment volume has also remained strong over time, with no recorded negative growth years, despite present alternatives. This, once again, shows resilience in growth.

I view the current regulatory headwinds as one of noise rather than actual indication of worsened future economics. Trump’s recent comments on the 10% credit card interest cap, for one, has limited regulatory backing, and does not directly lead to worsened economics for Visa, as they rely on payment volumes, rather than credit loan volumes. Credit card interest cap will hurt subprime consumers, but superprime consumers (Consumers with high credit scores) are the ones driving volumes for credit cards, with this group of consumers making up about 2/3 of the total accounts and 80% of credit volumes, based on the Philadelphia Fed. The banks will also see a massive hit to their interest income, and they may introduce fees to make up for the difference, all of which will still have to go through Visa and Mastercards’ payment rails. As for the CCCA, regulation headwinds are also currently exaggerated.

Valuation & future growth assumptions:

Base case scenario assumes growth rates as follows, with most other metrics assumed business as usual:

• VAS continues growing at double digit, driving continued expansion in processed transactions.

• Payment volume will be mostly tied to GDP and inflation growth, at c. 6%.

• Take rate for services revenue is assumed to be constant despite potential margin expansion ahead, driven by Tokenization and increased fraud protection.

• International transactions are expected to continue growing at projected market rate, with pricing decline due to pressures from cross-border A2A transactions.

Bear case scenario assumes growth rates as follows, with significant slowdown in growth:

• Services revenue to grow approximately in line with inflation, at 4%.

• Cross border and Visa direct transaction growth to see significant deceleration, in line with traditional payment growth despite secular tailwinds.

Bull Case:

Sensitivity Analysis Table:

The fiscal year 2025 earnings report for Warrior Met Coal (HCC), released on February 12, 2026, details a pivotal year defined by strategic expansion and operational resilience. While the company faced significant global pricing headwinds in the metallurgical coal market, it successfully transitioned from a period of heavy capital investment to one of record-breaking production. This transition was anchored by the ahead-of-schedule launch of the Blue Creek mine, a project that has fundamentally altered the company’s long-term growth profile and cost structure by raising production volume by 75%.

The most striking takeaway from the 2025 report is the company’s ability to drive volume growth in an unfavorable pricing environment. For the full year, Warrior Met Coal achieved record production of 10.2 million short tons, a 24% increase over 2024. This surge was primarily driven by the early commencement of longwall operations at the Blue Creek mine in October 2025 tat was roughly eight months ahead of the original timeline.

This operational success, however, was dampened by coal prices that declined significantly throughout the year at the first half. While production reached new heights, the average net selling price for the company’s coal dropped 29% annually to $132.62 per short ton. Consequently, full-year revenue declined 14% to $1.31 billion, and net income fell to $57 million. Production and other financial metrics were all in line with or exceeded the guidance of the management.

For the year ended December 31, 2025, the company experienced mixed performance compared to 2024, with higher production and sales volumes but significantly weaker financial results. While operational activity increased, declining selling prices negatively affected revenue, profitability, and cash flows.

In 2025, tons sold increased by 21% and tons produced rose by 24%, indicating stronger production output and higher sales volumes. This growth suggests improved capacity utilization and operational expansion. Additionally, the company successfully reduced its cash cost of sales per ton by 19%, demonstrating improved cost efficiency and cost control measures.

Despite these operational improvements, financial performance declined due to a substantial drop in pricing. The average net selling price per ton fell by 29%, which led to a 14% decrease in total revenue to $1.31 billion. Lower pricing significantly pressured margins, resulting in a sharp 77% decline in net income to $57.0 million. Similarly, adjusted EBITDA decreased by 43%, reflecting reduced operating profitability.

The decline in earnings also affected cash generation. Cash provided by operating activities fell by 38%, and free cash flow remained negative, worsening by 43% compared to the prior year. Earnings per share mirrored this trend, with diluted EPS dropping 77%, highlighting the reduced returns to shareholders.

Overall, although the company achieved higher production and sales volumes and improved cost efficiency in 2025, these gains were overshadowed by lower selling prices. As a result, the company reported declines across key financial metrics, underscoring the significant impact of market pricing on overall financial performance.

Warrior’s strategy in 2025 also involved a pivot toward the Asian market where the demand for metallurgical goal is expected to surge driven by the steel infrastructure buildout in India. In Q4, nearly 57% of sales were directed to Asia, reflecting a strategic alignment with growing demand in the Pacific Basin. This shift came with challenges, including higher freight rates and a slightly lower “gross price realization” (75% in Q4 vs. 83% in Q3), partly due to a higher mix of “High Vol A” coal which typically sells at a different price point than the premium low-volatility coal.

2026 Guidance

For the fiscal year 2026, Warrior Met Coal (NYSE: HCC) expects to make significant progress in its operational and financial performance as the company transitions into a new phase driven by the full-year contribution of its Blue Creek mine. According to the company’s guidance, 2026 is projected to be a year of growth in both production and sales volumes, as Warrior builds on its recent operational achievements and the ramp-up of its strategic growth project. Specifically, Warrior Met Coal forecasts coal sales of 12.5 to 13.5 million short tons and coal production of 12.0 to 13.0 million short tons for 2026, representing meaningful increases compared to expected 2025 levels. This reflects the full-year impact of the Blue Creek longwall operations, which commenced ahead of schedule and are now contributing to expanded capacity. This was higher than what I modelled, I initially modelled the company to ramp production to 10.5 Mill metric tonnes whereas management has guided an 11.8 Mill metric tonnes.

Now comes the part that is slightly bearish: Despite the anticipated growth in volume, the company acknowledges that the external pricing environment for steelmaking coal may remain challenging. Global market conditions, including continued pressure on pricing indices driven by weak steel demand, excess supply, and shifting trade dynamics, are expected to influence realized prices in 2026. Warrior continues to emphasize that many aspects of its outlook, including pricing assumptions, remain dependent on broader market conditions in the steelmaking coal sector. This I believe is simply management being conservative in their guidance given the uncertainty in the outlook of met coal prices. Their assumptions for their guidance include the spot prices of met coal to be from 185-215/Mst.

In terms of costs and capital allocation, Warrior Met Coal is guiding to cash costs of sales between $95 and $110 per short ton, highlighting disciplined cost management even as production scales up. Capital expenditures are expected to rise modestly for sustaining existing operations, with circa $105–$115 million allocated to support mine infrastructure and operational needs, while spending on the Blue Creek project itself declines as it nears completion adding an additional 50-75 million. This brings the overall capex of the company to 150-190 million which is significantly below what I have included in my model of around 290 million. This difference frees up 100 million in free cash flow for the company. The lowering of the FOB cost per Mt is also within my expectations and what I have modelled for the company at around 112/Mt.

Management has also provided insight into free cash flow expectations for 2026. While the company anticipates negative free cash flow in the first half of the year due to the ramp-up of production and working capital needs associated with Blue Creek, it expects to return to positive free cash flow in the second half as capital spending declines and operations normalize. This is relatively in line with my expectations and as I am expecting the company to return to shareholders via dividends and buybacks. I am delighted to hear that management is choosing to buyback shares instead of just giving dividends as they have done in the past. Share buybacks for warrior are more value accretive to the shareholders, especially for international investors. I also loved the fact that management is choosing to focus only on returning to shareholders at appropriate prices. What I have detested about the rival AMR is that they have been buying back regardless of the price and the cycles that they are in. WIth Warrior, I am more optimistic with the management’s ability to buyback shares at opportune rates. Currently, the contracting activity is strong, with approximately 90% of the target sales volume under contract, providing revenue visibility and reducing market uncertainty.

Overall, Warrior Met Coal’s guidance for 2026 underscores a transition toward higher volume growth and improved operational scale. While market pricing remains a headwind, the company’s expanded capacity, cost discipline, and strong contract coverage position it for improved performance relative to 2025. All in all, the company has exceeded my expectations and management has successfully shown that they are highly reliable given that most mining companies exceed in budget and delay the development of the mines. This year is framed as a pivotal phase in the company’s growth trajectory, driven by the successful commercialization of the Blue Creek asset and an operational focus on maximizing production efficiency and stakeholder value. The company still remains undervalued and still presents a very good opportunity. Blue Creek has yet to reach full production volume and thus Warrior still has another year of volume ramp up in 2027. I am absolutely astounded by the quality of the management in a small cap company. The execution of the company has been way better than I have expected and I believe that the conservative nature of the management is absolutely crucial to the success of the business given the cyclical nature of the metallurgical coal. The moats of the company will never be disrupted given it is a natural moat. More details of the moat can be read in my initial writeup can be found at

. For those who have read my initial write-up of the company, you should be aware that unless God decided to create a river next to your mine, shift your mine 30% closer to the nearest port, magically put you in a state with cool weather and damp conditions that reduce your risk of a mine fire, or place you in a red state that is pro-coal, you will never be able to achieve the FOB costs that Warrior has. Barge loadout abilities are the key to warrior’s ability to maintain such a low cost. As the lowest cost producer, you are guaranteed to be profitable even in the worst of times and that is extremely valuable as you could strategically acquire weaker competitors at absurdly low valuations or buyback shares in your company. You will also be able to survive and see times where met coal prices goes absolutely crazy. In situations like this, it is an understatement to say that Warrior could 10X. Going back to the massive spike from 2021-2022, met coal spot prices spiked all the way to 600/Mt. During that period due to the war with ukraine Europe and parts of Asia faced extreme shortages of natural gas and thermal coal. In that environment, lower-quality met coal and PCI coal (pulverized coal injection) were occasionally diverted into power generation where plants had the technical flexibility to do so. With the current Ai developments and rapid buildout of data centers, it is the market consensus that energy demand is going to rocket. Whilst I highly doubt that just energy demands would drive the met coal prices to such high prices, there are numerous events that could occur that will cause met coal prices to rocket (ie floodings in australia etc). I would love to be able to predict when it would occur but situation like these are whilst important, is unknowable and hence I do not include it in my analysis. At current prices, you don’t need this situation to take place to succeed and it remains a free “call option”. Given that met coal has no debt and a net cash position that represents 10% of the market and also has the lowest FOB costs in the industry, I am delighted to hold onto the company given that the likelihood of the company going bankrupt is highly unlikely compared to the other swing producers with higher FOB costs such as AMR and whitehaven coal mine. Given that the mines are in alabama, the risk of mine fires are minimal due to the climate and being right next to the black tombee river. Flooding is also unlikely to occur given that in the past 25 years there were only 3 years where there was flooding but flooding in general is less detrimental to miners given that they usually only halt operations for a couple months and do not severely damage the mines as compared to the mine fires. Warrior provides for me an uncorrelated return stream to the broader market and is currently still quite undervalued even though it has run over 100% since I first started acquiring shares in the business. I initially trimmed some shares at around 84 2 months back as I was uncertain if the company was able to complete BC on time and now, with the BC being fully complete, the uncertainty has been completely erased. I have added back my position that I had trimmed into the company and I believe that the shares are still attractive at these prices.

Base Case DCF