IREN: An update since march

• Back in March I wrote an analysis I’m very proud of, and one I think was genuinely thorough. It’s now been more than four months since I published it, and a lot has happened since , while the company has moved in a direction many didnt expect IREN to take. Since I wrote that analysis the stock has been on a proper rollercoaster, swinging between $35 and $70 several times. The share price is only a few percentage points higher than when I published in March, but the thesis and the risk/reward are much stronger today than they were just four months ago.

  In this updated analysis I cover the most important things that have happened to

  $IREN

  since then, and what I think investors can expect going forward.

  I go through the update chronologically, covering the news and the direction IREN has taken over the past four months and why it matters for the company. At the end, I will lay out what can be expected over the coming year, and the new direction the company has taken, which has increased the upside enormously.

  I always like to start with the “bad” news and then move on to the positive. Even though a lot of positive things have changed for IREN since last time, Im not going to hide the fact that there have been things I’ve had serious doubts about as an investor, and that I cant defend.

  The bad

  The $832M RSU package

  I’m not going to spend too much time and energy on this, because there iss little I can do about it now. I don’t support this enormous payout to the co-CEOs, and I see it as very greedy and damaging to both the stock and to profits. But I’ll go through what’s relevant for investors to know, and how it will affect the company going forward.

  On June 30, the board approved the largest compensation package in the company’s history: 9,099,328 RSUs to each of the two co-CEOs — 18.2 million shares in total, worth roughly $832M at the closing price of $45.73. The package vests in four equal annual tranches, with a two-year holding period before they can potentially sell. There will be no further RSUs for the co-CEOs until fiscal year 2031. The size corresponds to roughly 5% dilution at today’s price, so about 1% annual dilution over the coming years.

  What I find strange is that the package has no performance conditions attached. There are no revenue targets, no share price thresholds, no EBITDA requirements, the brothers simply need to still be employed on each vesting date to receive all of it, regardless of whether the stock doubles or halves in the meantime. But they will ofc have incentives to make the stock be as high as possible, to get as much $ as possible.

  The RSU package will also hit the P&L, and most of the “damage” happens in the first year, where 52% of the RSU value is front-loaded and drags the accounts down through larger losses. So the result is that just as IREN scales revenue dramatically over the coming quarters, there will be large losses attached to this RSU package. I recommend reading Agrippa’s Substack for an even more detailed explanation of the effect of this package.

  The Warriors sponsorship

  

  One piece of news that made investors furious was IREN’s sponsorship deal with the Golden State Warriors. The announcement came at a time when the stock was in a downtrend and investors were fed up with the lack of communication from management and IR, and the lack of contracts and execution. In a period where the company is doing $33.4M in revenue (I’ll get to the Q3 report shortly), $50M for a sponsorship looks like a lot, and unnecessary.

  I’ll try to stay as neutral as possible on this decision and cover both sides. Toward the end of this update I’ll cover the new direction and strategy IREN has taken, which does justify the spending tied to marketing. This deal is also something I view as fairly unimportant, so I’ll keep it short and simply summarize the negatives and positives of sponsoring the Warriors.

  Negative:

  • Poor capital allocation? The deal costs roughly $50M, which today is more than IREN did in AI revenue last quarter. That’s a lot of money right now, but it will mean much less in the future. The company is sitting on $7.6B in cash (as of 20.07.26). This was probably an opportunity IREN couldn’t let slip, and one they saw good ROI in for long-term value creation.

  • Unnecessary? Is it necessary for a company like IREN to market itself in order to sell to customers, if their product is good enough and demand for compute is as strong as it is today? Maybe there shouldn’t be any need to sell your name for . A big contract with Anthropic, OpenAI or another hyperscaler would probably have sold the name pretty well too ;)

  • Measurability. It’s natural to wonder how much ROI a logo on a jersey can really deliver for a company like IREN, but I assume we’ll get some answers from the company at the next quarterly report.

  Positive:

  • IREN’s new customers are in Silicon Valley: IREN is moving in a direction where they focus more on end customers like Fireworks AI, Together AI, Hume AI, Perplexity and similar companies. Many of these companies are located in Silicon Valley, where the Warriors are based. The Bay Area is the place for AI startups, and IREN wants to reach that target group with their “marketing.” There will be many more of these companies, and they can offer better margins for IREN, which is why this cost can be defended.

  • It can influence regulation and marginal decisions: IREN operates in a sector that is under heavy scrutiny right now. There are a lot of concerns tied to AI and the effect it will have on cost and for humanity. By being a player that supports the local community and contributes positively to society, this could be decisive in whether individual decisions go IREN’s way or not. To make the point concrete: imagine Texas is only going to approve one data center, and it’s between IREN and Nebius. The people making that decision know the IREN name well because they sponsor teams like the Warriors, whereas Nebius is a name none of the regulators recognize at all (hypothetically). That small bit of recognition can matter enormously for whether things go IREN’s way on the regulatory side. And by supporting the sports people care about, it can help turn the narrative around AI, and the direct view of the company itself. We saw this earlier this year when IREN sponsored the Sydney Swans in Australia, and a few months later got their 800MW data center approved in the country. IREN had struggled for a long time to get capacity approved there, but after the Sydney Swans sponsorship their 800MW was approved. Whether there’s a correlation or not I don’t know, but you shouldn’t dismiss the fact that these things can have a positive effect — and then maybe a $50M cost isn’t so bad?

  Later Horizon 1 delivery than investors expected

  Many had hoped Horizon 1, the first 75MW (of 300MW) data center IREN is delivering to Microsoft would be finished in May or early June. Those were investor expectations only; there is no deadline known to the market that says IREN was supposed to deliver by then. IREN specified in the May 7 report that Horizon 1 would be delivered in Q3.

  As I write this section, the date is 22.07.2026, and we are only days away from the handover of Horizon 1. IREN’s internal date for the Horizon 1 handover was July 19, and MSFT has five days to approve the handover from

  $IREN

  . So I expect this to happen very soon 😊

  @Deep_miners

  found a Facebook post where an IREN employee posted about a dinner celebrating the completion of Horizon 1 (see image).

  “I was so grateful to celebrate our amazing employees at the SWECS Emplyee Appreciation Dinner for the completion of Horizon 1 in Childress, TX”

  

  Found by @Deep_miners

  The last quarterly report

  The last quarterly report, released on May 7, brought some very good news, but also came with disappointments.

  Like many investors, myself included, I expected far greater AI revenue than what ended up being reality. AI revenue in Q3 came in at only $33.4M, while most estimates sat between $60M and . The big miss came from late billing at the Prince George data center. ;eaning revenue took longer to come through than I and others had estimated. Those sites should now be operational and will contribute positively to the coming quarterly report at the end of August. There’s still some uncertainty about how long they’ve been operational during this quarter, so there could be large swings in AI revenue. The miss was of course disappointing, but it means little in the big picture, since the site is now 100% operational.

  What was also odd was how the quarterly report was released. It took well over an hour longer than usual before the company published its numbers. It was going to be combined with the NVIDIA deal (which I’ll cover later), and it seemed like that deal was meant to “cover” the weak AI numbers. It also seemed like something was changed at the last second, because the report and the earnings call were a bit chaotic and unprofessional. As someone who has watched the earnings calls repeatedly over time, you could tell Dan wasn’t as calm and confident as before, he seemed more stressed and irritated. Probably tied to the problems and the late release of the numbers.

  Even though there have been some disappointments since last time, the upsides have been much bigger, which has strengthened my thesis on the company. And even though the quarterly report had “bad” numbers, the company delivered some very good news, which I’ll go through now.

  The positives:

  The NVIDIA deal

  Throughout my earlier analysis I pointed to IREN’s strengths: secured power, full vertical control, and a management team that prioritizes long-term value creation. NVIDIA confirms to me that IREN is sitting on something NVIDIA can’t get anywhere else. There’s a reason Sweetwater 1 will be NVIDIA’s flagship deployment of DSX infrastructure. When the world’s most valuable company picks you as a strategic partner over every other neocloud, it’s because you have something they can’t get elsewhere.

  None of the other neoclouds today are fully integrated the way IREN is, with up to 5.8GW of capacity. The competitors lease data centers from third parties, while IREN builds its own.

  The NVIDIA contract itself is based on a partnership across the 5GW portfolio and starts with a smaller 60MW gross cloud contract at Childress worth $3.4B over five years. In this deal IREN will manage a fully managed cloud service, this is not a bare metal contract. IREN will provide the software layer to the world’s most valuable company! It’s hard to imagine NVIDIA needing that, but there may be other reasons behind it. Maybe NVIDIA wants to test IREN before scaling Sweetwater 1 (1,400MW) into one of the world’s largest VR200 data centers, or their other sites, where IREN combined with their new acquisition Mirantis will offer software on top of the bare metal model. That would mean even higher margins for IREN. Why else would NVIDIA use IREN’s software, when they almost certainly have better software than IREN does today? I strongly recommend reading @Agrippa_Inv deep dive on IREN’s latest quarterly report, where he makes very good arguments for the question I’m raising here.

  IREN and NVIDIA also ended up with a warrant agreement, which gives NVIDIA the right to buy 30M shares at $70, conditional on delivering up to 600,000 GPUs (NVIDIA can buy shares after every 100,000 GPUs sold). There is also a commitment to roll out NVIDIA’s DSX architecture and DGX systems, with Sweetwater as the stated flagship, and across the 5GW pipeline.

  What I feel is underappreciated in the market is how dependent NVIDIA is on IREN, and on the other neoclouds in the future. All of NVIDIA’s largest customers today are hyperscalers, and most of them are in the process of building their own GPUs to compete with NVIDIA. NVIDIA understands that over time competition will intensify, and that these players will use their own GPUs instead of buying NVIDIA’s, which carry net margins above 60% today. NVIDIA depends on players like IREN growing large and becoming successful companies, because those are their long-term customers.

  NVIDIA has an enormous incentive to make IREN as successful as possible, so that there are future companies to buy their products. NVIDIA also depends on data centers being built as fast as possible so their GPUs get installed. If development stalls or takes too long, operators won’t have incentives to buy more GPUs, and that would punish NVIDIA hard. So I like to think that NVIDIA has enormous incentives to make IREN succeed, and I think this is undercommunicated and poorly understood by the market. NVIDIA will support companies like IREN as much as possible going forward.

  An important distinction between the NVIDIA and Microsoft contracts is that IREN has gone from being a company that only offers bare metal to now offering a cloud service to all types of customers. For me the headline number isn’t the most interesting part, it’s what the deal does to IREN’s business model. The Microsoft deal was bare metal, IREN delivered raw compute. The NVIDIA deal is a fully managed cloud service. If IREN manages to scale and deliver a competitive cloud service plus Mirantis services, IREN can extract much better margins from customers, and this is already something we’re seeing in the newest contracts announced on 20.07.2026 (covered shortly).

  One detail I think is under-communicated is how little CapEx IREN will have tied to the NVIDIA contract. Since NVIDIA’s contract involves air-cooled GPUs, the cost is around 1/3 of liquid-cooled data centers. Instead of paying $15M per MW as in the MSFT contract, it will probably cost around $5M per MW to retrofit their Childress sites. This means less debt, less dilution and higher margins for IREN. And even though CapEx is lower for the NVIDIA contract, IREN still gets a 30% higher lease rate from NVIDIA than from MSFT. As Agrippa has also shown in his X post, the NVIDIA deal will generate more profit than the MSFT deal, even though it’s 1/6 the size (60MW vs 300MW). Net margins of 26% on the NVIDIA contract versus 8% from MSFT (numbers from Agrippa).

  

  From @Agrippa_inv X post

  There’s also GPU CapEx of roughly $1,6B. That means a far smaller capital commitment than Microsoft, and given the job Anthony Lewis has done on the financing side (remember the sub 6% rates versus CoreWeave’s 10%), IREN doesn’t need to lever this hard or dilute shareholders to fill the equity portion. This is a deal that largely finances itself, especially when IREN is sitting on $7.6B in cash (20.07.2026).

  The negative I take from the NVIDIA deal and from the newest contract we got on Monday 20.07.2026 is the lack of data. Since becoming an IREN investor I’ve valued their transparency, because it gives an investor more information to make the best possible decisions. This is something I’ve always appreciated about IREN relative to competitors like Nebius, who share very little about their contracts. Compare the numbers and information IREN and Nebius shared after their respective MSFT contracts and you’ll see what I’m referring to. But with the newest contracts from IREN the information has become thinner and thinner, we’ve received no GPU count, no CapEx, no prepayment and no margins.

  All in all, though, I’m very happy with the partnership IREN has built with NVIDIA, and the contract they’ve secured with very good margins and low CapEx = less dilution and less debt.

  The Mirantis acquisition

  I’ve already touched on the fact that the most important shift with the NVIDIA deal is that IREN moves from pure bare metal to offering fully managed cloud. The problem is that you can’t deliver a cloud service without a software stack, and that’s exactly the capability IREN has lacked until now. The Mirantis acquisition is the answer to that.

  What IREN is doing is simply buying its way up the value chain for just 625M. What’s also very interesting is that the acquisition is being done 100% in shares. The owners and investors in Mirantis are taking 100% of their payment in IREN stock. That means they’re selling the company they’ve built and loved for over 20 years in exchange for shares in a fairly young AI infrastructure company. One thing is certain: the Mirantis owners have full confidence in IREN.

  And here lies the really big point. Think about each layer a compute customer needs: land and power, the data center itself, the GPUs, and finally the software layer that makes it all usable. Every single layer someone else owns is a layer of margin leaking out of the company. CoreWeave and Nebius lease their data centers from third parties. The hyperscalers don’t run software platforms for others. None of the competitors own the whole stack. After Mirantis, IREN does exactly that: from the land the data center sits on to the software the customer logs into , plugging every margin leak at once. This is what, over time, makes IREN the most profitable company in the sector, with margins no one else can match. The information I’ve heard is that the acquisition closes at the end of July, or early August at the latest.

  

  What is Mirantis?

  Mirantis is a software company of around 650 employees based in Campbell, California, with more than two decades in the open source world (a major OpenStack contributor, bought Docker Enterprise in 2019, heavy on Kubernetes). That means IREN is simultaneously buying 650 employees with deep expertise in exactly the software and AI-services side their customers are going to need.

  But what’s really interesting is their main product, k0rdent AI, a turnkey “Metal-to-Model” platform that handles the entire path from racked hardware to finished, usable AI services. This isn’t generic enterprise software, it’s a purpose-built platform for operationalizing AI at scale, with multi-tenant handling that lets you serve many different customers on the same infrastructure simultaneously. Probably something of what we’ll see at the Mackenzie facility (?). Mirantis also holds several NVIDIA certifications and is one of the founding partners in NVIDIA’s AI Cloud Ready program , in other words, a company with an already tight collaboration with NVIDIA, which benefits IREN enormously.

  

  The software layer is where the margins are

  Mirantis describes the point better than I can myself: building a GPU cloud is the easy part; the hard part is making it commercially profitable. A neocloud that has to build bare metal automation, orchestration, GPU virtualization and a service catalog from scratch spends months on engineering before it sees a single dollar of cloud revenue. k0rdent AI lets IREN skip that entire phase.

  And here’s a point I think is absolutely central, remember why the Q3 numbers disappointed? AI revenue came in at only $33.4M because billing and burnin at Prince George took longer than expected. That is exactly the type of delay, the time from racked hardware to paying customer that k0rdent AI is built to cut down. In other words: the acquisition directly solves the problem that frustrated me last quarter. Future data centers should be able to go from finished rack to revenue far faster than Prince George did.

  Even more important is what the platform does to margins over time. Mirantis describes a three-step maturity model, which is really a ready made map for how IREN can climb in profitability: first Bare Metal as a Service with API-driven lifecycle management (away from pure commodity leasing), then a multi tenant GPU cloud where virtualization and higher customer density push up GPU utilization and profit, and finally a selg service AI cloud with its own marketplace where customers provision resources themselves. That fits perfectly with the direction IREN has taken toward Silicon Valley customers like Fireworks, Together, Hume and Perplexity, who don’t want raw iron, they want a finished platform they can just plug into. And we’re already seeing it: the newest $2.8B contracts explicitly include “both bare metal and managed cloud services.”

  The valuation angle nobody talks about

  This is really what I find most exciting as an investor. Step 3 in that model, self-service with a marketplace, means IREN moves from being a pure leasing business toward becoming a software and platform business. And that’s a far more important change than just higher margins. Software and platform revenue has historically been priced at a completely different multiple than pure infrastructure leasing, because it’s more scalable, more sticky, and has better earnings quality. So if IREN actually climbs those three steps, Mirantis doesn’t just change how much IREN earns, it changes how the market should value the company in the first place. That’s a re-rating opportunity, not just a margin improvement, and I think it’s severely undercommunicated.

  Future-proofing on the software side

  What’s also overlooked is that k0rdent AI provides day zero support for both today’s Grace Blackwell and next generation Vera Rubin, with a single common API surface across GPU generations. Operators can go from GB300 to VR200 without redesigning infrastructure, changing their operating model or rebuilding their workflows. See where this points? Sweetwater 1 is set to be NVIDIA’s VR200 flagship. This may not be decisive for the SW1 build itself next year, but it could save IREN enormously when they later upgrade, for example, the Microsoft contract to future models like VR Ultra. That’s exactly the future proofing advantage the DSX architecture is about, just on the software side. IREN now owns it in-house for $625m!

  

  The NVIDIA connection runs deeper than people think

  One last thing worth highlighting, straight from Mirantis’ own material: they aren’t just a user of NVIDIA’s stack. They are a primary open source contributor to NVIDIA’s Infra Controller, work upstream with NVIDIA to define how the provisioning of AI data centers should evolve, and have integrated NVIDIA DSX OS. In other words: IREN didn’t buy a random software company. They bought a company that helps shape NVIDIA’s own AI factory framework. It’s hard to imagine a better piece to own when you’re simultaneously going to be NVIDIA’s DSX flagship.

  Potential downsides of the acquisition

  I’m not only going to comment on the upside, because this is not a guaranteed success. First, the acquisition hasn’t formally closed yet, it’s expected to complete at the end of July based on the information I’ve heard. So until it closes, this is unusable.

  Second, the other downside lies in IREN’s own operational execution. Integrating a company of 650 employees with its own culture and its own systems is not easy, and historically many acquisitions create problems and take longer than investors expect. IREN has also never operated a managed cloud service at scale, so it remains to be proven that they can actually implement the technology, and not just own the company that has it. Mirantis will operate for itself, still after the aqquisation. So it will make it easier to cooperate between eachother.

  All in all

  For $625M, IREN plugs the last margin leak in the value chain, connects technically straight into NVIDIA’s next-generation roadmap, and potentially opens the door for the market to value the company as a software platform rather than a pure infrastructure lessor. If they manage the integration, this is the piece that lifts IREN from “GPU lessor” to the most profitable, fully integrated cloud provider in the market. The upside since I wrote the March analysis has become much bigger, and in just four months the company has gained the ability to offer services to more companies, which opens the door to more customers, not just hyperscalers. It also creates the opportunity to capture margin across the entire value chain.

  APAC: 800MW

  As many know, IREN is today an Australian company (hopefully American in the future 😉), but until now the company has only operated data centers in North America (Canada and the US). That changed on June 3: IREN signed a transmission connection agreement supporting a planned 800MW data center at Bundey in South Australia, around 125 km northeast of Adelaide. This is the company’s first announced data center in Australia, and one of the largest data center projects announced in the Asia-Pacific region so far.

  Worth clarifying: what’s secured is the grid capacity. The agreement secures four 330kV feeder connections at the network operator’s substation, supporting up to 800MW without the need for network upgrades. IREN expects to begin early works and procurement in parallel with obtaining regulatory approvals and fulfilling the conditions of the agreement, so the final authority approvals remain outstanding.

  Energization is expected to start from 2028, which means the facility probably won’t begin generating revenue until late 2028 or early 2029 (my own assumption on timing). With this, IREN’s total power pipeline now passes 5.8GW.

  The strategically important part here is access to APAC. The site benefits from submarine fiber connectivity into the major regional demand centers, including Singapore, Indonesia, South Korea and Japan. Asia-Pacific is among the world’s fastest-growing sources of AI demand, with a significant gap between expected demand and available infrastructure. It also fits well with IREN’s clean-power profile: South Australia’s grid targets 100% net renewable energy by 2027.

  Taken together, this reduces geographic risk and opens up a far broader customer base across regions, which in turn reduces concentration risk in the company. The project is also expected to create over 200 ongoing skilled jobs and more than 500 positions during construction, which strengthens the political will behind the project.

  Nostrum Group in Spain (490MW)

  In June, IREN completed the acquisition of Ingenostrum S.L., better known as Nostrum Group, a Spanish developer of grid connected AI data centers. This is an important milestone, because it marks IREN’s very first entry into the European market, expanding their footprint from North America and Australia to now also include Europe.

  What IREN actually gets: around 490MW of secured, grid-connected power in Spain, plus a development pipeline beyond that (details were not disclosed), and a local team of more than 50 employees across development, engineering, construction and operations. Nostrum was founded in 2009 as an engineering company in the renewables sector, and pivoted toward data centers as its core business when it rebranded from Ingenostrum in 2025. IREN did not disclose the purchase price, in line with the declining transparency I’ve mentioned earlier.

  With Nostrum, IREN’s power portfolio passed 5GW (before the Australia deal later lifted it to 5.8GW). That means IREN has increased its grid-secured portfolio by 100% in just one year, from 2.9GW to 5.8GW!

  Why Spain?

  The country has become one of the most attractive locations in Europe for large scale AI data centers, with abundant and cheap renewable power, strong fiber connectivity, and a regulatory climate that is actually supportive of data center development, three things that are in short supply in many other European countries.

  What I like best about this acquisition is how it fits IREN’s playbook. They didn’t enter Spain on a greenfield basis from zero. They bought a team that already holds secured power positions and knows the local permitting landscape inside out. For someone like me, who has argued that IREN’s entire edge lies in delivery certainty and vertical control, this is exactly the same recipe, just exported to a new continent. You’re buying execution capability, not just power on paper.

  Strategically this does two things. First, it further reduces concentration risk, just like the Australia deal, the broader the geographic and customer footprint, the less vulnerable IREN is to any single market. Second, it opens the door to something I think is underrated: sovereign AI.

  European governments are increasingly focused on having AI compute on their own soil, for reasons of data sovereignty and security, and Nostrum’s CEO Gabriel Nebreda himself highlighted sovereign AI programs as a target group. That’s a structural tailwind IREN is now positioned to capture.

  All in all a small acquisition in both dollars and MW compared to the American footprint, but strategically a big step. IREN is no longer a pure North America company, they’re becoming a global AI infrastructure player. The timeline for actual revenue from Spain is probably a couple of years out, since the pipeline has to be built out, but the foundation is now in place.

  Raise Summit 2026 — IREN as main sponsor

  

  Much of what I’ve mentioned so far is well known to core investors, and it’s fairly public information. What I can offer that’s more recent comes from Raise Summit. Raise Summit is a large AI conference held in Paris and considered Europe’s biggest meeting place for people in the sector. What’s interesting about this year’s edition is that IREN was the main sponsor of Raise Summit 2026.

  Several IREN investors were invited to the conference by

  @HenriDelahaye who gave @LandoInvests

  several tickets after he wrote an analysis on this year’s Summit. It ended up being a big success for the investors, and we gathered a lot of information over those days. In the next sections I’ll go through what I think is most relevant for IREN investors, and what it will mean for the company.

  @LandoInvests

  made a summary of the most important takeaways from those days and hosted a Space with the other investors who attended, which contained a lot of important information.

  What’s really interesting are management’s statements that:

  • The contracts and deals available today are better than what was on the table in negotiations just two months ago. I read this as meaning the coming contracts could be even more profitable than the NVIDIA contract announced in May (two months ago). Demand continues to grow faster than supply, which makes IREN’s 5.8GW more valuable with every day that passes, ending in better lease rates, better terms and higher prepayments from customers.

  • The sweet spot for prepayments is between 25–40%, and in some cases up to 100%! Prepayments mean IREN has less need to dilute shareholders or take on more debt or convertible notes. So seeing prepayments trend up from the 20% in the MSFT contract is very positive. Everything suggests the company is in a much better position than it was just four months ago. As we saw in the newest deal from 20.07.2026, the GPU prepayment was 45%! It’s important to distinguish between prepayment as a share of contract value and as a share of GPU value (I cover this in more depth toward the end).

  • IREN will receive VR200 test units. IREN is one of few players that will be allocated test batches of NVIDIA’s newest model, VR200. That gives them roughly six months to test the chips as much as possible before they’re ready for mass installation across sites.

  • Horizon 1 handover July 17–19. Investors got confirmation from management that the internal date for the Horizon 1 handover to MSFT was between July 17 and 19. This is very big for the company, and will remove an enormous execution risk from the stock. IREN will have proven they can build the data centers of the future quickly. There aren’t many players in the market who have pulled off the same thing, and the first one is always the hardest. After the H1 delivery it becomes copy paste plus upgrades based on the learnings from H1, deliveries get standardized and will happen faster and hopefully more profitably.

  • The sweet spot for deals is now 3 years. As with the newest $2.8B contract, IREN has started taking on more enterprise customers instead of hyperscalers. The preferred contract length has therefore come down from 5 years to 3. This has both an upside and a downside: higher rates for a shorter lease period, but greater GPU residual value risk, whereas 5-year contracts offer lower lease rates but less risk tied to paying down the GPUs and other CapEx.

  • GB300 Exemplar Cloud status from NVIDIA. It was said that IREN has received GB300 Exemplar Cloud status from NVIDIA, which is a mark of quality. It’s a very strict designation, and IREN will be one of few players to receive it. This is expected to be announced shortly (days/weeks). It will help strengthen IREN’s brand and gives more confidence to customers and future customers.

  New $2.8B in contracts for IREN

  After a long wait, IREN has finally come out with news that, for some investors including myself, felt like an eternity. Even though we got a contract with NVIDIA, that didn’t feel the same as this.

  On July 20, IREN dropped big news: $2.8B in new, multi-year cloud contracts with a number of leading AI developers. At the same time they lifted the ARR target for end of 2026 from $3.7B to over $4B, of which roughly 85% is now under contract. This was well rewarded by the market, and IREN gained 20% the same day.

  Investors who have followed the company for a long time knew this had to happen. The increase from $3.7B to $4B was in the cards for anyone who has put in enough time to understand the company. I personally believe the EOY 2026 ARR will be raised further over the course of the year and land between $4.2B and $4,3B.

  The new contracts IREN has landed include only smaller AI enterprise companies. This also justifies why IREN has pushed hard to build brand through the Golden State Warriors sponsorship. They want more of these customers in the future, because they’re willing to pay more for compute than hyperscalers like MSFT and Google are. This is a customer group it makes more sense to grow alongside. By going only for hyperscalers, you risk that in the future those players no longer need IREN because of their own capacity buildout, leaving IREN with no customer and none of the good customer relationships they’re now building through these AI enterprise companies. It means IREN can extract better margins from these companies, while they’re also a better fit as long-term, sustainable customers.

  IREN also disclosed that they’re sitting on $7.6B in cash and cash equivalents. That corresponds to $21 per share in cash alone (not accounting for debt).

  IREN also received a 45% prepayment on the GPU cost tied to the contracts. That’s a big upgrade from earlier contracts (20% for MSFT). It shows that IREN can build scale without hyperscalers, because their customers have plenty of capital to cover IREN’s CapEx. This development is very positive for shareholders who fear dilution, and for those who fear the debt growing too large. A 45% GPU prepayment means IREN avoids financing large parts of the GPUs itself.

  What’s a bit odd is how IREN has changed the wording here. For the MSFT contract, prepayment was expressed as a share of the total MSFT contract size ($9.7B = $1.94B in prepayment), whereas in the newest contract it’s 45% of the GPU cost, not the total contract value. That means in practice the prepayment on the newest contracts sits around 35% based on my assumptions about total contract value.

  Again I have to point out the negative:

  We got aggregate numbers and a customer list, but no breakdown per contract. No GPU counts, no prices, no MW allocation, and no indication of who is bare metal versus managed cloud. Transparency continues to decline, and it makes it harder to model the individual deals precisely. But that’s a luxury problem when the direction is this clearly positive. I also understand this is less necessary when we’re most likely talking about smaller contracts with many more counterparties, but I will always root for transparency, and for as much information to investors as possible.

  What I was very happy to see, though, were the customers IREN has landed. They now have enterprise customers like: Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI, Hume AI, and a new leading AI developer. These are good, up and coming AI companies that IREN can grow alongside.

  What also interests me is how the newest contracts already include managed cloud services. That means IREN is already offering more services than they have before. The press release states that the contracts will include “both bare metal and managed cloud services.” Could it be that the Mirantis acquisition is already adding value to the contracts IREN signs?

  But all of this clarifies the strategy I think IREN will focus on going forward: less hyperscaler, more AI enterprise.

  The new strategy: from hyperscalers to AI enterprises

  This is perhaps the most important strategic change to understand going forward, and in my opinion what really shapes the IREN thesis from here. IREN is moving away from a primary focus on hyperscalers and toward a far broader base of AI enterprises.

  IREN is a company that has been through enormous change, and it keeps changing dramatically whenever the math says it makes sense. IREN has gone from being primarily a colocation player in its earliest AI days, to a bare metal player, to now being a player that also offers software through the Mirantis acquisition. What’s next?

  To understand why this matters so much, we have to look at the playbook itself.

  The old model, where the Microsoft deal is the classic example, works like this: to fill a large new data center, you lock in a giant customer 9–36 months before the facility is finished. The problem is that only the hyperscalers have the capital, the balance sheet and the visibility to commit that far in advance. And when you sign that early, you also lock in the price at that point in time, in a market where compute is getting scarcer and prices keep rising. If IREN had sold all its capacity in early 2026 as many of us wanted, we would have far worse terms and prices than what we see in the market today. Just look at the difference I showed when comparing the NVIDIA and MSFT contracts. The difference is night and day.

  IREN’s new playbook flips this. Instead of pre selling capacity years in advance, they keep more of the capacity open and sign contracts closer to the commissioning date. Doing this gives IREN better prices, better terms and better prepayments. Just look at the GPU lease rates SpaceX got by leasing out capacity that was available on short notice. The lease rates were around 2x today’s GPU prices with their Anthropic and Google contracts. You get rewarded for waiting as close to commissioning as possible, though there’s little data on this so far (only SpaceX for now).

  This isn’t just speculation on my part: it’s something management has confirmed directly. Look at their newest press release from July 20. They explicitly write that they remain selective in allocating capacity ahead of commissioning, and that they prioritize diversification across customers and platform layers. So this is something investors need to get used to: Contracts will come much closer to completion of the data centers than they do at competitors. Both Nebius and CoreWeave have sold most of their future capacity, while IREN waits.

  Why does this make sense today?

  The whole strategy rests on one assumption: that demand exceeds supply. And it does, by a wide margin. IREN says themselves that demand from hyperscalers, enterprises, AI developers and frontier labs continues to exceed both available and planned capacity. In such a supply-constrained market, IREN doesn’t need to derisk by selling off capacity early. They can hold back, let demand come to them, and take better prices when they do sign. It’s a seller’s market, and in that environment it’s simply dumb to lock yourself into yesterday’s prices.'

  The broader customer base

  Here’s where the real gain lies. By signing closer to completion, IREN opens the door to an entirely new group of customers. Many smaller AI companies : model developers, AI app providers and robotics companies, They cant simply commit to capacity 12–36 months in advance. They don’t know where they’ll be in three years; only the hyperscalers have that luxury. But when IREN signs close to commissioning, suddenly this whole broader market can participate. And that’s exactly what we see in the $2.8B contract: Perplexity, Figure, Together, Fireworks, Hume and the rest. These customers pay better prices, want managed cloud rather than raw iron, and they diversify the customer base so IREN isn’t at the mercy of one or two large hyperscaler customers.

  Short-term pain, long-term gain

  I’ll be honest about the downside, because it’s real. This strategy looks bad in the short term. Fewer announced contracts, a backlog that looks thin, and a market that grows impatient and punishes the stock, exactly as we saw during the correction and the rollercoaster between $40 and $70. Much of the frustration investors felt over the “lack of contracts” was in reality not weakness, it was discipline. Management deliberately left capacity open to secure better terms later.

  This is exactly the mindset I want to see from a management team: a willingness to take short term pain in exchange for long term value creation. Better margins, more customers, higher prices. Less risk, more upside. Its easy to sell everything off early to calm the market, but that’s not how you build the most profitable company in the sector.

  The road ahead: what investors can expect through 2026 and early 2027

  I’ll try to summarize what can be expected over the coming six months for IREN, including contracts, news and hires. As we saw after Google released its numbers, CapEx among hyperscalers keeps rising. Demand continues to increase dramatically, and Google continues to defend the use of enormous CapEx, which benefits IREN a great deal, and we can expect the other hyperscalers to follow.

  This updated analysis has already become much longer than I intended, so I’ll use more headers and keep the explanations simple, listed in roughly the order I think things will happen:

  • Handover of Horizon 1. As mentioned earlier, this should be announced within the coming days, based on the data and information I’ve received.

  • GB300 Exemplar Cloud designation from NVIDIA. This should also come shortly; I expect a press release on this before August.

  • FY2026 quarterly report. This will probably be IREN’s last bad quarter, and it will be used as a quarter where they front-load a lot of CapEx and write-downs. I won’t spend much time on this, I think

    @fransBakker9812

    has made a good summary:

  

  • Horizon 5–6 get a contract. IREN has started groundwork to build the Horizon 5–6 data centers, which will house NVIDIA’s newest model, Vera Rubin 200. What’s interesting about VR200 is that the newest tests done by CoreWeave and NVIDIA show that first measured performance delivers 10x more tokens per megawatt than Blackwell. That means the revenue IREN will generate over the coming year will increase DRAMATICALLY, if this holds true for the Horizon 5–6 facilities and SW1. I think we can expect revenue per MW at these VR200 facilities to be roughly $20M/MW (speculation). That means H5–6 and SW1 alone could generate $6B in ARR on my assumptions, though these aren’t available numbers today, so be critical of them. For the remaining data centers with GB300 and B300 models I assume roughly $13M/MW, which is what’s used in this calculation/illustration. I’ll let the numbers speak for themselves:

  • 

  So we shouldn’t rule out that IREN manages to deliver over $15B in EOY 2027 ARR, roughly a 4x in ARR growth YoY (from $4B). I’m very optimistic about IREN’s future 😉

  • The Mackenzie facility generates cash. Rumors have spread that IREN has already received many of the GPUs destined for Mackenzie, and that the retrofit of the data centers from BTC mining to GPU data centers is going smoothly. I expect these to start generating cash for IREN during September or early October. Mackenzie is where IREN’s newest contracts ($2.8B) are being deployed. It’s not yet known whether all capacity at Mackenzie is leased out, but much of it is.

  • Increase of EOY 2026 ARR: IREN recently raised ARR guidance from $3.7B to over 4B. Over the course of the year, as the remaining capacity gets contracted, ARR will rise further to $4.2B–$4.3B. That means IREN will have raised guidance through the year from $3.4B to $4.2B, a pretty hefty increase :)

  • Contract on SW1. This is pure speculation, but I think toward the end of autumn or the end of the year, IREN will announce its customer for the giant facility at SW1, which will be 300MW of VR200. This is NVIDIA’s flagship site. IREN has already come a long way with the SW1 build (I recommend subscribing to Frans Bakker to get access to the newest satellite images he buys of Childress and SW1). This will be an enormous contract, where the contract value will be larger than today’s market cap of around $15B!

  • Contracting out the remaining air-cooled data centers at Childress. After the May report, IREN announced they would retrofit 250MW from BTC mining to data centers. Of those 250MW, 60MW is already sold to NVIDIA through the $3.4B contract. So we can expect a further 190MW of capacity to be sold during the year. Assuming the same contract value per MW, a 190MW contract would come to roughly $10.8B in contract value over a five-year period. These air-cooled data centers are much faster to bring up than liquid-cooled ones, and can be expected to be ready in the first half of 2027. There’s enormous demand for this capacity, and I expect very good lease rates and terms. I expect the customers IREN lands for these 190MW won’t be a hyperscaler, but many more AI enterprise companies.

  • Canal Flats. Canal Flats is IREN’s smallest data center at 30MW in Canada, and IREN has said it’s a 2027 project. This site will run air-cooled GPUs, probably B300/GB300, and won’t get much focus from management given its size. But I expect a contract on it toward the end of the year or early in 2027.

  • Bitcoin becomes history. I think 2026 will be the last year where Bitcoin mining is a noticeable part of revenue. I think IREN will generate some BTC into 2027 through the Canal Flats facility (30MW), but it will be a very small part of revenue, so completely unimportant. IREN will announce that they’ve moved away from BTC entirely and will focus on AI after Canal Flats becomes an AI data center.

  • The news that will break short-term investors: a new ATM. I think that during this year or early 2027, IREN will announce that they’ve used up their existing $6B ATM and opened a new one at $10 billion. I have confidence that IREN can use this capital to create added value, at shareholders’ expense in the short term. This will create a lot of negativity, just as the previous ATM did. So consider this a warning to everyone who fears this: sell, if you can’t handle further dilution. But I also think this will be the last ATM of that size, after which they can scale further using cash flow from the data centers then in operation, plus debt tied to infrastructure and GPUs. The flywheel will hopefully be well underway after that ATM.

  Closing thoughts

  Four months, and the stock is essentially flat. But that’s the whole point of this update: the price hasn’t moved, and the business underneath it has changed completely.

  In March the thesis was secured power, vertical control, and one very large customer. Today IREN owns the entire stack from land to software, has NVIDIA as a strategic partner with structural incentives to make them succeed, has expanded from one continent to three, has grown the grid-secured portfolio from 2.9GW to 5.8GW, has broadened the customer base from hyperscalers to a group of AI enterprises that pay better and are willing to prepay 45% of the GPU cost, and is sitting on $7.6B in cash — $21 per share before you value a single megawatt.

  Meanwhile the market has spent those four months being annoyed about a sponsorship, the $6B ATM, and a quarter of delayed billing.

  I’m not going to pretend the risk is gone. The RSU package is greedy and unearned, transparency is getting worse rather than better, IREN has never run a managed cloud at scale, the Mirantis integration has to actually work, and the ATM I’m expecting will hurt when it comes. Every one of my numbers for 2027 rests on assumptions that could turn out to be wrong.

  But the direction is the thing. Every single decision the company has made over these four months has moved it further up the value chain and further away from being a commodity lessor. That’s the pattern I’m buying, and it’s why I’m holding a stock that’s flat with far more conviction than I held it in March.

  (Stock price when this was written = $41,28)

  None of this is written in a vacuum, and a few people deserve credit for the work behind it:

  • $Agrippa_inv

    — for the deep dives on the RSU package and the latest quarterly report, and for the NVIDIA vs MSFT margin work I lean on above. If you’re not subscribed, fix that. Check out his Substrack:

  Agrippa Investments

  Institutional-Grade Research & Analysis on Hyper-Growth Stocks.

  By 𝐀𝐠𝐫𝐢𝐩𝐩𝐚

  • @fransBakker9812

    — for the FY2026 quarter breakdown, and for buying the satellite imagery of Childress and SW1 that lets the rest of us actually see the build progress.

  • @LandoInvests

    — for the Raise Summit write-up and for hosting the Space afterwards, which is where a lot of the management commentary in this piece comes from.

  • @HenriDelahaye

    — for getting investors into Raise Summit 2026 in the first place.

  • @Deep_miners

    — for digging up the Horizon 1 completion post that nobody else caught.

  This is what makes theI RENcommunity worth being part of. Go follow them.

  T his is not investment advice.

• I own shares in

  , and everything above is my own analysis and my own assumptions — be critical of the numbers. I also used some AI to correct the article and translate from Norwegian til English, some segments also have been written by the help of AI.)

Lemonade was founded in 2015 by Daniel Schreiber (current CEO) and Shai Wininger (CEO/CTO). Both have been entrepreneurs for most of their lives and have built successful companies since the late 1990s. Shai co-founded Fiverr but left the company due to its culture, wanting to create something bigger and different. Interestingly, the idea of starting an insurance company was the result of a misunderstanding during a meeting Daniel had with a friend.

Shai is the company’s technological genius with a deep understanding of technology and artificial intelligence, perfecting their products. Daniel describes himself as a generalist: a leader who sets the vision and coordinates the entire orchestra. He is also a skilled storyteller.

Since 2015, Lemonade has built its fully digital AI system and continuously improved it; today they are likely on version 14+.

In today’s market, “AI” has become a buzzword that many companies use purely for marketing purposes, without any real substance. When Daniel was presenting the company’s strategy to the board and management, the entire meeting was about how they would use technology and AI to revolutionize the insurance industry. (This photo is from 2015, from their first strategy presentation.)

From 2015, their first strategy presentation

The company is fully integrated – they do not use any third-party services or purchase licenses from others. Shai and his team have built everything from the ground up, and the results have become evident in recent years. As of March 2026, the company has approximately 1,280 employees.

Why Insurance?

Daniel and Shai chose insurance because the market is enormous – several trillion dollars in annual TAM – in an industry that hasn’t been fundamentally renewed in over a hundred years, while trust in traditional insurers remains low. The average age of publicly traded insurance companies in the U.S. is over 130 years. These large companies have profited from complex systems requiring many employees. Lemonade, by contrast, has a single integrated system, and the use of automation and AI enables the company to scale rapidly.

How Lemonade Uses AI and Automation

Claims and Policies

Lemonade has two AI agents, Maya and Jim, that handle all new policies and claims. Maya processes 98% of new insurance applications, while Jim covers 55% of claims. A full 30% of all claims are processed and paid out in just 2–3 seconds. Claims volume has increased 2.5x over the past three years, while Lemonade has reduced headcount – efficiency has roughly tripled.

Telematics and Customer Pricing

Lemonade uses AI models that analyze all available customer data to predict risk. Traditional insurers typically select customers based on age and credit score, but Lemonade uses telematics to measure actual driving behavior. Safe drivers receive lower premiums, while higher-risk drivers pay more.

New in Q4 2025: Lemonade has launched autonomous pricing for Tesla Full Self Driving (Supervised). Autonomously driven miles are priced at 50% of the human-driven per-mile rate. As FSD safety continues to improve, prices will continue to decline. This positions Lemonade at the forefront of the transition to autonomy-aware car insurance.

Operational Scaling

Because Lemonade is built on a robust technology foundation, it can grow its customer base without increasing headcount at the same rate. Since Q3 2022, the company has added 1.2 million customers while headcount declined by 6%. Lemonade’s LAE costs have fallen from 13% to 6% (Q4 2025), with a target of further reduction to 3.5%. Leading U.S. insurers operate at around 9%, while traditional carriers sit at 12–15%.

Picture from @NFTlunatic

Lemonade’s Strategy

Lemonade focuses on young customers – 70% of its customers are under 35. The strategy is to find the lowest-risk customers and provide such a strong product experience that they remain customers for life, upgrading from renters/pet insurance to homeowners and then car and home insurance. Over 50% of all new car insurance policies come from existing customers – zero marketing cost for these sales.

2026 Strategy:

The company is investing in three AI-driven strategic initiatives:

(1) an enhanced cross-sell platform,

(2) a more automated pricing engine for faster and more accurate pricing changes, and

(3) the next evolution of the growth engine toward real-time, autonomous optimization across channels and creative. An Investor Day has been announced for New York City in November 2026.

Q4 2025 Results – Record-Breaking Quarter

The fourth quarter of 2025 was Lemonade’s strongest ever, characterized by accelerating growth, underwriting excellence, and increasing operating leverage.

Key Metrics for Q4:

9 consecutive quarters of accelerating IFP growth: The IFP YoY growth rate has increased every quarter from 18% (Q3 2023) to 31% (Q4 2025). Simultaneously, the TTM gross loss ratio has fallen from 88% to 64% over the same period. This combination has resulted in a more than quintupling of gross profit.

Full Year 2025 vs. 2024

Full Year Metrics FY 2025 | FY 2024

Revenue $737.9M | $526.5M

GEP $1,050.8M | $827.3M

Gross Profit $293.4M | $166.9M

Gross Profit Margin 40% | 32%

Gross Loss Ratio 64% | 73%

Net Loss ($165.5M) | ($202.2M)

Adj. EBITDA ($118.1M) | ($149.7M)

Adj. Free Cash Flow $48.8M | $47.6M

Growth Drivers – Q4 2025 Update

Pet Insurance

IFP: $439M | Growth: 55% | TTM GLR: 70% (+1 pp)

Pet is now Lemonade’s largest product line by IFP – achieved in just 5 years. The growth rate has been in the mid-50s for two consecutive years. Lemonade has the fastest-growing share of Google search volume among U.S. pet insurance companies. The cost per claim has fallen from $44 in 2021 to $14 in 2025 – a level the company believes is industry-leading.

Car Insurance

IFP: $187M | Growth: 53% | TTM GLR: 70% (–23 pp)

2025 was a year of deliberate product optimization for car insurance, with 2026 set to mark the broader breakout. Conversion rates improved by as much as 60% compared to a year earlier. Both new and renewal business show improved underwriting. As of summer 2025, the product was available in only 42% of the U.S. market, with a Pennsylvania launch upcoming (covering 9 of the 10 largest U.S. cities). U.S. car insurance represents an annual TAM of approximately $350 billion.

Autonomous Pricing: Lemonade has launched autonomous pricing for Tesla FSD (Supervised), where autonomously driven miles are priced at 50% of the human-driven rate. The company believes the expansion of autonomous driving plays directly to its strengths: usage-based pricing, real-time data, and flexible coverage.

Europe

IFP: $60M | Growth: 150% | TTM GLR: 74% (–16 pp)

Europe delivered its tenth consecutive quarter of triple-digit IFP growth. Strong momentum across both direct and price comparison channels. Europe’s regulatory environment allows for rapid iteration in pricing and underwriting. Lemonade operates in the U.S., UK, Germany, the Netherlands, and France. Europe still accounts for only ~5% of total IFP, leaving substantial room for further expansion.

Homeowners Multi-Peril

IFP: $530M | Q4 GLR: 39%

Remains the largest single line in IFP terms. Strong underwriting with a Q4 gross loss ratio of 39%, down from 55% in Q4 2024. Premium per customer has declined somewhat ($247 vs. $266 in Q4 2024), which may reflect product mix and non-renewal of underperforming policies.

Reinsurance

The company has significantly reduced its use of reinsurance. Ceded written premium fell from $513.9M in FY 2024 to $407.8M in FY 2025. In Q4 2025, only $55.2M was ceded, versus $124.5M in Q4 2024 – a 56% reduction. This means a larger share of premiums is retained by Lemonade.

The consequence: revenue is growing faster than IFP (53% vs. 31% in Q4), and a larger share of margins flows to the company. The risk is somewhat higher in catastrophic events, but management has indicated they can adjust the reinsurance level as needed. This decision reflects increased confidence in their own models and underwriting capabilities.

Increased Marketing Spend

Growth spend increased to $53.4M in Q4 2025, up from $36.0M in Q4 2024 (+48%). For the full year 2025, total sales and marketing reached $224.4M versus $166.3M in 2024 (+35%). Marketing is directed at young customers with lower CAC, who generate high long-term returns through cross-selling. The company targets an LTV:CAC ratio of 3:1.

Balance Sheet and Liquidity

Cash, Cash Equivalents & Restricted Cash $396.8M

Investments (Fixed Maturities + Short-term) $722.9M

Total Cash & Investments $1,119.7M

Borrowings Under Financing Agreement $158.1M

Regulatory Surplus (Restricted) ~$250M

Total Assets $1,925.7M

2026 Guidance

Key takeaways from guidance:

• IFP growth of 32% guided for 2026 – 200 bps above prior public comments

• Revenue growth of ~60% (driven by both IFP growth and lower reinsurance cession rate)

• Adj. EBITDA loss improves by >50% compared to 2025

• Positive Adj. EBITDA expected in Q4 2026, with 2027 expected to be the first full year of Adj. EBITDA profitability

• These expectations were first communicated in 2022 – underscoring the predictability of the model

Current Valuation

As of March 27, 2026, LMND trades at approximately $60.70, implying a market capitalization of roughly $4.6 billion. With $1.12 billion in cash/investments and $158M in debt, the enterprise value (EV) is approximately $3.6 billion.

Share Price (03/27/2026) $60.70

Market Capitalization ~$4.6B

Enterprise Value (EV) ~$3.6B

P/S (FY25 Revenue) ~6.2x

P/S (FY26E Revenue ~$1.19B) ~3.9x

EV/GEP (FY25) ~3.4x

EV/GEP (FY26E ~$1.36B) ~2.6x

52-Week High / Low $99.76 / $24.31

At the current price, Lemonade is not cheap, but the valuation sits well below my estimated long-term fair value ($150). Use

@Ironic_Ape

website to calculate LMND fair value:

https://www.ironic-ape.com/lmnd

The stock trades at ~3.9x FY26 expected revenue – still higher than traditional insurers (0.5–1.5x) but moderate compared to high-growth technology companies (10–20x).

Analyst consensus is Hold, with price targets ranging from $44 (KBW, Sell) to $85 (Morgan Stanley, Overweight). Morgan Stanley upgraded the stock on March 17, 2026 following the Tesla partnership announcement. Truist Securities initiated coverage with a Buy recommendation in January 2026.

Key catalysts ahead:

• Positive Adj. EBITDA in Q4 2026 – could trigger re-rating from institutional investors

• Tesla FSD insurance and autonomous pricing as a growth driver

• Car insurance as a rapidly scalable line (expanding from 42% to full U.S. coverage)

• Continued European expansion (only ~5% of IFP today)

• Investor Day in New York City, November 2026

Risks

• The company remains unprofitable on a net income basis (estimated EPS of -$1.76 in 2026)

• Higher retained risk following reduced reinsurance increases downside exposure to catastrophes (the $6.9M California FAIR Plan assessment related to the January 2025 wildfires illustrates this)

• Annual Dollar Retention has declined from 88% to 85% – partly driven by deliberate non-renewal of underperforming policies, but worth monitoring

• Competitive intensity in the insurance market, particularly from other insurtechs and traditional carriers investing in technology

• Dependence on AI models continuing to improve and deliver better pricing accuracy

• A delayed path to profitability would likely be punished severely by the market

Summary

Q4 2025 confirms that Lemonade’s business model works and scales. The company is firing on all cylinders: accelerating growth (31% IFP, 53% revenue), strong underwriting improvement (gross loss ratio down 11 pp to 52%), and clear operating leverage (1.2M new customers with 6% fewer employees). Gross profit has more than quintupled in under two years.

What sets Lemonade apart is its technology stack, purpose-built from day one. While legacy insurers bolt AI onto decades-old systems, Lemonade’s entire operation – from underwriting and claims to customer acquisition and fraud detection – runs on proprietary AI infrastructure. This is not cosmetic AI; it is the reason the company can scale revenue 53% while shrinking headcount. As AI capabilities compound, this structural advantage should only widen, making Lemonade increasingly difficult to replicate.

With $1.12 billion in cash/investments, positive adjusted free cash flow, and a clear path to Adj. EBITDA profitability in Q4 2026, financial risk is contained. The three growth drivers – Pet, Car, and Europe – provide diversified and sustainable growth. Tesla FSD insurance opens a new strategic dimension.

The valuation is not cheap compared to traditional insurers, but once the company reaches profitability, a re-rating is likely. With guided ~60% revenue growth in 2026 and a clear profitability timeline, this is fundamentally about execution risk – and so far, management has delivered consistently on its targets.

(If you are interested to learn more about this company and the technology, you should watch their Investor day 2024, its really good!)

Today, Iren uses 100% renewable energy for its operations, primarily located in Canada and Texas. These regions provide them with plenty of surplus energy, allowing them to utilize excess green power. This also puts them in a highly favorable position to help balance the local power grids while increasing energy efficiency. Their mining capacity reached 50 EH/s in the summer of 2025. It will not be expanded beyond this 50 EH/s cap; all future growth will come from AI. Just two years ago, the company had only 5.5 EH/s, highlighting their massive growth in recent years.

The company has also optimized its data centers for both Bitcoin mining and AI. This dual strategy provides much more stability and optionality for the business. Unlike other miners, Iren does not “HODL” (hold) the Bitcoin they mine. This provides better downside protection in the event of a major Bitcoin crash, though it slightly reduces the upside if the price were to skyrocket.

2025 was the year the company made a major push into AI and data centers. Today, they have issued guidance of $3.4B in ARR (Annual Recurring Revenue) from AI (140k GPUs), and roughly $1B from Bitcoin mining, assuming they maintain their 50 EH/s. Over the course of a single year, Iren will go from an ARR of $28 million to at least $3.4B. This means their AI revenue will multiply by 121x if management reaches that $3.4B target. I actually believe they will guide closer to $5B toward the end of 2026, but I will explain that later. So far, Iren has secured the following AI customers: Microsoft, Together.AI, FluidStack, and Fireworks AI, and they have already managed to become an NVIDIA “preferred partner.”

CEO, Management, and Experience

Dan (CEO) previously owned and managed Palisade Investment Partners, an asset management firm investing in infrastructure. The company specialized in the energy sector—including wind, solar, and gas—and grew to $6B in assets under management. Dan was also a Vice President at Macquarie, an investment bank with a $60B market cap. Will held a similar position at Macquarie, where he was responsible for commodities and global markets.

The brothers each hold about 4.2% of the total outstanding shares. The incentives are clearly there for both of them to run this company as profitably and efficiently as possible.

One reason I can maintain such a large allocation in this company is my strong trust in the management team. This trust is based on the results they have delivered over the past few years, as well as their values and long-term focus, which I will get into next.

My confidence in management is built on their values and long-term perspective. Dan and Will’s primary goal is to create long-term shareholder value, and they refuse to be distracted by competitors or short-term strategies. In 2024 and 2025, it became highly popular for Bitcoin companies and other smaller firms to buy up large amounts of Bitcoin, often diluting shareholders to fund these purchases. Following the success of MicroStrategy, spearheaded by Michael Saylor, other companies like Mara and CleanSpark—and the Bitcoin industry in general—began holding the Bitcoin they mined, while also buying more through debt and shareholder dilution. Iren was the only company that abstained from this, despite heavy pressure from shareholders and the broader market. They stuck to their long-term plan, to the ultimate benefit of their long-term shareholders. This decision likely saved investors a lot of money down the road, as those other companies later took a severe beating. I highly recommend watching this clip where Dan explains why they don’t HODL Bitcoin: https://www.linkedin.com/posts/iren_daniel-roberts-discusses-irens-non-hodl-activity-7179488519839465472-nAjG/

Another reason I trust Dan is that he doesn’t announce unrealistic “pipelines” (future data centers without an approved grid connection agreement). Other competitors casually throw out future pipeline projections without actually having approved access to the power grid. There’s no point in me writing too much about this; I think this 2-minute video explains the message perfectly without needing further commentary: Pipeline announcements (see from 34:50 to 36:50).

Clip from https://www.youtube.com/@McNallieMoney

Another reason for my high confidence in the management is simply looking at the results they have delivered so far. Iren grew from 0.8 EH/s in January 2022 to 50 EH/s by the summer of 2025. They grew faster than any other competitor for two years in a row. In 2023, they increased their hashrate by 250%, and they smashed that record in 2024 with a 402% growth rate (from 5.6 EH/s to 28.1 EH/s). Not only did they grow much faster than their competitors, but they were also by far the most profitable Bitcoin mining company—and actually the only one to turn a profit! Dan Roberts said this in an interview: “Targets look bullish and ambitious, but as you know, we don’t publish something unless we’re super confident we at least beat it,” and I think that sums up the management perfectly. They are focused on executing and never promise something they can’t deliver.

See the images and figures below for their Bitcoin mining results. As seen in Figure 2, Iren is compared to CleanSpark, which in my opinion is the second-best Bitcoin miner based on the numbers they report.

As the numbers in Figure 2 show, Iren has gross margins of 74.5% compared to CLSK’s 56.7%. After all costs are paid, Iren had a cost to mine one Bitcoin of $66,964, while CLSK was at $89,053 (this was at a time when the Bitcoin price was very high). The numbers speak for themselves; Iren has executed better than anyone else, and the hope is that they will be able to transfer this performance over to the AI/HPC segment.

Figure 1: Q2 2025, sourced from Agrippa Investments

Figure 2: April 2025, sourced from McNallie Money

The Rest of the Management Team

Iren has a highly competent team with decades of experience in the energy, finance, and data center sectors. All executives at Iren have over 20 years of experience in their respective fields. Iren’s most recent hire is John Gross (CIO, Chief Innovation Officer), whose appointment was announced on February 17, 2026. John has over 20 years of experience in data centers and specializes in high-density and liquid-cooled infrastructure. He possesses knowledge that is crucial for Iren as they design data centers for future GPUs, which will require high rack density and advanced cooling.

Iren also has a CTO (Chief Technology Officer) with over 20 years of experience in the cloud and data center sector. Denis Skrinnikoff has been part of the team since 2021, specializing in enterprise architecture. “Denis has been involved in the design, construction and management of Tier 3 data centers, acquisitions and mergers, and managing business units, most recently as Senior Director, Cloud and Colocation at TeraGo and previously as Senior Director of Technology at RackForce Networks, Inc.”

The role I consider the most important for Iren moving forward is the CFO (Chief Financial Officer). Iren hired Anthony Lewis for this position on July 1, 2025. The reason I see this role as crucial is that Iren will need to raise a lot of capital in the future, and they depend on favorable terms and interest rates to maximize shareholder value and the bottom line. Iren benefited from Anthony Lewis’s track record and experience almost immediately. Within a few weeks, Anthony managed to negotiate 100% financing for $100M worth of GPUs at an interest rate below 10%. This means the GPU purchases will not require cash upfront, making it easier for Iren to scale without tying up capital or diluting shareholders. The deal is structured with a 36-month loan period, and as long as Iren’s returns exceed the interest rate, they generate a profit without deploying their own capital.

The value of Anthony Lewis was also evident when Iren negotiated the financing for $5.8B worth of GPUs related to their Microsoft deal (November 2025). Here, Iren secured $3.6B in financing at an interest rate below 6%. By comparison, CoreWeave did something similar but at a 10% interest rate. These are massive differences that will compound significantly over time. The reason Iren secures such low interest rates is due to the company’s low debt ratio, which I will cover later.

Iren also has an impressive incentive plan, where the CEO is paid out if the stock price reaches $370, $650, $925, and $1,850 (2023). All the right incentives are in place for the company to grow into a massive enterprise. It has been changed to $57, $100, £142, $285 due to diluation adjustion (10/06/2025, see under).

(from https://iren.gcs-web.com/static-files/73181eb1-1ae4-4eb8-b0c9-a70077e37f2e page 42, note (3)

Over the past few years, Iren has built a highly competent team that is ready to scale into the most profitable, efficient, and forward-thinking data center company in the industry.

Their Success in Bitcoin and Why They Started With Mining

In late 2017 and early 2018, Will and Dan decided that Bitcoin was a “0-to-1” (revolutionary) invention and wanted to work with it. At the same time, the brothers realized their background gave them a massive advantage in the sector. This is because both had an exceptional understanding of the energy sector, particularly wind and solar power.

Since the beginning, Iren has employed a slightly different strategy than its competitors. Iren focuses on building its data centers from the ground up. This means they buy the land and build all the infrastructure themselves. As a result, the company is vertically integrated, giving them 100% control over their own facilities. Competitors in Bitcoin mining have typically pursued growth by acquiring smaller companies or through mergers and acquisitions (M&A). Iren has carried this greenfield strategy over into AI/HPC, which sets them apart from competitors who lease their facilities—something that eats into margins (I’ll get into this more later).

“Bitcoin mining was Step 1. Not because it was the end goal, but because it was the smartest first step. Unlike any other data center service, Bitcoin mining doesn’t require an external customer. You mine for yourself, for your shareholders. It let IREN build a track record of development execution, generate cash flow, and prove out its infrastructure capabilities at massive scale before layering in more complex, customer-facing services.”

Was It a Pivot to AI, or Was This Planned All Along?

From day one, Iren has been clear and transparent about wanting to do whatever is most profitable for the company and its shareholders. Iren has used Bitcoin mining as a fantastic opportunity to scale the company to its current size, where they own 4.5 GW of power capacity even before the AI revenue has fully kicked in. Over the past few years, it has become abundantly clear that it makes more economic sense to scale up in AI rather than Bitcoin mining. It yields better returns, while also providing shareholders with more stability and better access to capital.

Over the last year, almost all Bitcoin mining companies have transitioned to running AI/HPC alongside their mining operations. Because of this, it’s natural to assume that Iren also “pivoted” into AI, but this is where many get it completely wrong. Since the company’s inception, Iren has talked about AI/HPC and how it would eventually become a natural progression for the business. Iren has been “playing around” with AI since 2023 and purchased 248 NVIDIA H100s in August 2023. They have been leasing these out to Poolside since 2024.

This is the comment Dan Roberts (CEO) gave in the press release following the purchase of those H100s: “Leveraging our next-generation data centers into generative AI is an exciting opportunity, particularly given current industry shortages in rack space and compute. We believe demand for sustainable computing is unlikely to go away, and feel we are uniquely positioned to capture ongoing growth in the broader industry; whether that be ASICs for Bitcoin mining, or GPUs for generative AI and beyond.”

The transition to AI/HPC has been meticulously planned. Their Bitcoin mining centers in Canada (Canal Flats, British Columbia, and Mackenzie) had until recently operated exclusively for Bitcoin mining. However, these data centers were built from the start to accommodate AI/HPC, making the transition very seamless without any significant additional costs. See the video here:

https://x.com/IREN_Ltd/status/1971576440392216991

Dan has explained their strategy across many podcasts, and since 2020 (the oldest clip I could find), he has been talking about how AI/HPC is the future and how they would eventually transition into that market once it made economic sense—which has proven to be right now. I highly recommend watching this 2-minute video

Sourced from https://www.youtube.com/@familyofficecommunity3516

So, Bitcoin was never the end goal; it was simply the starting point for building a massive portfolio with access to power. Bitcoin made it easy for Iren to scale up without needing an external customer (you mine Bitcoin, with no middleman). At the same time, Bitcoin mining proved that Iren could successfully build large-scale data centers and generate strong cash flow. Bitcoin mining was merely step one before Iren began moving into AI/HPC. I personally believe that 2026 will be the last year Iren mines Bitcoin, and that from 2027 onward, they will be a 100% neocloud company.

Iren’s Path Forward in AI

As mentioned, Iren has been operating and testing AI/HPC for many years alongside their primary Bitcoin mining operations. Management has been clear about this transition, stating it would happen once it offered a better risk/reward ratio for the company and its shareholders. That time has now arrived, and Iren is on its way to becoming a 100% AI company, leaving the Bitcoin segment behind. If you visit their website today, it’s actually very difficult to tell that they even do Bitcoin mining.

Just a year ago, the information and signals provided by management suggested that Colocation (CoLo) was the natural transition. Colocation essentially means that Iren builds the data centers for the customers, while the customers buy the GPUs and install them in Iren’s facilities. This is the strategy Cipher Digital has pursued. This solution provides stable, steady cash flow and is less capital-intensive than the direction Iren has taken over the past year.

Instead, Iren has chosen to pursue a strategy called Infrastructure as a Service (IaaS). This means they finance and build the data centers, while also purchasing and owning the GPUs and all remaining equipment. Iren thus owns all the hardware and leases it out to a large customer like Microsoft, Meta, or Google (hyperscalers). Iren will not offer software or similar services; they provide solely the physical infrastructure a company needs to run their LLM models or similar workloads. This strategy is highly capital-intensive and requires significant debt financing.

The reason Iren chose IaaS over CoLo is the vastly superior payback period. The payback period for CoLo is nearly double that of IaaS. However, IaaS requires much more work and capital from management, which means higher risk and a larger potential downside. There is nothing stopping Iren from running a combination of both models, but according to management, any move must make economic sense to maximize shareholder value. The agreement with Microsoft has so far proven that this IaaS model makes sense. As I will cover later in the analysis, Iren managed to raise $5.8B in capital to finance 76,000 NVIDIA GB300 GPUs at an effective interest rate around 3%. They received a 20% prepayment from MSFT ($1.94B) and financed the remaining $3.6B at an interest rate below 6%. In practice, this means Iren can halve its payback period compared to Cipher Digital, financed at an interest rate of just over 3%. This makes tremendous economic sense, and I am glad management is pursuing this direction as long as the economics hold up.

Iren offers two different types of data centers:

• Air-Cooled: These are data centers that rely solely on air cooling. They are much cheaper to build than liquid-cooled data centers, but they cannot run NVIDIA’s future GPUs. Models like the Vera Rubin 200 (VR200) require a rack density approaching 200 kW. Rack density refers to the power density per rack, meaning how much power each individual rack consumes or requires. Iren has managed to run some of NVIDIA’s latest models (B200/B300) in their air-cooled data centers in Canada, which is incredibly impressive in itself (I’ll get back to this later). However, as mentioned, the newest models will require a rack density of 200 kW or more. The upper limit for air-cooled data centers is typically an 80 kW rack density. Therefore, future data centers built to handle these newer models will be equipped with liquid cooling.

• Liquid-Cooled: These data centers utilize liquid cooling to meet the massive demands of the latest and future models from NVIDIA and other GPU/TPU manufacturers. Liquid cooling drastically increases the cost of the data center compared to air-cooled facilities. Iren has previously built air-cooled data centers for $700K/MW, but with their latest Microsoft contract, the cost will be $14-16M/MW!

Future data centers will primarily feature liquid cooling. However, on their latest earnings call (Q2 2026, which corresponds to Q4 2025 for most regular publicly traded companies), Iren stated they had received increased demand for air-cooled data centers because they offer a lower “time to power.” In other words, they can deliver compute faster, which is critical since these companies need capacity now.

Iren holds what is perhaps one of the largest secured MW portfolios in the world, with 4.5 GW announced so far. Iren does not announce any sites until they are 100% certain they have contracts for grid connection. This means they have 4.5 GW of data centers with secured future access to power grids like ERCOT and SPP.

It is important to distinguish between the different types of MW. Last quarter, Iren announced a new 1,600 MW gigacenter; this is secured MW. This means they have secured future power for the site. Once they actually gain access to the grid, it becomes energized MW. For instance, Iren has their massive 1,400 MW site that will be energized in April 2026. Finally, Iren currently has their three data centers in Canada which are monetized MW. These are data centers that went from having secured power, to being energized, and are now generating revenue.

Today, Iren’s portfolio primarily consists of energized and secured power. But as most are aware, the demand for data centers is massive, and CAPEX (capital expenditures) among hyperscalers is increasing enormously.

In Q4 2025 alone, hyperscalers increased their AI/data center CAPEX guidance for 2026 to staggering amounts:

• Amazon: Increased to $200B (up from $131B in 2025)

• Alphabet (Google): Increased to $175-185B (up from $91B in 2025)

• Meta: Increased to $115-135B (up from $72B in 2025)

• Microsoft: Increased to $100-150B (up from $116B in 2025)

This means these companies alone have planned AI/data center expenditures of approximately $600B for the year 2026.

Their Energy Portfolio (4.5 GW)

Iren is likely the publicly traded company with the most secured power today at 4.5 GW. Current estimates from Cargoson suggest that total global data center capacity stands at 122 GW. Measured against today’s global capacity, Iren alone holds 3.7% of the world’s data center capacity. Iren’s biggest competitor, CoreWeave, has 3.1 GW of contracted power and aims to have 5 GW of AI infrastructure by 2030. Iren’s other competitor, Nebius, has increased its capacity to 3 GW. However, what separates Iren from these two competitors is that Iren owns its data centers outright, rather than leasing from a third party like CoreWeave and Nebius do.

Currently, Iren has an AI capacity of only 460 MW, and the resulting revenue hasn’t fully hit their quarterly reports yet. However, it will start increasing drastically from their Q3 2026 report (which corresponds to Q1 2026 for the rest of the market). In Q2 2026 (Q4 2025 for the broader market), Iren only had $16M in AI revenue.

This 460 MW of active capacity comes primarily from Canada (160 MW) and Childress (300 MW). The Childress capacity represents only 300 MW of the 750 MW total capacity contracted out. That 300 MW is currently contracted to Microsoft in a $9.7B deal over 5 years, where Iren will run 76,000 GPUs (GB300). These 4 data centers totaling 460 MW (just 10% of their total capacity) will generate $3.4B in ARR (Annual Recurring Revenue) alone.

Considering that just 10% of their data center portfolio generates $3.4B in ARR, Iren could potentially generate $35B in ARR by 2030 using their full 4.5 GW capacity, assuming identical conditions (though there is every reason to believe the unit economics for future deals will improve over time, which I will discuss later).

The Different Data Centers and Their Unique Features

Iren currently operates across 6 different locations, comprising a total of 7 distinct data centers. The reason there are 7 data centers across 6 locations is that Sweetwater 1 will have a capacity of 1.4 GW and will be ready by April 2026, while Sweetwater 2 will have 600 MW and will be ready by the end of 2027.

Their currently announced data center portfolio is located exclusively in Canada and the US. Iren has several smaller data centers in Canada and some massive ones in the US. There are a few differences between these facilities, and I will go through each one and explain why they are unique. I will cover them in chronological order based on when capacity comes online, starting with the capacity Iren already has today in Canada, where they have announced 40,000 GPUs for 2026.

The smallest data center Iren has today is Canal Flats in Canada, which sits at just 30 MW. The other Canadian data centers are Prince George at 50 MW and Mackenzie at 80 MW. What separates the Canadian data centers from the ones in the US is their PUE. PUE stands for Power Usage Effectiveness and is a metric used to explain how energy-efficient a data center is. It measures how much of the power is used for the actual IT equipment/servers versus how much is used for cooling and overhead. In Canada, Iren boasts a PUE of 1.1, which is arguably the best in the industry. In practice, this means only 10% of their power costs go toward cooling. So the energy overhead is just 10%. By comparison, Iren’s contract with Microsoft announced a PUE of 1.5. This practically means that the margins at Iren’s Canadian facilities are much higher (refer to the image below).

The reason for these massive differences is the climate and temperature variation between the US and Canada. Canada is much cooler, meaning less power is required to cool and maintain the GPUs.

Currently, the data centers in Canada also operate exclusively as an AI Cloud. This means customers rent the GPUs and services directly from Iren, typically for a period of months or longer. This is in contrast to the Microsoft contract, which is an IaaS (Infrastructure as a Service) agreement where MSFT purchases the capacity for a 5-year term.

Iren currently offers the following GPUs:

To date, Iren has announced roughly 40,000 GPUs for Canada during 2026. This capacity primarily involves B200 and B300 models utilizing air cooling. Very few competitors can manage to run these models using air cooling, which is another factor that sets Iren apart from the competition.

Iren has managed to build AI data centers for $700K/MW, achieving an 80 kW rack density capable of running some of NVIDIA’s latest models with air cooling instead of liquid cooling, all while maintaining a PUE below 1.1 (in Dan’s own words). That is a lot of numbers and information in one sentence, but it is incredibly impressive. For comparison, Iren is building out the Microsoft deal at a cost of $14-16M per MW. The difference between $700K and $14-16M is extreme. The trade-off is that Iren’s Microsoft deal features a rack density of 200 kW or more, and those data centers will handle the absolute latest GPU models like the VR200, which demand rack densities approaching or exceeding 200 kW.

I really must emphasize how impressive it is that Iren can achieve an 80 kW rack density in an air-cooled data center. This was largely seen as unattainable, but Iren has achieved it at their Prince George facility. The advantage of pulling this off with air cooling rather than liquid cooling is that the data centers are far cheaper and much faster to build. I actually asked Gemini Pro, “What is the maximum rack density an air-cooled data center has achieved?” and the answer I got was 40 kW. The only other example I found of another player breaking the 40 kW barrier is Intel, with a 43 kW rack density.

Iren’s own website states that their maximum rack density is 70 kW across all their data centers in Canada, though Dan himself has mentioned that they hit 80 kW at Prince George. But even a 70 kW rack density across all their Canadian data centers is a massive achievement, especially alongside an average PUE of 1.1!

Canada is essentially a cash machine for Iren, featuring sky-high margins and low CAPEX/OPEX.

As the image of Prince George above shows, Iren is building a liquid-cooled data center. Iren claims this is not an expansion of Prince George, but I and several others have our doubts. I believe this might be related to the further expansion of PG, which I’ll discuss below.

Iren’s Data Centers in Canada and Their Economics

Canal Flats (Canada):

• Size: 30 MW

• Operational: Yes

• PUE (Power Usage Effectiveness): 1.1

• ARR: $272.73M

• EBITDA Margins: 85%

• Payback Period: 3.6 years

• Pre-Tax Profit (Annual): $99.12M

Mackenzie (Canada):

• Size: 80 MW (To be upgraded to 100 MW)

• Operational: Yes

• PUE: 1.1

• ARR: $727.27M

• EBITDA Margins: 85%

• Payback Period: 3.6 years

• Pre-Tax Profit (Annual): $232.5M

Prince George (Canada):

• Size: 50 MW (To be upgraded to 80 MW)

• Operational: Yes

• PUE: 1.1

• ARR: $500M

• EBITDA Margins: 85%

• Payback Period: 2.7 years

• Pre-Tax Profit (Annual): $232.5M

If you are observant, you might have noticed I wrote that Mackenzie and Prince George are set to be upgraded from 80 to 100 MW and 50 to 80 MW, respectively. This is information you have to look very hard to find. It is public information, but it is highly obscured—meaning it is likely not priced in, in my opinion. I found this detail in an interview with Glen Harrison (VP of Operations in Canada) on a podcast with only 280 subscribers and 480 plays (see from 9:25 to 9:43).

Clip from https://www.youtube.com/@unblockingthechain

The clip is only 10 seconds long, where Harrison explains that they are going to expand these data centers. This will mean a lot for Iren in terms of revenue and margins coming out of Canada. Prince George alone gives Iren an ARR of $500M with just 50 MW, which means we can expect closer to $800M in ARR from Prince George in the near future (6–18 months).

Mackenzie is also slated for an upgrade from 80 to 100 MW, which will increase its ARR from $730M to $910M. This means that from Canada alone, Iren has a potential ARR of $1.98B. Their current guidance for Canada sits at $1.43B. That represents a potential increase of over $500M in ARR.

In the image below, you can see a tender Iren put out regarding the delivery of steel for “Mackenzie 100MW,” as highlighted in the text. As it also states, the delivery or completion date is April 1, 2026. This strongly suggests that the 100 MW expansion will be operational before too long.

To summarize the data centers in Canada:

• Low PUE = lower costs = higher margins.

• Running AI Cloud at these facilities.

• Slated to host 40,000 GPUs (based on current announcements).

• 3 centers currently at 30 MW, 50 MW, and 80 MW—where the 50 MW will become 80 MW, and the 80 MW will (potentially) become 100 MW.

• Currently only air-cooled data centers, meaning they cannot run the absolute newest GPU models that require over 130 kW rack density. However, a liquid-cooled data center is being built in Prince George right now.

• Iren has achieved an 80 kW rack density in Canada—the best in the industry—allowing them to run B200/B300 GPUs using only air-cooled facilities.

Data Centers in the US

Now let’s move on to the massive data centers Iren has in the US, or more specifically, Texas. Texas is known for being a state with far fewer regulations, where data centers are welcomed with open arms. Iren’s data centers in Texas are located far away from populated areas, so no noise or operational disturbances bother anyone. This is highly positive, though the downside is a longer commute for the people working for Iren. Iren’s most recent update noted that there were 2,000 workers on-site at Childress, preparing the data centers for Microsoft.

(The image below shows 8 cranes at Iren’s Childress facility, sourced from Frans Bakker’s X profile).

Iren currently holds 3 massive sites with approved power capacity. These data centers have capacities of 750 MW, 2,000 MW, and 1,600 MW, bringing their total capacity in Texas to 4,350 MW. None of these specific mega-expansions are fully completed today, and only 300 MW out of the 4,350 MW is currently “sold” or leased to customers. Both Childress and Sweetwater 1 & 2 are connected to the ERCOT power grid, the operator managing the Texas electricity grid.

Childress is Iren’s largest active site today, featuring 750 MW that is already energized. Childress is the facility Iren is working on the hardest right now, following their November 2025 contract with Microsoft for 300 MW (200 MW of critical IT load, due to a 1.5 PUE). This contract was valued at $9.7B over 5 years, where MSFT leases both the data centers and the GPUs from Iren.

This provides Iren with $1.94B in annual revenue. This specific data center footprint will consist of 4 distinct phases of 75 MW each, which is why they are referred to as Horizon 1 through 4. As of writing this (March 2, 2026), Iren is well underway with Horizon 1, which will likely start generating revenue in May 2026. Horizon 4 is expected to be completed by the middle of H2. I will go deeper into this contract later in the analysis.

Iren still has the remaining 450 MW of Childress’s 750 MW capacity available today. This capacity is currently being used for Bitcoin mining, but rumors suggest this is on track to be pivoted to AI/HPC. Some clever individuals likely tracked down Iren’s Bitcoin wallet address and noticed their mining hashrate had dropped from 50 EH/s to roughly 35 EH/s, indicating major changes are underway—though this cannot be officially confirmed yet. For those interested, here is the Bitcoin address: https://mempool.space/address/3EeHsUEnyYQkjdV7zT2v4KHid6B16RjkHn

I expect an announcement regarding a contract leasing out this remaining capacity will come sometime in 2026, with construction kicking off in 2027 (after they finish Horizon 1-4). This facility could potentially result in a $15B deal over 5 years (50% larger than Horizon 1-4), assuming a contract similar to the Microsoft deal.

Here are the numbers from the MSFT deal and my projections for Horizon 5-10 (the remaining 450 MW at Childress):

Horizon 1-4 (West Texas)

• Size: 300 MW

• Operational: Soon (Contracted with $MSFT)

• Construction: Expected completion around October 2026 with H4

• PUE: 1.5

• ARR: $1.94B

• EBITDA Margins: 85%

• Payback Period: 5.7 years

• Pre-Tax Profit: $389.5M

Horizon 5-10 (West Texas)

• Size: 450 MW

• Operational: Yes, but currently for Bitcoin mining, not AI/HPC

• Construction/Contract: H1 2027???

• PUE: 1.5

• ARR: $4.05B

• EBITDA Margins: 85%

• Payback Period: 4.4 years

• Pre-Tax Profit: $1,313.02M

Sweetwater 1 in West Texas

So far, we have walked through all the data centers Iren currently has operational or under construction. These data centers (Canada + Childress) account for 910 MW. These 960 MW (including the +50 MW upgrade in Canada) alone can generate $7.97B in ARR ($1.94B from Horizon 1-4, $4.05B from Horizon 5-10 (VR200?), and $1.98B from Canada).

However, Iren also has approved capacity for 3,600 MW across their massive Sweetwater 1 (SW1) facility at 1,400 MW, Sweetwater 2 (SW2) at 600 MW, and last but not least, their newest announcement from the Q2 2026 report: 1,600 MW in Oklahoma.

Sweetwater is viewed as Iren’s signature project. This is the site where Iren will prove what they are truly capable of, and a single 1,400 MW facility attracts many major players who want parts of—or the entirety of—the site. Sweetwater 1 (1,400 MW) was announced by Iren in November 2023 in their monthly update. SW1 is scheduled to be energized in April 2025/2026, and it will likely become one of the largest operational data centers in the world. Initial groundworks and terrain preparation have already begun—perhaps making it ready for a future customer and contract?

Below is a satellite image of Sweetwater 1 taken on February 15, 2026. In the image, you can identify 3 primary stations (main substations = the heart of the power supply), which are expected to support 750 MW (450 MW gross IT load) available at 34.5 kV/36.5 kV during the latter part of H1 2026. Developing this facility will likely require a continuous workforce of over 2,500 people for at least 12–18 months. Building out a potential 1,400 MW will take a long time. It will take Iren roughly 9 months to build Horizon 1-4 (300 MW), so SW1 will demand more workers and take up to 2 years to fully build out. It will likely be completed in phases, just like the MSFT deal. In the MSFT deal, the first 75 MW will probably be ready in May, and the final 75 MW (of the 300 MW total) will be finished in October.

While the exact construction timeline for SW1 remains unclear, once completed, it will be one of the largest data centers globally.

The big question investors have regarding Sweetwater 1 and future data centers is what strategy Iren will choose. Will they pursue IaaS for the entire 1,400 MW, opt for Colocation for all of SW1, or potentially use a hybrid approach? The possibilities are vast, but the signal Dan has sent over the last few quarters is that IaaS offers a significantly better payback period than Colocation, even though it requires much more capital and potential shareholder dilution.

As Iren stated in their Q2 2026 earnings call:

• “When you’ve got scarcity around how many data centers you can physically bring online, every incremental 200 MW can deliver either $300 million-ish through a colocation or multiples of that in the billions under a cloud contract. So when we look at the monetization opportunity for our platform and growth for shareholders in creating value, the cloud opportunity creates a lot more upside, as we see it. And in terms of colocation versus cloud, we believe that AI is going to continue. We believe that data center demand is going to continue to compound, so recovering capital back in short order and being able to compound those returns, as distinct from holding effectively a bond exposure against a hyperscaler, is the area we want to play. And if investors want bond-like exposures, they can buy colocation companies, they can buy bonds in the hyperscalers.”

• “But we believe we’re offering a high conviction, but very well-managed risk exposure to the sector through this AI cloud business. Now, again, we’re not dogmatic. Things can change quickly. We get a compelling colocation deal, we will absolutely pursue it, but right now, AI cloud looks very compelling for all those reasons.”

The reason Dan sees a much better risk/reward in financing and purchasing the GPUs themselves is based on how their contract with MSFT played out. They received a 20% prepayment on the $9.7B deal ($1.94B) from MSFT, and raised the remaining capital ($3.6B) at under 6% interest. This means the average interest rate Iren pays to own the GPUs outright is just over 3%. As Dan himself said in Q2 2026:

“we’ve now secured 95% of the cost of the GPUs at an average interest rate of around 3% when you factor into the prepayment. So we essentially got the GPUs for next to nothing. So I think in terms of capital intensity, it ticks that box.”

If Iren can continue executing this strategy, it will result in vastly higher margins and better returns for investors compared to simply using the facilities for Colocation.

Job Postings for Iren at Sweetwater 1

There are currently 10 job postings from Iren regarding the Sweetwater facility. This can often be a strong indicator of the direction a company is heading. The latest jobs were posted on March 2, 2026, and these are the roles:

• The Procurement Manager will lead procurement activities for a new data center site, with an initial focus on supporting capital construction and infrastructure projects. This role will play a critical part in delivering large-scale mechanical, electrical, and critical infrastructure packages on schedule and within budget.

• The Contracts Coordinator supports the full lifecycle management of capital and operational contracts across data center construction and live site operations. This role plays a critical coordination function between Procurement, Legal, Finance, Project Management, and external contractors to ensure contracts are properly drafted, negotiated, executed, tracked, and archived in accordance with company policies.

It is up to interpretation what to make of this, but it is interesting regardless. I personally expect an agreement will materialize before energization (April 2026).

The players I believe would be interested in SW1 are: Google, Amazon, Meta, OpenAI, Anthropic, xAI, and Microsoft? (Though I doubt the latter since they already have Childress). I personally hope and believe it will be Google.

These are the numbers I expect/assume Iren could achieve for SW1 with an IaaS agreement:

Sweetwater 1 (West Texas)

• Size: 1,400 MW

• Operational: No (Energized April 2026)

• Construction/Contract: Throughout all of 2026, 2027, and likely 2028 as well.

• PUE (Power Usage Effectiveness): 1.5

• ARR: $16.1B

• EBITDA Margins: 85%

• Payback Period: 3.5 years

• Pre-Tax Profit: $7.014B/yr

For those curious, here are the calculations made for SW1 and how I arrived at these numbers (I used @jiahanjimliu’s model, which can be found here: https://jiahanliu.github.io/IREN-Community/):

Sweetwater 2

Iren announced an additional 600 MW in Sweetwater, which became their SW2 facility, on March 17, 2025. This facility is 600 MW (+500 acres) and will be energized at the end of 2027. This means the data center will not be 100% operational until towards the end of 2028, but it will be built out gradually in the same manner as Horizon 1-4, likely also in 75 MW batches. Sweetwater 2 is located 28 miles from SW1, and these data centers will be connected via direct fiber.

Oklahoma

Oklahoma is their most recently announced data center at 1,600 MW. This data center was announced in Iren’s Q2 2026 report (February 5, 2026). This surpassed the SW1 data center’s 1,400 MW and will become Iren’s largest when it becomes fully operational toward 2030. The data center is scheduled to be energized during 2028, meaning it overlaps nicely with their SW1 facility, which will likely be completed and operational by that time.

This center uses SPP for grid exposure, compared to the other centers (SW1, SW2, and Childress) in Texas, which use ERCOT. The only information I have about this potential data center comes from a meeting in Pittsburgh to discuss a tax break for Iren in OK. Present at the meeting were Giles Walsh (VP of Operations), David Shaw (Head of Data Center Operations), and Jason Date (Commercial Team). Iren presented their estimates for earnings and revenue, and these figures would equate to 25% of the county’s revenue. Discussions also indicated that the OK site would be launched in 3 phases, with the first phase being 600 MW. The meeting was said to be positive, with no conflicts arising between the parties. The next meeting will be held on April 6, 2026.

Future Energy Portfolio (Multi-GW)

I won’t speculate on how many unannounced sites Iren will secure power from. However, as Iren themselves wrote following their OK announcement (1,600 MW), they are still working on a “multi-GW development pipeline,” which leads one to assume that further data center announcements are on the horizon.

In their Q2 2026 earnings call, Kent Draper also had this to say regarding ERCOT’s Batch Zero meeting: “In addition to that, yeah, there are other projects in our portfolio that are potentially also included in Batch Zero.”

I interpret this to mean that Iren might have more data centers in Texas that haven’t been announced yet, meaning Iren could potentially have more MW energized in the near future (1-6 months). Several large investors in Iren, such as Agrippa, believe that their future unannounced portfolio sits at over 3.5 GW.

Another point is that I believe Iren will potentially start building data centers in regions outside of Canada and the US. Dan Roberts posted this early this morning when I woke up (March 3, 2026):

He specifically says, “Building AI cloud around the world today,” which suggests they are talking about multiple data centers across multiple continents. There are some rumors about potential data centers in Australia, Japan, Singapore/Indonesia/Malaysia, and possibly Europe. These are just rumors and nothing more, but I do expect that new data centers outside the US and Canada will be coming from Iren in the future.

In-Depth Analysis of the $9.7B Microsoft Contract

In November 2025, Iren announced its first IaaS contract with Microsoft for a total value of $9.7B over a 5-year period. In practice, this means that Iren will build the data centers and supply all the equipment and GPUs. Iren remains the sole owner of all components and equipment, and Microsoft leases all of this from Iren for a period of 5 years. The agreement covers 300 MW with 200 MW of critical IT load.

Contract Specifications and Financials

Here is a breakdown of the contract specifications and financial details:

• Total Value: $9.7B ($1.94B in ARR) over 5 years.

• EBITDA Margins: 85%.

• Upfront Payment: A 20% prepayment from Microsoft ($1.94B), which is highly lucrative for IREN.

• GPU Capital Expenditure: $5.8B allocated for GPUs (including servers, InfiniBand, cabling, etc.) via an agreement with Dell. This covers 76,000 Nvidia GB300s, with a potential upgrade path to the VR200.

• 

Costs (per MW):

• $9M–$11M related to data center infrastructure.

• $3M for the deployment of a 100 MW supercluster architecture with flexible rack densities (130–200 kW), capable of running Nvidia’s VR200 in the future.

• $2M to expedite the Horizon 1-4 data centers (an additional premium for faster delivery).

• Totaling $14M–$16M per MW. (For context, IREN’s previous guidance was $6M–$7M per MW before securing the MSFT contract; see image).

This brings the total cost for Horizon 1-4 to $3B, excluding the $5.8B cost associated with the GPUs.

To summarize, the total capital expenditure to build out the data center infrastructure and purchase the GPUs will be $8.8B.

On paper, securing a $9.7B contract against a cost of $8.8B might not look like a stellar deal. However, it is crucial to remember that IREN will retain full ownership of a fully developed data center with a 20- to 25-year lifespan. Furthermore, they will retain the potential residual value of the GPUs (estimated at 10–20%), equating to $580M–$1.16B.

Trust and Credibility

One of the aspects I value most about this Microsoft contract is the trust IREN has secured and the opportunity it provides them to prove their capabilities. The MSFT contract covers only 300 MW out of their 4,500 MW capacity—just 6.67% of their approved grid-connected data centers. In my opinion, this is the defining contract that will demonstrate whether IREN can deliver a high-quality product to hyperscalers (like Google, Microsoft, Amazon, etc.) strictly on schedule.

I believe IREN had to accept slightly less favorable unit economics on this initial deal to prove their execution capabilities. Doing this with only 6.7% of their portfolio is a highly respectable strategic move. Beyond the numbers, Microsoft’s choice to partner with IREN provides massive validation for the company. As management themselves stated regarding the contract:

• “This deal matters for reasons beyond the economics. Microsoft does not sign nine-figure, multi-year contracts with companies it doesn’t trust. This is a de facto institutional endorsement of IREN’s development capabilities, infrastructure quality, and operational reliability. It’s the kind of validation that cannot be manufactured.”

And:

• “A counterparty like Microsoft, they don’t enter into these deals lightly. It was many months of detailed due diligence. And the due diligence that they did was on every aspect around our business. The business itself, the management team, the datacenter design, the site management, the delivery capabilities that we have, our ability to stand up large scale GPU clusters and operate them. It gives us an air of credibility now, and even since the deal was announced we had a large amount of inbound inquiries. And the other discussions we were already having with other hyperscalers, you know have significantly ramped up.” (See 8:21 to 10:44)

Clip from: https://www.youtube.com/@McNallieMoney

As CCO Kent Draper explains, this heavily legitimizes the company and has smoothed the path for negotiating subsequent deals. If they successfully deliver for MSFT, it serves as ironclad proof to other prospective clients that a former Bitcoin mining company can be trusted to deliver top-tier AI/HPC data centers.

The Math Behind the Deal

I believe the agreement must be divided into two distinct parts to accurately understand the return on investment. It doesn’t make sense to value this deal based purely on a 5-year timeframe when the data center infrastructure has a lifespan of at least 20 years.

These calculations were performed by Agrippa Investment, and I highly recommend reading and subscribing to his Substack (linked at the bottom). The reasons I haven’t done these calculations myself are:

• Agrippa has already modeled this out in a highly thorough and well-organized manner.

• I personally lack the time and specific expertise to run all these different financial models.

GPU Calculations (Assisted by AI + Agrippa):

• Unlevered IRR:

  • 10–12% (low double digits) assuming the GPUs have $0 residual value after 5 years.

  • 17–20% (high teens) assuming the GPUs retain 20% residual value (based on a 6.25-year depreciation schedule).

• Levered IRR:

  • 25–30% if they secure $2.5B in debt financing (assuming $0 residual value).

  • 35–40% if debt financing is increased to $3B (assuming $0 residual value).

  • 35–50% if we also factor in a 20% residual value for the equipment, depending on whether the loan is $2.5B or $3B.

Data Center Infrastructure Calculations:

• Internal Rent: The company calculates an internal rent of $130 per kW per month (plus an annual 3% escalator). This generates approximately $1.56M in rent per megawatt (MW) in the first year.

• Operating Expenses (OPEX): Approximately $291,000 per MW is allocated to the colocation segment (the remainder is shifted to the cloud segment).

• Financing: Agrippa assumes a 50/50 equity-to-debt split for the data center CapEx, utilizing a 10-year loan at a 7.5% interest rate.

• Earnings per MW: Initially, this generates an average profit of roughly $500,000 per IT MW.

  Kent Draper, CCO, Return on MSFT deal:

Clip from: https://www.youtube.com/@McNallieMoney

IREN’s Own Guidance on the MSFT Deal:

• Unlevered IRR in the “low double digits” (est. 11–13%).

• Assuming a 20% residual value for the chips (a 6-year GPU lifespan), Unlevered IRR jumps to the “high teens” (est. 17–20%).

• Levered IRR of 25–30% with zero GPU residual value and $2.5B in financing.

• Levered IRR of 35–40% assuming $3B in financing.

• When assuming a 20% residual value for the GPUs, the Levered IRR approaches 50%.

As we know from their Q2 2026 reporting, IREN secured $3.6B in financing at a lower interest rate than most anticipated (under 6%).

As Kent Draper notes, the deal is justifiable simply through the direct returns IREN earns on the contract. However, when you factor in the credibility aspect, this agreement alone is the catalyst for their transition from a pure-play Bitcoin miner into an advanced, fully-fledged neocloud provider—and the company’s valuation is following suit. The rationale is straightforward: the market assigns higher valuation multiples to AI companies than Bitcoin mining companies, especially since AI/HPC operations offer superior returns and fundamental stability.

The VR200 Renegotiation in the Microsoft Contract

What wasn’t highlighted in the initial Microsoft announcement can be found in their latest 10-Q associated with the Q2 2026 report. The contract includes a renegotiation clause to upgrade the GPUs from the GB300 to NVIDIA’s newest models, the Vera Rubin (VR200), slated for launch in early H2 2026. This means the data centers IREN is building for Microsoft are designed to handle next-generation models that aren’t even in production yet.

This proves IREN’s long-term vision; their data centers are engineered for the future, not just the present (Tier 3 data centers). The contract stipulates that these negotiations can occur if Microsoft provides a 4-month notice, giving the companies 30 days to discuss new terms tied to these latest models.

I believe this benefits both parties, with IREN standing to gain the most due to the recent surge in rental prices for older NVIDIA models. This will likely be highly relevant for Horizon 3 and 4, which won’t be completed until around September or October. Therefore, I see a strong probability that Horizon 1 and 2 will run on GB300s, while Horizon 3 and 4 will deploy VR200s following negotiations with MSFT. Regardless, this is excellent news that highlights IREN’s capability to build data centers with highly demanding technical specifications, pushing rack densities closer to—or above—200 kW.

Images of rental prices from @matthew_sigel

The image above illustrates the rental rates for older NVIDIA models. We can see that demand for these GPUs has drastically increased, driving rental prices up by nearly 25% in 2026 alone. Even though these chips are three years old, demand remains robust because their quality and capacity are still highly relevant. This massive need for compute capacity drives up the residual value of older GPUs, which in turn increases the potential future residual value of the hardware IREN currently holds.

A Glimpse into IREN’s Future

Up to this point, I have covered what I consider most relevant regarding IREN’s past. In this final section of the analysis, I will outline what is relevant moving forward, and what investors can expect from the company in terms of upcoming news and contracts.

As I write this (10:07 AM, March 4, 2026), it has been months since the company released any major news. The last announcement was the 1,600 MW site in Oklahoma, but such news has little impact on the stock price until the company can deliver and prove its current and future contracts through tangible cash flow. Currently, only $16M related to AI/HPC has appeared in their quarterly reports, but this will increase drastically in the coming months and years. Aside from Oklahoma, investors haven’t heard any significant updates since the Microsoft deal in November 2025.

The core thesis for IREN’s future is my expectation that all upcoming contracts will be an improved version of the MSFT deal—yielding higher revenue per MW and a stronger bottom line for the company.

I initially published this analysis in Norwegian a few days ago (March 4, 2026), and there has been major news since then. Therefore, I have decided to include it in this update:

IREN Expands AI Cloud Capacity to 150,000 GPUs (04.03.26)

Iren is increasing its GPU count and ARR guidance from 140,000 GPUs to 150,000 GPUs. Consequently, the ARR is increasing from $3.4B to $3.7B—a $300 million increase driven by these 10,000 additional GPUs.

The recent purchase consisted of 50,000 NVIDIA B300s. These GPUs will be delivered in phases throughout H2 2026 and will be deployed to Mackenzie, British Columbia, and Childress for air-cooled data centers. Note that Childress is explicitly mentioned here. Iren has already leased out 300 of the 750MW in Childress, and the remaining 450MW are not yet operational for AI/HPC. This suggests that this capacity will be built out as air-cooled data centers. The advantage of this approach is a significantly lower CAPEX ($2M-$4M/MW), which results in a faster time-to-power. This indicates that either a contract for the B300 GPUs in Childress will materialize in the coming months, or Iren plans to run its own cloud solution on them.

These 50,000 GPUs had a total cost of $3.5B, which equates to an average cost of $70,000 per GPU.

Out of these GPUs, approximately 33,000 are heading to Canada and 17,000 to Childress. This means that the 17,000 GPUs in Texas will require roughly 50MW of power.

ATM for 6B (04.03.26)

Iren also announced a new $6B (40% of mcap)ATM alongside the purchase of 50,000 GPUs. An ATM is a capital-raising method that Iren has utilized many times in the past. The dilution hasn’t happened yet, and it is not guaranteed that the facility will even be tapped into. This is entirely at Iren’s discretion. The market reacted negatively to the news, sending the stock down 8% the following day. Nobody likes getting diluted, but remember that the two individuals taking the biggest dilution hit are Dan and Will with their 5% ownership stakes. An ATM can be a solid way for companies to fund growth without taking on excessive debt. The real question is whether the company can convert this capital into further growth, and given their track record, I have no doubt they will succeed.

Dilution is tough, but it will be necessary for the company to keep growing at the pace demanded by this sector. What makes this specific ATM interesting is that I don’t see any immediate reason for them to use it right now, unless it serves as a security backstop for a potential deal. Perhaps a counterparty needs assurance that Iren has the capital to build out the data centers, making the ATM act as a financial guarantee?

The last time Iren initiated an ATM, it was for $1B—just weeks before they signed a contract with Microsoft. Today’s ATM is 6x larger, only 6 months later, at roughly the same market cap.

Before that, the last time I recall them tapping into the ATM was at $6 per share, and the stock proceeded to 10x over the following six months. Dilution might sound intimidating, but when utilized strategically, it is a fantastic tool for creating shareholder value!

There will be a lot of speculation forward, but what ultimately matters here is whether you trust that Iren is making the right decisions for the company’s future and shareholders, I believe so.

Anticipated Catalysts and News for 2026

I expect the following developments in 2026, with many likely to materialize in the coming months:

1. Canada (Software Solution?)

• I believe this will either involve Microsoft—given their strong interest in Canadian data centers—or a player just below the hyperscaler tier (like Fluidstack, Together AI, etc.). I expect news on this very soon; it wouldn’t surprise me if it’s announced before I even finish this analysis. As previously described, the Canadian data centers are much smaller than the US sites, but they can start generating revenue very quickly because the facilities are already operational—they just need to plug in the GPUs.

Their Q2 2026 report also mentioned an IREN software solution for the first time. IREN hasn’t offered this before. Dan Roberts explained this on the Q2 2026 earnings call:

“We do have an internal software capability. I think we probably downplay it a bit, partly in response to the market seeming to overplay it, but we’ve got that capability. To give you additional comfort, one of the contracts we are negotiating at the moment is a multi-billion-dollar contract where we need to bring a software solution. So it is not holding us back... The reality is exactly as what Kent said, you are dealing with the largest technology companies in the world. To pretend that you can be better at software and jam something down their throat when that is their competitive moat and that is their expertise, it’s just not congruent with reality.”

What this actually means for IREN and what exactly the software entails is hard to say, but I expect this “multi-billion-dollar” deal will involve a moderate software solution originating from their Canadian data centers.

2. Sweetwater 1

• As mentioned in the analysis, Sweetwater 1 is set to be energized in April 2026 (IREN has interchangeably referred to this as Q2 and April). Since April is only a month away, I expect IREN is closing in on a contract for at least the first 300 MW to be built. The company has had two years to negotiate contracts for this site, and I hope they can finalize this soon. Personally, I believe this will be a much larger contract than the MSFT deal, based entirely on the VR200 rather than the GB300. I also expect the CapEx/MW to be lower than the MSFT deal, assuming no extra costs for security or fire prevention arise. Post-Microsoft, IREN has learned how to build with maximum efficiency and profitability, and they will apply these lessons to reduce costs for future build-outs like SW1.

3. Handover of Horizon 1 to Microsoft

• I expect IREN to meet its deadlines and officially hand over Horizon 1 to MSFT in May, complete with GPUs ready to generate revenue for both parties. I also anticipate they will simultaneously announce that they are on track for Horizons 2 through 4.

4. Horizon 5–10

• Following a Sweetwater contract and the completion of Horizon 1-4 for MSFT, I believe IREN will begin building data centers for the remaining 450 MW in Childress. This capacity is currently used for Bitcoin mining, which I expect the company to phase out during 2026 or early 2027. This would result in a data center 50% larger than the MSFT Horizon 1-4 build (300 MW). Could be a aircooled datacenter after management comment in Q2 2026.

5. Expansion of Mackenzie and Prince George

• I expect an announcement during 2026 confirming the expansion of Mackenzie (from 80 to 100 MW) and Prince George (from 50 to 80 MW). This will begin impacting the financials as early as 2026. They have shared that this is underway, and I believe 2026 is the year it materializes.

  6. GPU Acquisitions from Other Neoclouds

• There is an ongoing shortage of GPUs, and deliveries have taken longer than expected. Rumors suggest IREN might purchase GPUs from competitors who lack the necessary infrastructure to install them. This could accelerate their “Time to Power” and positively impact the balance sheet.

  7. More Compute and an Increased 2026 ARR Guidance

• As mentioned earlier, I interpret that IREN potentially has more data centers in “Batch 0” in ERCOT. This means that over the next 12 months, new data centers could be announced that are ready to be monetized or built shortly. Simultaneously, I believe the company will raise its ARR guidance from $3.4B to a range closer to $5B–$7B.

8. NVIDIA Takes a Stake in IREN (Pure Speculation)

• NVIDIA recently invested $2B in competitor CoreWeave (CRWV), and I firmly believe a similar investment will happen with IREN once they have more time to prove their execution capabilities to the market. This is purely speculation, but I think it could happen within 24 months.

  9. Relocating Headquarters to the US

• Given that major contract negotiations and the bulk of operations are based in the US, I believe the company will relocate its headquarters from Sydney to the States. This move alone could allow the company to be included in more passive index funds, leading to increased exposure and stock stability. Management has discussed this previously, and I believe it could be a positive catalyst for the stock in 2026 (+5–10%). (Note: This is not confirmed by IREN).

Future Price Target (Bull Case for IREN)

Using @jiahanjimliu’s model, I have outlined a bull case for what Iren might look like heading toward 2030, assuming the company successfully delivers on its secured capacity. I highly recommend taking some time with this model to input your own assumptions and data to match your expectations for bear, base, and bull scenarios: https://jiahanliu.github.io/IREN-Community/

This involves many assumptions and should in no way be considered financial advice or a recommendation. In my opinion, this represents the best-case scenario for $IREN:

To make this as clear as possible, I will break down each individual data center. At the end, everything is summarized, including the underlying assumptions I have factored in.

Data Center Breakdown

• Canal Flats (Canada)

  • Capacity: 30 MW

  • Status: Operational

  • PUE (Power Usage Effectiveness): 1.1

  • ARR (Annual Recurring Revenue): $272.73M

  • EBITDA Margin: 85%

  • Payback Period: 3.6 years

  • Pre-Tax Profit (Annual): $99.12M

• Mackenzie (Canada)

  • Capacity: 80 MW

  • Status: Operational

  • PUE: 1.1

  • ARR: $727.27M

  • EBITDA Margin: 85%

  • Payback Period: 3.6 years

  • Pre-Tax Profit: $232.5M

• Prince George (Canada)

  • Capacity: 50 MW

  • Status: Operational

  • PUE: 1.1

  • ARR: $500M

  • EBITDA Margin: 85%

  • Payback Period: 2.7 years

  • Pre-Tax Profit: $232.5M

• Horizon 1-4 (West Texas)

  • Capacity: 300 MW

  • Status: Soon to be operational (Contracted with $MSFT)

  • Construction: Expected completion around October 2026 (including H4)

  • PUE: 1.5

  • ARR: $1.94B

  • EBITDA Margin: 85%

  • Payback Period: 5.7 years

  • Pre-Tax Profit: $389.5M

• Horizon 5-10 (West Texas)

  • Capacity: 450 MW

  • Status: Not operational

  • Construction/Contract Target: H1 2027 (?)

  • PUE: 1.5

  • ARR: $4.05B

  • EBITDA Margin: 85%

  • Payback Period: 4.4 years

  • Pre-Tax Profit: $1,313.02M

• Sweetwater 1 (West Texas)

  • Capacity: 1,400 MW

  • Status: Not operational (Energized in April 2026)

  • Construction/Contract Target: Throughout 2026, 2027, and likely 2028

  • PUE: 1.5

  • ARR: $16.1B

  • EBITDA Margin: 85%

  • Payback Period: 3.7 years

  • Pre-Tax Profit: $6,874.04M

• Oklahoma

  • Capacity: 1,600 MW

  • Status: Not operational

  • Construction/Contract Target: 2028/2029

  • PUE: 1.5

  • ARR: $14.4B

  • EBITDA Margin: 85%

  • Payback Period: 4.3 years

  • Pre-Tax Profit: $4,721.86M

• Sweetwater 2 (West Texas)

  • Capacity: 600 MW

  • Status: Not operational

  • Colocation: Yes

  • Construction/Contract Target: Late 2027

  • PUE: 1.5

  • ARR: $0.73B

  • EBITDA Margin: 85%

  • Payback Period: 8.2 years

  • Pre-Tax Profit: $432M

Summary & Valuation

To summarize, these 8 data centers—for which Iren already holds approved capacity—would generate the following:

• Annual Revenue: $38.73B

• Annual EBT (Earnings Before Tax): $14.33B

• Annual Net Profit: $13.74B

Market Cap & Share Price Calculation:

• $13.74B × 35 (P/E Ratio) = $480B Market Cap

• Assuming a future shareholder dilution of 40%.

• $480B / 572.8M shares (current 409M shares × 1.4).

Calculated Share Price: $840

Underlying Assumptions:

• Operational capacity is set at 4,500 MW.

• All data centers (excluding SW2) will operate as IaaS (Infrastructure as a Service). This means Iren purchases the compute chips and leases out the data centers alongside the GPUs. This is highly capital-intensive and will place heavy demands on management.

• All data centers successfully secure contracts.

• The residual value of the GPUs will be 15% after 5 years (Example: Iren buys $5.8B in NVIDIA GPUs from Dell. After a 5-year agreement with Microsoft, the hardware can be sold for $870M).

• Future contracts will require 20% prepayment, like the Microsoft agreement.

• Capital raised to finance GPUs and CapEx will carry a 6% interest rate. (Iren raised capital at under 6% for the MSFT deal, but it is reasonable to expect rates to rise as the company takes on more debt).

• Data center lifetime is estimated at 20 years.

• The contract for Sweetwater 1 (SW1) will be 20% more lucrative than the MSFT contract.

• Upcoming data centers will secure contracts 15% better than MSFT (based on leasing prices for the Hopper model).

• Cost per MW remains around $15M (This is a conservative estimate; costs will likely decrease toward $12M/MW).

• A PUE of 1.5 for all non-Canadian data centers (also a conservative estimate).

• Shareholder dilution of 40% (This figure could be significantly higher depending on execution and macroeconomic conditions).

• GPU lifespan is set at 5 years.

• No additional data centers are announced by Iren, despite strong indications that more are coming.

• No future expansion of Mackenzie (from 80 MW to 100 MW) in May 2026, nor expansions of other data centers.

• Zero future Bitcoin mining.

• A P/E multiple of 35.

• No renegotiation of the MSFT agreement to upgrade to potential new models, such as the VR200.

(This thesis relies heavily on assumptions, and the actual outcome is guaranteed to differ from this scenario. Investing in Iren carries massive risk; the company needs to raise closer to $100B in capital to fully monetize and operate IaaS across all its facilities. Ultimately, the core investment thesis rests entirely on trust in management—which I personally have.)

Iren’s Competitive Advantages

1. Fully Integrated Supply Chain

Unlike the vast majority of its competitors, Iren is 100% fully integrated. They own and have built all their data centers themselves. This prevents margins from bleeding out to third parties. Competitors like Nebius and CoreWeave rely heavily on leasing capacity from third parties, which erodes profitability. Iren secured this power capacity long before the current market shortages arose, and well before the ChatGPT/LLM wave hit.

As Dan (co-CEO) stated in a podcast: “Why put your destiny in the hands of a 3rd party? Control it yourself because people let you down. And if you want something done, you’re better off doing it yourself.” Not only do they avoid the friction and disappointments of dealing with third parties, but the company also retains all the margins, which are estimated to be around 20%.

2. Compute Capacity

• Iren possesses 4,500 MW of secured power. This is a massive advantage that allows them to bypass the long interconnection queues currently plaguing the power grid. As competitor Nebius noted during their earnings call regarding their Meta contract: “The size of that contract was limited to what power they had available. If they had more, they could have sold more.” All available capacity is instantly sold, and customers are lining up.

A year ago, the ERCOT waitlist was 63 GW. By December 2025, it had grown to 226 GW, and it has likely increased since. The capacity simply isn’t there to meet demand, and Iren is sitting on an enormous portfolio. OpenAI recently partnered with Oracle and Softbank on “The Stargate Project” for 10 GW by 2029, at a cost of up to $500B. When Iren’s SW1 data center (1,400 MW) is energized in April 2026, it will be a highly attractive asset for heavyweights like Google, Meta, and Anthropic.

3. 100% Renewable Energy

Since day one, Iren has operated exclusively on renewable energy. This is a crucial asset for clients who are bound by strict corporate climate goals. It positions Iren as a preferred partner for companies seeking to operate as green and sustainably as possible to bolster their ESG reporting.

4. Purpose-Built for Extreme Performance

• Iren has already built data centers that push the boundaries of what is technologically possible today (such as the 80 kW rack density in Prince George). They are now building facilities designed to handle the absolute latest VR200 models. This is highly impressive and proves the company has accumulated deep expertise through its build-outs. It also reflects the strength of a management team with decades of combined industry experience.

5. Community Impact and Strategic Location

Iren is known for strengthening the local communities in which they operate. When building new data centers, they prioritize local labor and materials. For instance, all interior furnishings in their data centers are purchased locally, and Iren actively provides grants to local charities.

Furthermore, Iren locates its facilities far from residential areas. This minimizes issues related to noise and visual disruption for locals. Recently, there have been civilian protests against data center expansions by other operators, driven primarily by fears of noise, as well as rising water and electricity prices. This risk is virtually nonexistent for Iren, especially given that their largest data centers are situated in remote areas of Texas.

Competitors

I consider CoreWeave and Nebius to be Iren’s biggest competitors. Both are neocloud providers executing a strategy similar to Iren’s. What sets them apart is that Nebius claims to have a software component integrated into their AI initiatives. The significance of this is difficult to determine, as Nebius keeps their cards very close to their chest regarding their contracts.

Nebius also holds a massive $17.4B contract with Microsoft over 5 years (300–350 MW IT load). Information is incredibly scarce; the only thing to go on are rumors that this will involve over 100,000 GB300 GPUs, and no CapEx has been announced in connection with the deal. This lack of transparency is exactly why I stay away from the company, even though I have personally owned the stock in the past (at $25). The fact that information is withheld makes me feel they have something to hide. Transparency is crucial to me as an investor, and it is something I highly value with Iren. Nebius themselves have guided for over 3 GW of capacity for 2026.

CoreWeave is the most established player in the industry, with 810 MW of active power capacity. They have expanded their total contracted power to 3.1 GW (3,100 MW). The company currently has over 250,000 GPUs and brought 100,000 GPUs online in 2025. They also carry a high debt burden, with a Debt-to-Equity ratio of 8.9. In comparison, Iren sits at 1.5, and Nebius at 1.1 (These figures are sourced from Fiscal.AI).

To compare Q4 revenues: CoreWeave reported $1.6B, Nebius reported $228M, and Iren reported only $16M. So, CoreWeave currently has 100x more revenue than Iren. However, I believe that over the coming years, Iren will become the largest and most profitable neocloud company. With that, I also expect Iren’s market capitalization to eventually surpass both CoreWeave’s and Nebius’s.

Valuation just based on contracted power: (removed $5B from Nebius MCAP because of ownerships in Clickhouse, Avride etc)

Iren has a much lower valuation based on their contracted power, even though they also own the data centers themselves. My opinion is that if Iren manages to execute over the next few years, this gap will close, and Iren will become the largest neocloud company measured by Mcap, due to vertical integration (higher margins) + largest capacity.

Key Risks for Iren

There are numerous risks associated with investing in such a “hot” sector (data centers/AI). Here are the primary risks I believe could challenge Iren’s future:

• Access to Capital:

  For Iren to expand at the required pace in this sector, they are dependent on raising as much cheap capital as possible. While Iren has proven capable of this so far, the macroeconomic picture can change rapidly.

• Debt & CapEx:

  The total cost to build out all of Iren’s data centers and potentially purchase all the GPUs carries an enormous price tag of around $130B (assuming the same CapEx profile as the MSFT agreement). This is 10x their current market cap. While this expenditure will be spread over many years, it entails significant risk. It is therefore crucial to have trust in the management team to make the right decisions for the company’s long-term health.

• Declining Demand:

  While I am unsure if this is realistic in the coming years, it remains a possibility. For Iren, this would mean increased difficulty in negotiating favorable contracts, leading to margin compression and hits to the bottom line. It would also reduce the underlying value of their data centers and GPUs, thereby lowering the company’s overall valuation.

• Execution Failure:

  If Iren fails to deliver data centers to clients on time and at the expected quality, it would severely damage their reputation. This would hinder their ability to secure future clients and contracts, carrying highly negative consequences for the company.

• Energy Portfolio Concentration (ERCOT & SPP):

  Their power suppliers are primarily ERCOT in West Texas (Childress, SW1, SW2) and SPP (Oklahoma). If systemic issues were to arise in these grids, Iren’s heavy concentration in Texas could be a negative factor. However, I struggle to pinpoint exactly what this risk would materialize as, given that Iren holds firm contracts for this power.

• Shareholder Dilution:

  To scale further, dilution is to be expected. Management has been clear that they will try to avoid it as much as possible, using it only when strictly necessary. We have previously seen Iren dilute shares at last year’s low of $6, a move that was viewed very negatively by many investors. Iren did announce a new $6B ATM (at the market offering” 04.03.2026. There is no confirmation that this will be used in the near future, but it could mean dilution for shareholders down the line. At the current market value, it would correspond to a 40% dilution if fully utilized today (13B mcap).

Conclusion

To summarize, I am highly optimistic about this company, which is why it currently makes up over 50% of my personal portfolio. I believe the next 36 months will be incredibly strong for the company and its share price. Iren is sitting on scarce resources that everyone wants a piece of. If Iren manages to capitalize on this, the company could grow exceptionally large (into the hundreds of billions).

This analysis should not be construed as financial advice or a recommendation; it is for entertainment purposes only. My views on the company are valid only as of the time of writing and could change rapidly.

I have spent an enormous amount of time on this analysis, and I hope you, the reader, find some value in it. Remember to do your own due diligence and research on the company.

Iren’s share price at the completion of this analysis (March 6, 2026) is $38.5

My position:

I bought my first shares in Iren in May 2024 at a price of $6.87. The company was only a very small position for me then, but during the downturn last April (from $12 to $5), I had done extensive research and bought the stock with both hands. This made the company my largest position. Since then, I have continued to invest a massive amount of time into researching the company, and I have subsequently increased my average cost basis from $7 to $21—and I continue to buy the stock to this day.

Special Thanks:

I want to give a special thanks to Frans Bakker (@FransBakker9812)—I recommend subscribing to his X profile, as there is a wealth of information to be found there. I also want to thank Agrippa Investments (@Agrippa_Inv) for his Substack (highly recommended for Iren investors):

Agrippa Investments

Institutional-Grade Research & Analysis on Hyper-Growth Stocks.

By 𝐀𝐠𝐫𝐢𝐩𝐩𝐚

and Jim Lui (@jiahanjimliu) for his insights and financial model related to Iren: https://jiahanliu.github.io/IREN-Community/.
And also McNallieMoney for the amazing podcast content with the Iren team, check them out: https://www.youtube.com/@McNallieMoney

Microsoft, Together.AI, FluidStack og Fireworks AI, og har allerede klart å bli NVIDIA’s «preferred partner».

CEO, management og erfaring

Dan (CEO) har eid og styrt et selskap kalt Palisade Investment Partners, som er et forvaltet selskap som investerer i infrastruktur. Selskapet var spesialisert innenfor energi sektoren, som inkluderer vind, sol og gass. Selskapet vokste seg til $6b. Dan har også vært visepresident for Macquarie, som er en investment bank med mcap på $60b. Will hadde lignende posisjon i Macquarie som ansvarlig for råvarer og globale markeder.

Brødrene sitter med omtrent 4,2% av totale aksjer hver. Så insentivene er der for begge brødrene om å drive dette selskapet så lønnsomt og effektivt som mulig.

Noe som gjør at jeg kan ha en så stor allokering til dette selskapet er at jeg har stor tillit til management, og årsaken til dette er basert på deres resultater de siste årene, men også deres verdier og langsiktighet som jeg kommer inn på senere.

Tilliten jeg har til management bygger på deres verdier og langsiktighet. Formålet til Dan og Will er å skape langsiktige aksjonærverdier, og lar seg ikke bli påvirket av konkurrenter og kortsiktige strategier. 2024/2025 ble det veldig populært for bitcoin selskaper og andre mindre selskaper å kjøpe opp store mengder bitcoin, og gjerne utvanne aksjonærer for å kjøpe bitcoin. Etter suksessen til Microstrategy med Michael Saylor i spissen, begynte andre selskaper som Mara og Cleanspark, og generelt alle andre bitcoinselskaper å holde bitcoinsa de minet, men kjøpte også gjennom lån og utvanning av aksjonærer. Iren var det eneste selskapet som ikke gjorde dette, samtidig som aksjonærer og markedet la press på at de måtte gjøre dette. De hadde en langsiktig plan, og forholdt seg til dette, til fordel for aksjonærer på langsikt. Denne avgjørelsen har trolig spart aksjonærer mye penger i etterkant, der slike selskaper senere har fått juling. Anbefaler å se på dette klippet der Dan forklarer hvorfor de ikke HODL bitcoin.

https://www.linkedin.com/posts/iren_daniel-roberts-discusses-irens-non-hodl-activity-7179488519839465472-nAjG/

En annen grunn til at jeg har tillit til Dan er at han ikke annonserer pipeline (fremtidige datasentre uten godkjent connection agreement). Andre konkurrenter slenger ut med fremtidig pipelines som de ikke har godkjent tilgangen til gridden (strømnettverket). Er ikke noe poeng i at jeg skriver mye om dette, mener at videoen på 2 minutter forklarer budskapet uten videre kommentar.

Pipeline annonsering (se fra 34:50 til 36:50)

En annen årsak til at jeg har stor tillit til management er ved å se på resultatene de har levert hittil. Iren vokste fra 0,8 EH/s i januar 2022 til 50 EH/S sommeren 2025. De vokste raskere enn noen andre konkurrenter har klart to år på rad. I 2023 økte de sin hashrate med 250%, og knuste rekorden i 2024 med 402% vekst i 2024 (5,6 EH/s til 28,1 EH/s). Ikke nok med at de vokste mye raskere enn konkurrentene, de var også soleklart det mest lønnsomme bitcoin mining selskapet, og faktisk det eneste lønnsomme! Dan Roberts sa dette i et intervju: «Targets look bullish and ambitous, but as you know, we dont publish something unless were super confident we at least beat it» og synes det oppsummerer management bra. De skal levere, og lover ikke noe de ikke kan levere på.

Se bildene og figurene under for deres resultater innenfor Bitcoin mining:

Som en ser på figur 2, blir Iren sammenlignet med Cleanspark, som er etter min mening den nest beste bitcoin mineren basert på tallene de leverer.

Som en ser på tallene i figur 2 har Iren gross marginer til 74,5% sammenlignet med CLSK sine 56,7%. Etter alle kostnader er betalt hadde Iren en bitcoin kostpris på $66,964, samtidig som CLSK hadde $89,053 (dette var når bitcoin kursen var svært høy). Tallene snakker for seg selv, Iren har levert bedre enn noen andre, og håpet er at iren vil klare å overføre prestasjonene over til AI/HPC segmentet.

Figur 1 Q2 2025, Hentet fra Agrippa

Figur 2 April 2025 fra McNallie Money

Resterende management:

Iren har et svært kompetent team med mange tiår med erfaring innenfor energi, kapital og datasenter sektoren. Alle lederne innenfor Iren har over +20 års erfaring innenfor sitt felt. Den nyeste ansettelsen av Iren er John Gross (CIO, Chief innovation officer) som ble lansert 17 februar 2026. John har over 20 års erfaring innenfor datasenter, og er spesialisert innenfor «high density and liquid cooled infrastructure». Han sitter med kunnskap som er veldig viktig for Iren for å kunne skape datasentrene til de fremtidige GPU-ene som vil kreve høy rack density og kjøling.

Iren sitter også med en CTO (Chief technologi officer) som har over 20års erfaring innenfor cloud og datasenter sektoren. Denis Skrinnikoff har vært en del av teamet siden 2021 med en spesialisering innenfor virksomhetsarkitektur. «Denis has been involved in the design, construction and management of Tier 3 data centers, acquisitions and mergers, and managing business units, most recently as Senior Director, Cloud and Colocation at TeraGo and previously as Senior Director of Technology at RackForce Networks, Inc.”

Den rollen som jeg anser som den viktigste for Iren fremover er CFO (Chief financial officer). Der ansatte Iren Anthony Lewis 1 juli 2025. Årsaken til at jeg anser denne rollen til den viktigste, er at Iren må hente mye kapital fremover, og er avhengig av gode terms og renter på dette for å oppnå mest mulig aksjeverdier og bunnlinje. Resultatene og erfaringen til Anthony Lewis tjente Iren på med en gang. Innen få uker klarte Anthony å forhandle frem 100% belånte GPU-er til en rente under 10% til $100m. Dette betyr at kjøpene av GPU-ene ikke vil kreve cash upfront, og gjør det lettere for Iren å skalere uten bruk av kapital og utvanning av aksjonærer. Avtalen er strukturert på den måten at det er en låneperiode på 36 måneder, og så lenge Iren tjener mer enn renten, går de i pluss uten bruk av kapital. Iren så også viktigheten av Anthony Lewis når Iren skulle forhandle finansering til $5,8B av GPU-ene fra Microsoft avtalen Iren har med de (November 2025). Her hentet Iren $3,6B finansering til under 6% rente. For å sammenligne har Coreweave gjort det samme, og de gjorde det samme til 10% rente. Her er det snakk om store forskjeller som vil bety mye over tid. Årsaken til at Iren får så lave renter er gjeldsgraden selskapet har, som jeg kommer inn på senere.

Iren har også en imponerende incentivplan, der CEO-en blir betalt hvis aksjekursen når 370$, $650, $925$ og $1850. Så alle incentivene er der for at selskapet skal vokse seg til et gigantisk selskap.

Iren har altså over de siste årene bygd et svært kompetent team, som er klar for å bygge seg opp til det mest lønnsomme, effektive og fremtidsrettede datasenterselskapet.

Deres suksess i bitcoin, og hvorfor de endte opp med mining:

Will og Dan bestemte seg mot slutten av 2017/2018 at bitcoin var en 1-0 oppfinnelse, og ønsket å jobbe mot/med denne oppfinnelsen. Samtidig forstod brødrene at de hadde en enorm fordel innenfor sektoren med sin bakgrunn. Årsaken til det er fordi begge hadde eksepsjonell forståelse av energisektoren, og spesielt vind og sol energi.

Iren har siden starten kjørt en litt annen strategi enn konkurrenter. Iren fokuserer på å bygge opp datasentrene sine fra bunn. Altså at de kjøper opp landet, og bygger opp alt av infrastruktur selv. Dette gjør at selskapet er fullintegrert som gjør at de har 100% kontroll over egne anlegg. Konkurrenter innenfor bitcoin mining har kjørt en strategi gjennom å kjøpe opp mindre selskaper eller kjøre en fusjon (M&A). Iren har tatt med seg denne strategien inn til AI/HPC, noe som skiller seg ut ifra konkurrentene som leier anleggene, noe som spiser på marginene (Kommer mer innpå dette senere).

“Bitcoin mining was Step 1. Not because it was the end goal, but because it was the smartest first step. Unlike any other data center service, Bitcoin mining doesn’t require an external customer. You mine for yourself, for your shareholders. It let IREN build a track record of development execution, generate cash flow, and prove out its infrastructure capabilities at massive scale before layering in more complex, customer-facing services.”

Var det en pivot til AI, eller var dette planlagt lenge?

Iren har siden dag 1 vært tydelig og åpen for at de vil gjøre det som lønner seg for selskapet og aksjonærene. Iren har brukt bitcoin mining som en fantastisk mulighet til å skalere selskapet til størrelsen det har blitt til i dag, der de eier 4,5GW med kapasitet før AI omsetningen har begynt. De siste årene har det blitt veldig tydelig at det gir mer økonomisk mening å vokse seg stor innenfor AI enn bitcoin mining. Det gir bedre avkastning, samtidig som det gir aksjonærer mer stabilitet og bedre tilgang på kapital. Det siste året har nesten alle bitcoin mining selskaper gått over til å drive med AI/HPC ved siden av deres bitcoin mining. Dette gjør at det er veldig naturlig å tenke at Iren også har «pivota» inn i AI, men her tar mange helt feil. Iren har siden starten av selskapet snakket om AI/HPC og om hvordan dette over tid vil bli en naturlig overgang for selskapet. Iren har siden av 2023 «lekt» seg med AI og kjøpte 248 NVIDIA H100 august 2023. Dette har de siden leid ut til Poolside fra 2024.

Dette er kommentaren til Dan Roberts (CEO) ga etter pressemeldingen etter kjøpet av H100:
“Leveraging our next-generation data centers into generative AI is an exciting opportunity, particularly given current industry shortages in rack space and compute. We believe demand for sustainable computing is unlikely to go away, and feel we are uniquely positioned to capture ongoing growth in the broader industry; whether that be ASICs for Bitcoin mining, or GPUs for generative AI and beyond.”

Overgangen til AI/HPC har vært nøye planlagt, og deres bitcoin mining sentre i Canada (Canal Flats, British Columbia, og Manckenzie) har siden nylig drevet med kun Bitcoin mining. Datasentrene var også bygd for å kunne drive med AI/HPC, som gjorde at overgangen var veldig enkel, og ikke hadde noe nevneverdig ekstra kost, se video:

https://x.com/IREN_Ltd/status/1971576440392216991

Dan har gjennom mange podcaster forklart deres strategi, og har siden 2020 (eldste klippet jeg fant) snakket om hvordan AI/HPC er fremtiden, og hvordan de skal over tid over til det markedet når det gir mening økonomisk, som har visst seg å være nå. Anbefaler å se videoen på 2 minutter. (se fra 20:50 - 22:50)

(Klipp fra 2020)

Bitcoin var altså ikke det endelige målet, men var starten for å bygge et gigantisk portefølje med tilgang på kraft. Bitcoin gjorde det enkelt for Iren å bygge seg større uten et behov for en kunde (Miner bitcoin, ingen mellommann her). Samtidig visste bitcoin mining at Iren over tid klarer å bygge store datasentre og genere kontantstrøm. Bitcoin mining var altså det første steget før Iren begynte å bevege seg over til AI/HPC. Jeg tror personlig at 2026 blir det siste året Iren kommer til å mine bitcoin, og fra 2027 være 100% et neocloud selskap.

Veien videre til AI for Iren

Iren har som nevnt drevet/testet Ai/HPC i mange år ved siden av deres primære bitcoin mining del. Overgangen har vært tydelig fra management, og at dette vil skje når det gir bedre risk/reward for dem og aksjonærer. Den tiden har ankommet og Iren er på vei til å bli et 100% AI selskap som har lagt bitcoin delen bak seg. Hvis en går inn på deres nettsider, er det veldig vanskelig å finne ut at de driver med Bitcoin mining. Informasjonen og signalene som har blitt gitt av management kun for et år siden var at Colocation var den naturlige overgangen. Colocation (CoLo) betyr i praksis at Iren bygger datasentrene for kundene, også kjøper kundene GPU-ene og installerer disse i Iren sine datasentre. Dette er strategien Cipher Digital har gått for. Denne løsningen gir stabil og jevn cash flow, og mindre kapitalkrevende enn den retningen Iren har gått for det siste året.

Iren har valgt å gå for en strategi som heter infrastructure as a service (IAAS). Det betyr at de finansierer og bygger datasentrene, samtidig som de også kjøper og eier GPU-ene og resterende utstyr. Iren blir altså eier av alt utstyr, og leier dette ut til en stor kunde som Microsoft, Meta eller Google (Hyperscaler). Iren vil ikke tilby noe programvare eller lignende, men kun den fysiske infrastrukturen som et selskap skal trenge for å drive deres LLM modeller, eller lignende. Denne strategien er kapitalkrevende, og krever mye gjeldsfinansiering. Årsaken til at Iren velger å gå for IAAS framfor CoLo er at dette gir mye bedre payback periode. Payback perioden på CoLo sammenlignet med IAAS er nesten det dobbelte, men vil kreve mye mer arbeid og kapital av management. Høyere risiko, med høyere nedside. Men det er ingenting som stopper Iren fra å kjøre en kombinasjon av dette, men ifølge Iren må dette gi mening økonomisk for å skape mest aksjonærverdi. Avtalen med Microsoft har hittil bevist at denne modellen (IAAS) gir mening. Som jeg kommer innpå senere i analysen klarte Iren å hente $5,8B kapital til finansieringen av 76 000 Nvidia GB300 GPU-er til en rente på 3%. De fikk 20% prepayment av MSFT (1,94B), og resterende ($3,6B) til en rente under 6%. i praksis betyr dette at Iren kan halvere paybackperioden sin sammenlignet med Cipher Digital, til en rente på litt over 3%. Dette gir økonomisk mening, og er glad for at management går for denne retningen så lenge det gir økonomisk mening.

Iren tilbyr to ulike former for datasentre:

Air cooled: Dette er datasentre som benytter seg av kun luftkjøling. Dette er mye billigere å bygge ut enn liquid cooled datasentre, men det er ikke mulig å kjøre de fremtidige GPU-ene til Nvidia på slike systemer. Modeller som Vera Rubin 200 (VR200) krever rack density oppimot 200kW. Rack density er strømtettheten per rack (stativ), altså hvor mye strøm hver enkel rack forbruker/ har behov for. Iren har klart å kjøre noen av de nyeste modellene til Nvidia (B200/B300) på deres air cooled datasentre i Canada, som i seg selv er svært imponerende (viser til dette senere). Men som nevnt over vil de nyeste modellene kreve en rack density på 200 eller mer. Air cooled datasentre ligger grensen på 80kW rack density. Derfor vil de fremtidige datasentrene som blir bygd for å håndtere disse laget med liquid cooling.

Liquid cooled: Er datasentre som benytter seg av væskekjøling. Dette er for å svare de store kravene fra de nyeste og fremtidige modellene til Nvidia og andre GPU/TPU aktører. Væskekjøling gjør at kostnaden til datasentre øker drastisk, sammenlignet med air cooled datasentre. Iren har tidligere bygd air cooled datasentre til 700K/MW, og med deres nyeste kontrakt med Microsoft blir kostnaden 14-16M/MW!

De fremtidige datasentrene vil være hovedsakelig liquid cooling, men Iren sa på deres nyeste earnings call (Q2 2026, Q4 2025 for vanlige børsselskaper) at de hadde fått mer etterspørsel etter air cooled datasentre, fordi dette gir lavere time to power. Altså gir mer compute raskere, fordi behovet er NÅ for disse selskapene.

Iren sitter med kanskje et av de største secured MW porteføljene i verden med 4,5GW annonsert foreløpig. Iren annonserer ingen sites, før de er 100% sikre på at de har kontrakter på tilgang til strømnettet. Dette betyr at de har 4,5GW med datasentre med fremtidig tilgang til strømnett som Ercot og SPP.

Det er viktig å skille mellom de ulike typene MW. Iren annonserte på deres forrige kvartal et nytt gigasenter på 1 600MW, dette er secured MW. Altså de har sikret datasenteret med kraft/strøm i fremtiden. Når de får tilgang på strømnettet kalles dette energized MW. Iren har jo deres enorme 1 400MW som blir energized i april 2026. Iren har i dag deres tre datasentre i Canada som er Monetized. Altså datasentre som har vært secured power, til å bli energized og deretter blir monetized MW. Datasentre som generer penger.

I dag består Irens portefølje hovedsakelig av energized og secured power. Men som de fleste har fått med seg er etterspørselen etter datasentre stort, og capex (kapitalutgifter) blant hyperscalers øker enormt.

Bare i Q4 2025 økte hyperscalers CAPEX knyttet til AI/datasentre for 2026 til enorme summer:

Amazon økte deres til $200B fra $131B for 2025

Alphabet (Google) økte til $175-185B fra $91B for 2025

Meta økte til $115-135B fra $72B for 2025

Microsoft økte til $100-150B fra $116B for 2025

Dette betyr at disse selskapene alene har planlagt kostnader knyttet til AI/datasentre på ca $600B for året 2026.

Deres energiportefølje (4,5GW)

Iren er trolig det selskapet på børs som sitter med mest secured power i dag til 4,5GW. Dagens estimater fra Cargoson estimerer at total kapasitet som brukes til datasentre er i dag på 122GW. Målt i dagens kapasitet sitter Iren alene med 3,7% av verdens datasenter kapasitet målt etter dagens kapasitet. Iren’s største konkurrent Coreweave sitter med contracted power på 3,1GW, og har som mål å ha 5GW med AI infrastruktur innen 2030. Iren’s andre konkurrent er Nebius som har økt deres kapasitet til 3GW, men det som skiller Iren fra disse to konkurrentene er at Iren eier datasentrene selv, og ikke leier fra en 3-part som Coreweave og Nebius gjør.

Iren har i dag en AI kapasitet på kun 460MW, og omsetningen har ikke truffet inn på deres kvartalsrapporter enda, men vil begynne å øke drastisk fra deres Q3 2026 rapport (tilsvarer Q1 2026 for resten av markedet). I Q2 2026 (Q4 2025 for resten av markedet) hadde Iren kun $16M i omsetning fra AI.

De 460MW aktive kapasiteten kommer fra hovedsakelig Canada (160MW), og Childress (300MW). Childress kapasiteten er kun 300 av 750MW kapasitet kontraktet bort. 300MW er i dag i kontrakt med Microsoft til en avtale på 9,7B over 5 år, der Iren skal kjøre 76K GPUer (GB 300). Disse 4 datasentrene på 460MW (10% av total kapasitet) vil gi Iren 3,4B i ARR (Annual recurring revenue) alene.

Med at kun 10% av deres datasenter portefølje generer $3,4B i ARR, kan potensielt Iren generere $35B i ARR fra 2030 med deres totale kapasitet på 4,5GW hvis en antar like forutsetninger (har all grunn til å tro at unit economics for avtalene vil bli bedre over tid, diskuteres senere).

De ulike datasentrene og hva som er ulikt:

Iren sitter i dag med 6 ulike lokasjoner på deres datasentre der det er totalt 7 ulike datasentre. Årsaken til at det er 7 datasentre på 6 lokasjoner er at Sweetwater 1 vil være på 1,4GW og vil være klart fra april 2026, imens Sweetwater 2 vil være på 600MW og være klart fra slutten av 2027.

Dagens annonserte datasenter portefølje er kun lokalisert i Canada og USA. Iren har flere små datasentre i Canada, og noen store i USA. Det er noen ulikheter på disse datasentrene og vil gå for meg hver og en, og hvorfor de er unike. Jeg kommer til å ta det i den rekkefølgen som Iren vil ha kapasitet på, og starter med de Iren allerede har kapasitet på i dag, som er Canada, der de har annonsert 40 000 GPU-er for 2026.

Det minste datasentre som Iren har idag er Canal Flats I Canada som er på kun 30MW. De andre datasentrene i Canada er Prince George på 50MW, og Mackenzie på 80MW. Det som skiller datasentrene i Canada fra de Iren har i USA er deres PUE. PUE står for Power Usage Effectiviness, og er et mål som brukes til å forklare hvor energieffektivt et datasenter er. PUE forklarer hvor mye av strømmen som brukes til datautstyret/servere (IT), kontra hva som brukes til nedkjøling og andre ting. I Canada har Iren en PUE på 1,1 som er kanskje bransjens beste. Det betyr i praksis at kun 10% av deres strømkostnader går til kjøling. Så ekstrakostnaden er på kun 10%. For å sammenligne har Iren sin kontrakt med Microsoft annonsert en PUE på 1,5. Dette betyr at i praksis er marginene til Iren sine anlegg i Canada mye høyere. (ref bildet under)

Årsaken til de store forskjellene kommer av klimaet og temperaturforskjellene i USA og Canada. Canada er mye kjøligere, som betyr at det trengs mindre strøm til kjøling og vedlikehold av GPU-ene.

Datasentrene i Canada driver også i dag foreløpig kun med AI Cloud, som betyr at kunder leier GPU-ene og tjenesten fra Iren direkte, gjerne over en periode på måneder eller lenger. I motsetning til kontrakten med Microsoft som er en IAAS (infrastructure as a service), der MSFT kjøper kapasiteten over 5 år.

Iren tilbyr disse GPU-ene i dag:

Iren har i dag annonsert rundt 40K GPU-er til Canada i løpet av 2026. Denne kapasiteten innebærer hovedsakelig B200 og B300 modeller som er Air cooled. Det er svært få konkurrenter som klarer å kjøre disse modellene ved air cooling, og er noe som gjør at Iren skiller seg ut ifra konkurrentene.

Iren har klart å bygge AI datasentre til 700k/MW, oppnådd 80kW rack density som kan kjøre noen av de nyeste modellene til Nvidia med Air cooling istedenfor Liquid cooling, med en PUE på under 1,1 (Dan’s egne ord). Dette ble mye tall og informasjon i en setning, men dette er ekstremt imponerende. For å sammenligne bygger Iren nå med Microsoft avtalen til en kostnad på 14-16M per MW. Forskjellen på $700K og $14-16M er ekstremt. Forskjellen er at Iren sin Microsoft avtale har en rack density på 200 eller mer, men også at datasentrene håndterer de nyeste GPU modellene som VR200, som krever oppiot/over 200kW rack density. Jeg må faktisk presisere hvor imponerende det er at Iren klarer å oppnå 80kw/rack density i air cooled datasentre. Det ble sett på som uoppnåelig, men Iren har klart det i deres Price George anlegg. Fordelen med å klare det med kun aircooled istedenfor liquid cooled, er at det er mye billigere datasentre og bygge, og det går raskere å bygge også. Spurte Gemini pro om Hvor «mye er maksimal rack density et air cooled datasenter har oppnådd?». Svaret jeg fikk er 40 kW rack density. Eneste andre eksempelet jeg fant av andre aktører som har klart over 40kW rack density er Intel med 43kW rack density.

På Irens egne nettsteder står det at deres max rack density er 70 kW på alle deres datasentre i Canada, man Dan har selv nevnt at de har klart 80kW i Prince George.

Men 70kW rack density på alle deres datasentre i Canada er i seg selv en stor prestasjon med PUE gjennomsnitt på 1,1!

Canada er altså en pengemaskin for Iren, med skyhøye marginer og lav capex/opex.

Som bildet av Prince George viser til over, bygger Iren et Liquid cooled data senter. Iren sier selv at dette ikke er en ekspansjon av Prince George, men jeg og flere har våre tvil til det. Jeg tror dette kan ha med den videre ekspansjonen av PG som jeg nevner under.

De ulike datasentrene til Iren i Canada og med deres economics:

• Canal Flats (Canada):

Størrelse: 30MW

Operasjonell: Ja

PUE (Power Usage Effectiviness): 1,1

ARR: $272,73M

EBIDTA marginer: 85%

Payback periode: 3,6 år

Resultat før skatt (Årlig): $99,12M

• Mackenzie (Canada):

Operasjonell: Ja

Størrelse: 80MW (Skal oppgraderes til 100MW)

PUE (Power Usage Effectiviness): 1,1

ARR: $727,27M

EBIDTA marginer: 85%

Payback periode: 3,6 år

Resultat før skatt: $232,5M

• Prince George (Canada):

Størrelse: 50MW (Skal oppgraderes til 80MW).

Operasjonell: Ja

PUE (Power Usage Effectiviness): 1,1

ARR: $500M

EBIDTA marginer: 85%

Payback periode: 2,7 år

Resultat før skatt: $232,5M

Hvis du er observant så ser du kanskje at jeg har skrevet at Mackenzie og Prince George skal oppgraderes fra 80 til 100MW og 50 til 80MW. Dette er informasjon en skal lete lenge etter for å finne. Dette er offentlig informasjon, men er veldig vanskelig å finne = ikke priset inn etter min mening. Denne informasjonen fant jeg i dette intervjuet med Glen Harrison (VP of operations in Canada) i en podcast med 280 abonnementer og 480 avspillinger. (se fra 9:25 til 9:43)

Klippet er kun 10 sekunder, der Harrison forklarer at de skal ekspandere disse datasentrene, noe som vil bety mye for Iren, med tanke på omsetningen, og marginene som kommer fra Canada. Prince George alene gir Iren en ARR på $500M med kun 50MW, noe som gjør at en kan forvente nærmere 800M i ARR fra Prince George innen kort tid (6-18 måneder).

Mackenzie skal også oppgraderes fra 80 til 100MW noe som vil øke ARR fra $730M til 910M i ARR. Dette betyr at Iren alene fra Canada har en potensiell ARR på $1,98B. Dagens guidance fra Canada ligger på $1,43B. Altså en økning på over $500M i ARR. Bildet under ser du et anbud som Iren har angående levering av stål til «Mackenzie 100MW» som markert i teksten. Som det også står er delievery or completion 1 april 2026. Dette kan vel tyde på at 100MW vil bli operativ om ikke altfor lenge.

For å oppsummere datasentrene i Canada:

· Lav PUE = lavere kostnad = høyere marginer.

· Kjører AI Cloud på disse sentrene

· Skal hoste 40 000 GPU-er (det som er annonsert)

· 3 sentre på 30MW, 50MW og 80MW der 50 skal bli 80MW, og 80 skal bli 100MW (potensielt).

· Kun aircooled datasentre, så kan ikke kjøre de nyeste GPU modellene som krever over 130kW rack density. Blir bygd liquid cooled datasenter i Prince George nå.

· Iren har oppnådd 80kW rack density i Canada, bransjens beste, noe som gjør at de kan kjøre B200/B300 kun ved bruk av air cooled datasentre.

Datasentrene i USA

Da er det over til de store datasentrene som Iren har i USA, eller mer presist Texas. Texas er kjent for å være en stat med mye mindre reguleringer, og datasentre tas imot med åpne armer. Datasentrene til Iren i Texas er plassert langt borte fra befolkningen, så ingen lyd eller støy plager noen. Dette er veldig positivt, men det negative med dette er at det er en lengre kjøretur for arbeiderne som jobber for Iren. Den nyeste oppdateringen til Iren var et det var 2000 arbeidere på Childress, som arbeidet for å gjøre klar datasentrene til Microsoft. (bilde under viser 8 kraner på Childress anlegget til Iren, bildet hentet fra Frans Bakker sin X profil)

Iren sitter med 3 store områder som de har fått godkjent kapasitet på. Disse datasentrene har kapasitet på 750MW, 2000MW og 1600MW. Som totalt bringer en kapasitet på 4 350MW i Texas. Ingen av disse datasentrene står klare i dag, og kun 300MW av 4 350MW er «solgt»/leid ut til kunder. Både Childress og Sweetwater 1 og 2 er tilkoblet Ercot sitt strømnett. Altså den operatøren som styret strømnettet i Texas.

Childress er det største anlegget Iren har i dag, og er 750MW som allerede er energized. Childress er det datasentre som Iren jobber hardest på om dagen, etter at de signerte en kontrakt med Microsoft i November 2025 på 300MW (200MW critical IT load, pga 1,5 i PUE). Denne kontrakten var på 9,7B over 5 år, der MSFT leier datasentre+GPU-ene av Iren.

Dette gir Iren 1,94B i årlig omsetning. Dette datasentre vil bestå av 4 ulike områder med 75MW hver. Derfor disse ulike kalles Horizon 1-4. Iren er på tidspunktet jeg skriver dette (02.03.2026) godt i gang med Horizon 1, som trolig vil begynne å genere penger i mai 2026. Horizon 4 antas å være ferdig midten av H2. Jeg skal gå i dybden på denne kontrakten senere i analysen.

Iren har resterende 450MW fra Childress sin 750MW kapasitet tilgjengelig idag. Denne kapasiteten brukes nå til Bitcoin mining, men ryktene sier at dette er på vei til å byttes om til AI/HPC. Noen glupe personer har trolig funnet Iren sin bitcoin adresse, der en så at deres bitcoin mining andel hadde falt fra 50 EH/s til omtrent 35 EH/s, noe som tydet på at det blir gjort store endringer, men dette kan på ingen måte bekreftes enda. Her er i alle fall bitcoin adressen til de som skulle være interessert:
https://mempool.space/address/3EeHsUEnyYQkjdV7zT2v4KHid6B16RjkHn

Jeg forventer at det vil komme en annonsering med en kontrakt der Iren har leid ut dette i løpet av 2026. Der byggingen vil startet i løpet av 2027, altså etter at de er ferdig med Horizon 1-4. Dette anlegget kan altså mulig bli en avtale på $15B over 5 år (50% større enn Horizon 1-4), med en antagelse om lik avtale som med Microsoft.

Dette er tallene fra MSFT avtalen, og antagelsene mine for Horizon 5-10 (resterende 450MW på Childress):

Horizon 1-4 (West Texas)

Størrelse: 300MW

Operasjonell: Snart (Kontrakt med $MSFT)

Konstruksjon: (ferdig rundt oktober 2026 med H4)

PUE (Power Usage Effectiviness): 1,5

ARR: $1,94B

EBIDTA marginer: 85%

Payback periode: 5,7 år

Resultat før skatt: $389,5M

Horizon 5-10 (West Texas)

Størrelse: 450MW

Operasjonell: Ja, men ikke til AI/HPC (bitcoin pdd)

Konstruksjon/kontrakt: H1 2027???

PUE (Power Usage Effectiviness): 1,5

ARR: $4,05B

EBIDTA marginer: 85%

Payback periode: 4,4 år

Resultat før skatt: $1313,02M

Sweetwater 1 i West Texas

Hittil har vi altså godt gjennom alle de datasentrene som Iren er operasjonelle på eller under konstruksjon. Disse datasentrene (Canada + Childress) står altså for 910MW.

Disse 960 (+50MW oppgradering i Canada) MW alene kan genere 7,97B i ARR ($1,94b fra Horizon 1-4, $4,05B fra Horizon 5-10 (VR200?) og $1,98B fra Canada.)

Iren har også godkjent kapasitet på 3 600MW fra deres gigantiske anlegg i Sweetwater 1 (SW1) på 1 400MW, Sweetwater 2 (SW2) på 600MW, og sist men ikke minst deres nyeste annonsering fra Q2 2026 rapporten på 1 600MW i Oklahoma.

Sweetwater blir sett på som Iren’s signatur. Det er på dette datasentre Iren skal bevise hva de kan få til, og 1 400MW tiltrekker seg mange aktører som ønsker seg deler eller hele anlegget. Sweetwater 1 (1400MW) ble annonsert av Iren November 2023 i deres monthly update. SW1 skal bli energized april 2025, og vil trolig bli en av verdens største operasjonelle datasentre. Byggingen av SW1 har begynt smålig med terreng-behandling, for å gjøre klart til en fremtidig kunde og kontrakt?

Under ser du et satellittbilde av Sweetwater 1 som ble tatt 15 februar 2026. På bildet er det identifisert 3 primary stations (hoved transformatorstasjon = hjertet av strømforsyningen), som kan forventes til å være 750MW (450MW gross IT load) som vil være tilgjengelig på 36,5kV iløpet av siste del av H1 2026. Utbyggingen av dette anlegget vil kreve trolig mer enn 2500 arbeidere kontinuerlig over en periode på hvert fall 12-18 måneder. Å bygge ut potensielt 1 400MW vil ta lang tid. Iren vil bruke omtrent 9 måneder på Horizon 1-4 (300MW), så SW1 vil kreve flere arbeidere, og vil ta oppimot 2 år å bygge helt ut, men det vil gjøres ferdig i bolker på samme måte som ved MSFT avtalen. I MSFT avtalen blir de første 75MW klare trolig i mai, og de siste 75 av 300MW blir ferdig i Oktober??

Så utbyggelse perioden til SW1 er uklart, men når det står ferdig blir det et av verdens største datasentre.

Det som blir det store spørsmålet investorer har knyttet til Sweetwater 1 og fremtidige datasentre er hva slags strategi Iren kommer til å velge. Kommer de til å gå for IAAS på hele 1 400 MW eller blir det Colocation for hele SW1, eller potensielt en kombinasjon av disse to. Mulighetene er mange, men signalet Dan har delt de siste kvartalene er at IAAS gir en mye bedre payback periode enn colocation, selv om det vil kreve mye mer kapital og potensiell utvanning enn colocation vil gjøre. Iren sier i deres Q2 2026 earnings call:

“When you’ve got scarcity around how many data centers you can physically bring online, every incremental 200 MW can deliver either $300 million-ish through a colocation or multiples of that in the billions under a cloud contract. So when we look at the monetization opportunity for our platform and growth for shareholders in creating value, the cloud opportunity creates a lot more upside, as we see it. And in terms of colocation versus cloud, we believe that AI is going to continue. We believe that data center demand is going to continue to compound, so recovering capital back in short order and being able to compound those returns, as distinct from holding effectively a bond exposure against a hyperscaler, is the area we want to play. And if investors want bond-like exposures, they can buy colocation companies, they can buy bonds in the hyperscalers.”

“But we believe we’re offering a high conviction, but very well-managed risk exposure to the sector through this AI cloud business. Now, again, we’re not dogmatic. Things can change quickly. We get a compelling colocation deal, we will absolutely pursue it, but right now, AI cloud looks very compelling for all those reasons.»

Årsaken til at Dan ser mye bedre risk/reward ved å finansiere og kjøpe GPU-ene selv er gjennom hvordan deres kontrakt med MSFT gikk. De fikk 20% av 9,7B i en prepayment (1,94B) av MSFT, og hentet den resterende kapitalen ($3,6B) til under 6% rente, som betyr at gjennomsnittskosten Iren må betale i renter for å eie GPU-ene selv er på litt over 3%, som Dan selv sier i Q2 2026:
«we’ve now secured 95% of the cost of the GPUs at an average interest rate of around 3% when you factor into the prepayment. So we essentially got the GPUs for next to nothing. So I think in terms of capital intensity, it ticks that box.”

Hvis Iren kan fortsette med denne strategien, vil det gi mye høyere marginer og bedre avkastning for investorer, istedenfor å bruke anleggene til colocation.

Jobbutlysninger for Iren på Sweetwater 1:

Det er foreløpig 10 jobb utlysninger fra Iren angående Sweetwater anlegget. Det kan ofte være en indikator på hvilken retning selskapet skal. De nyeste jobbene ble utlyst 02.03.26 og dette er jobbene:

The Procurement Manager will lead procurement activities for a new data center site, with an initial focus on supporting capital construction and infrastructure projects. This role will play a critical part in delivering large-scale mechanical, electrical, and critical infrastructure packages on schedule and within budget.

The Contracts Coordinator supports the full lifecycle management of capital and operational contracts across data center construction and live site operations. This role plays a critical coordination function between Procurement, Legal, Finance, Project Management, and external contractors to ensure contracts are properly drafted, negotiated, executed, tracked, and archived in accordance with company policies.

Opp til tolkning hva som legges i dette, men interessant er det uavhengig. Jeg forventer jo en avtale vil komme før energization (April 2026).

De aktørene jeg tror vil være interessert i SW1 er:
Google, Amazon, Meta, OpenAI, Anthropic, xAI, Microsoft? (Tviler siden de har Childress allerede). Jeg personlig håper på og tror det vil bli Google.

Dette er tallene jeg forventer/antar Iren kan få for SW1 med en IAAS avtale:

Sweetwater 1 (West Texas)

Størrelse: 1 400MW

Operasjonell: Nei (Energized april 2026)

Konstruksjon/kontrakt: Gjennom hele 2026, 2027 og sikkert 2028 også.

PUE (Power Usage Effectiviness): 1,5

ARR: $16,1B

EBIDTA marginer: 85%

Payback periode: 3,5 år

Resultat før skatt: $7 014M/yr

Utregningene som er gjort for SW1, og tallene jeg har kommet frem til for de som er nysgjerrig: (har brukt @jiahanjimliu sin modell som kan brukes her https://jiahanliu.github.io/IREN-Community/

Sweetwater 2:

Iren annonserte ytterligere 600MW i Sweetwater, som ble deres SW2 anlegg 17 mars 2025. Dette anlegget er på 600MW (+500 acres) og vil bli energized i slutten av 2027, som betyr at datasentre ikke vil være 100% operasjonelt før mot slutten av 2028, men vil bli gradvis bygget ut på samme måte som Horizon 1-4, trolig også i 75MW bulker. Sweetwater 2 ligger 28 miles unna SW1, og disse datasentrene vil kobles gjennom direkte fiber.

Oklahoma:

Oklahoma er deres nyeste annonserte datasenter på 1 600MW. Dette datasenteret ble annonsert i Q2 2026 rapporten til Iren (5 februar 2026). Dette overgikk SW1 datasentre på 1 400MW, og vil bli Irens største når det blir operasjonelt mot 2030. datasenteret skal bli energized iløpet av 2028, så det overlapper godt med deres SW1 anlegg, som trolig vil være ferdig og operasjonell innen den tid. Dette senteret bruker SPP som grid exposure, sammenlignet med de andre sentrene (SW1 og 2, og childress) i Texas som bruker ERCOT. Eneste informasjonen jeg har om dette potensielle datasenteret er fra et møte i Pittsburgh for å diskutere en skattelette for Iren i OK. Tilstedet på møte var Giles Walsh (VP of operations), David Shaw (Head of data center operations), og Jason Date (Commercial team). Iren presenterte deres estimater for inntjening og omsetning, og disse tallene ville tilsvare 25% av fylkets revenue. Diskusjonene gikk også over at OK skulle bli lansert i 3 faser der første fase er på 600MW. Møte ble sagt til å være positivt, og at det ikke oppstod noen konflikter mellom partene. Neste møte vil være 6 april 2026.

Fremtidig energiportefølje (Multi GW)

Jeg skal ikke spekulere I hvor mye uannonserte steder Iren vil få kraft fra, men som Iren selv skriver etter deres lansering av OK (1 600MW) er denne fortsatt på «multi GW development pipeline», noe som gjør at en kan anta at det vil komme videre annonsering av datasentre.

I deres Q2 2026 earnigns call sier også Kent Draper dette angående ERCOT sin batch Zero møte:

“In addition to that, yeah, there are other projects in our portfolio that are potentially also included in Batch Zero.”

Dette tolker jeg som at det kan hende Iren har flere datasentre i Texas som ikke har blitt annonsert enda, som kan gjøre at Iren potensielt har flere MW som vil bli energized innen kort tid (1-6 måneder).

Flere store investorer i Iren, som Agrippa tror at fremtidig uannonsert portefølje ligger på over 3,5GW.

En annen ting, er at jeg tror at Iren vil potensielt begynne å bygge datasentre i andre deler enn Canada og USA. Dan Roberts la ut dette i dag tidlig da jeg sto opp (03.03.2026):

Han sier her «Building AI cloud around the world today”, noes om tyder på at det er snakk om flere datasentre på flere kontinenter. Det er noen rykter om potensielle datasentre i Australia, Japan, Signapore/Indonesia/Malaysia og mulig Europa. Dette er bare rykter, og ikke noe mer, men forventer at det vil komme nye datasentre utenfor USA og Canada fra Iren i fremtiden.

Dybde analyse av kontrakten med Microsoft $9,7B (det positive og negative)

I November 2025 lanserte Iren sin første IAAS kontrakt med Microsoft til en pris på $9,7B over en periode på 5 år. Dette betyr i praksis at Iren skal bygge opp datasentrene og alt av utstyr og GPU-er. Så Iren blir sittende som eier av alle komponenter og utstyr, og Microsoft leier alt dette i en periode på 5 år av Iren. Avtalen er på 300MW med 200MW i IT load.

Dette er spesifikasjonene i kontrakten, og det finansielle:

Samlet verdi på 9,7B (1,94B i ARR) (5 år)

EBIDTA marginer på 85%

Prepayment fra Microsoft på 20% ($1,94B) (Veldig lukrativt for Iren)

GPU payment på $5,8B (inkluderer servere, InfiniBand, cabling osv) gjennom avtale med Dell (76 000 Nvidia GB300, potensielt oppgradering til VR200?)

Kostnader (per MW):

$9-11M knyttet til datasenter infrastrukturen.

$3M til Utrulling av 100 MW superklyngearkitektur og fleksible rack-tettheter (130–200 kW) (som fremtidig kan kjøre Nvidia’s VR200.

$2M for å fremskynde Horizon 1-4 datasentrene (ekstra kost for å øke tempoet).

14-16M i kostnad per MW. (tidligere guidance fra Iren var på $6-7M/MW før de inngikk kontrakt med MSFT, se bilde)

Dette betyr en total kostnad på Horizon 1-4 til $3B, uten kostnaden knyttet til GPU-ene på $5,8B.

Så for å oppsummere vil kostnadene totalt bli $8,8B for å bygge ut datasenter infrastrukturen, og for å kjøpe GPU-ene.

Dette på papiret ser ikke ut som en god kontrakt der Iren får $9,7B, til en kostnad på $8,8B, men det er viktig å huske på at Iren vil sitte igjen med et ferdig utbygd datasenter med levetid på 20-25 år. Og de vil også sitte igjen med den potensielle restverdien i GPU-ene (10-20%?) som vil tilsi $580- $ 1160M i restverdi.

Tillit

Noe av det jeg verdsetter høyest med denne kontrakten med MSFT er at Iren har fått tillit, og muligheten til å vise hva de kan få til. MSFT kontrakten er kun på 300MW av 4 500MW, altså 6,67% av deres godkjente grid connected datasentre. Dette er etter min mening den kontrakten som vil vise om Iren klarer å levere et produkt til Hyperscalers (Google, MSFT, Amazon osv) på tidsfrister og kvalitet. Jeg mener at Iren måtte ta litt dårligere economics på den første avtalen for å bevise at de kan prestere. Å gjøre dette med 6,7% av deres portefølje er respektabelt. Ikke nok med det, men at MSFT velger å samarbeide med Iren, legitimerer selskapet stort. Som management selv sier angående kontrakten med MSFT:

“This deal matters for reasons beyond the economics. Microsoft does not sign nine-figure, multi-year contracts with companies it doesn’t trust. This is a de facto institutional endorsement of IREN’s development capabilities, infrastructure quality, and operational reliability. It’s the kind of validation that cannot be manufactured.”

Og

“A counterparty like Microsoft, they don’t enter into these deals lightly. It was many months of detailed due diligence. And the due diligence that they did was on every aspect around our business. The businsess itself, the management team, the datacenter design, the site management, the delivery capabilities that we have, our ability to stand up large scale GPU clusters and operate them. It gives us an air of credibility now, and even since the deal was announced we had a large amount of inbound inquiries. And the other discussions we were already having with other hyperscalers, you know have significantly ramped up.”

Se fra (8:21 til 10:44)

Som Kent Draper (CCO) selv forklarer, legitimerer dette selskapet stort, og har gjort det mye enklere å forhandle frem andre avtaler. Hvis de klarer å levere til MSFT er det et bevis til andre kunder at det er «trygt» å stole på et tidligere bitcoin mining selskap til å levere AI/HPC datasentre i det ypperste laget.

Matematikken bak avtalen:

Jeg mener at man må dele avtalen inn i to deler for å forstå avkastningen for selskapet bak avtalen, fordi det gir ikke mening å prise denne avtalen basert på en 5års periode, når datasenteret/infrastrukturen har en levetid på minst 20 år.

Utregningene er gjort av Agrippa Investment, og jeg anbefaler virkelig å lese/ abborene på hans substrack (helt nederst):

Årsaken til at jeg ikke har gjort det selv er fordi:

· Agrippa har allerede gjort det svært grundig og oversiktlig

· Har ikke kompetansen, eller tiden personlig til å regne på alle disse ulike måtene.

Utregningen av GPU-ene: (Ved hjelp av AI)

Ubelånt avkastning (Unlevered IRR)

• 10–12 % (Lav tosifret) dersom man antar at GPU-ene har $0 i restverdi etter 5 år.

• 17–20 % (Høy tosifret) dersom man antar at GPU-ene har 20 % restverdi (6,25 års avskrivning).

Belånt avkastning (Levered IRR)

• 25–30 % dersom de sikrer 2,5 mrd. dollar i gjeldsfinansiering (uten restverdi).

• 35–40 % dersom gjeldsfinansieringen økes til 3 mrd. dollar (uten restverdi).

• 35–50 % dersom man i tillegg forutsetter 20 % restverdi på utstyret, avhengig av om lånet er på 2,5 eller 3 mrd. dollar.

Utregningen av datasenter infrastrukturen:

• Intern leie: Selskapet regner en intern husleie på 130 dollar per kW per måned (pluss en årlig prisjustering på 3 %). Dette gir omtrent 1,56 millioner dollar i leie per megawatt (MW) det første året.

• Driftskostnader (OPEX): Allokeres med ca. 291 000 dollar per MW til colocation-delen (resterende legges på sky-delen).

• Finansiering: Agrippa legger til grunn en 50/50 splitt mellom egenkapital og gjeld for datasenter-CAPEXen, med et 10-års lån til 7,5 % rente.

• Inntjening per MW: I starten genererer dette en gjennomsnittlig profitt på omtrent 500 000 dollar per IT MW.

(se fra 15:33 til 19:43)

(Kent Draper (CCO) på Podcast) (3-4 minutter totalt)

Irens egen guidance på MSFT avtalen:

Unlevered IRR på “low double digits” (11-13%?)

Antar 20% restverdi på databrikkene (6 års levetid på GPU-ene) Unlevered IRR på “high teens” (17-20%?)

Levered IRR 25-30% på null restverdi av GPU-ene med 2,5B i finansiering.

35-40% hvis en antar $3B i finansiering.

Med en antagelse om 20% restverdi av GPU-ene blir Levered IRR på nærmere 50%.

Som vi vet fra deres Q2 2026, så hentet Iren 3,6B i finansering, til en lavere rente en de fleste antok (under 6%).

Som Kent Draper sier kan avtalen forsvares gjennom avkastningen Iren får på kontrakten, men hvis en legger til tillits aspektet er denne avtalen i seg selv årsaken til overgangen fra et Bitcoin mining selskap til et avansert og full utviklet neocloud provider, og prisingen av selskapet følger etter. Årsaken til det er høyere prising av et AI selskap enn et Bitcoin mining selskap, samtidig som AI/HPC gir bedre avkastning og stabilitet for selskapet.

VR 200 reforhandling i MSFT kontrakten

Det som ikke kommer frem i deres annonsering av Microsoft avtalen er det man kan finne i deres nyeste 10Q knyttet til deres Q2 2026 rapport. Der står det at kontrakten med Microsoft inkluderer en reforhandling av avtalen til å endre GPU-ene fra GB300 til de nyeste modellene til NVIDIA som blir lansert i starten av H2 2026. Disse modellene heter Vera Rubin (VR200). Dette betyr at datasentrene som Iren har bygd for Microsoft er bygd for å håndtere de modellene som allerede ikke er i produksjon enda. Et bevis på at Iren tenker langsiktig, og at Irens datasentre er planlagt og laget for fremtiden, og ikke kun nå (tier 3 datasentre). Kontrakten sier at disse forhandlingene kan forekomme hvis Microsoft sier ifra 4 måneder i forveien, og selskapene har 30 dager på å diskutere nye «terms» knyttet til de nyeste modellene.

Dette tror jeg vil egne begge parter, og Iren vil tjene mest på dette, grunnet den nyere økningen av leieprisen på de gamle Nvidia modellene. Dette tror jeg vil være relevant for Horizon 3 og 4, som ikke vil være ferdig før mot september/oktober. Så jeg ser en sannsynlighet for at GB300 blir kjørt på Horizon 1 og 2, også blir VR200 kjørt på Horizon 3 og 4 etter forhandlinger med MSFT. Uansett er dette en god nyhet som viser til kvaliteten og evnen Iren har på å bygge datasentre med svært krevende tekniske detaljer for å løfte rack density til nærmere eller høyere 200kW.

Bildet over viser til leieprisene knyttet til de eldre NVIDIA modellene. Det vi ser er at etterspørselen etter disse GPU-ene har økt drastisk, og leieprisene har økt nesten 25% i 2026 alene. Disse brikkene er 3 år gamle, og etterspørselen er stor fortsatt, selv om kvaliteten/kapasiteten på brikkene er veldig store. Behovet for kapasitet er stor, og det er også med på å øke restverdien av gamle GPU-er som vil øke potensiell den fremtidige restverdien av GPU-ene Iren sitter på i dag.

Et innblikk i fremtiden til Iren

Hittil har jeg dekket det meste jeg mener er relevant for fortiden til Iren. Den siste delen av denne analysen vil jeg dekke det jeg mener vil være relevant for selskapet fremover, og hva du/jeg som investor kan forvente fremover av selskapet, enten i form av nyheter eller kontrakter.

Når jeg skriver dette er klokka 10:07, 4 mars 2026. Det er mange måneder siden selskapet har annonsert noen form for nyheter. Sist nyhet var annonsering av Oklahoma på 1 600MW, men denne nyheter har liten påvirkning på aksjekursen før selskapet klarer å levere/bevise nåværende og fremtidige kontrakter gjennom cash flow. Foreløpig har kun 16M dukket opp på deres kvartalsrapport knyttet til AI/HPC, men dette vil øke drastisk de kommende månedene/årene. Utenom Oklahoma har ikke investorer hørt noe nytt siden Microsoft avtalen i November 2025.

Hoved essensen av fremtiden til Iren, er at jeg forventer at alle kontrakter vil være en forberedt MSFT avtale. Høyere revenue/MW, og mer bunnlinje for Iren.

Nyheter jeg forventer vil komme i 2026, der mange vil komme de neste månedene.

1. Canada (software?)

Jeg tror dette enten vil være Microsoft, siden de har visst stor interesse for datasentre i Canada, eller potensiell en aktør under Hyperscaler ligaen (Fluidstack,together AI også lignende). Jeg forventer nyheter om dette ganske snart, og skal ikke se bort ifra at dette er annonsert før jeg er ferdig med denne analysen. Som jeg har beskrevet i analysen er datasentrene i Canada mye mindre enn de i USA, men de kan ganske fort begynne å generere penger, fordi disse sentrene er allerede oppe å går, trenger bare å koble på GPU-ene.

Det ble også i deres Q2 2026 rapport for første gang snakket om en software løsning fra Iren. Iren har ikke tilbudt dette før, årsaken til det er:

“we do have an internal software capability. I think we probably downplay it a bit, partly in response to the market seeming to overplay it, but we’ve got that capability. To give you additional comfort, one of the contracts we are negotiating at the moment is a multi-billion-dollar contract where we need to bring a software solution. So it is not holding us back. It will not hold us back. The reality is exactly as what Kent said, you are dealing with the largest technology companies in the world. To pretend that you can be better at software and jam something down their throat when that is their competitive moat and that is their expertise, it’s just not congruent with reality.” Dan Roberts, Q2 2026 earnings call.

Hva dette egentlig betyr for Iren, og hva som legges i software er vanskelig å si, men forventer at denne «multi billion» vil være en moderat software løsning som vil komme fra deres Canada datasentre.

2. Sweetwater 1

Sweetwater 1 blir som nevnt i analysen energized i april 2026, Iren har flere ganger referert til dette som Q2 og April om hverandre. Men siden april kun er en måned unna i dag, forventer jeg at Iren nærmer seg en kontrakt på i alle fall de første 300MW som skal bli bygd. Selskapet har hatt 2 år på å forhandle kontrakter på denne siten, og jeg som investor forventer/håper at de kan få til dette innen den tid. Dette tror jeg personlig blir en mye større kontrakt enn MSFT avtalen, der hele avtalen vil gå ut ifra VR200, og ikke GB300. Jeg forventer også at Capex/MW blir redusert fra MSFT avtalen så lenge det ikke forekommer ekstrakostnader knyttet til sikkerhet/brann eller lignende. Iren har etter Microsoft avtalen forstått hvordan de skal bygge mest effektiv og lønnsomt, og vil bruke dette til å redusere kostnader ved videre bygging som ved SW1.

3. Annonsering av overlevering av Horizon 1 til Microsoft

Forventer at Iren holder sine tidsfrister med Microsoft og at Horizon 1 vil bli overlevert til MSFT i Mai med GPU-er som er klare for å genere penger for Iren og MSFT. Forventer også at de i samme slengen annonserer at de er i rute for de resterende 2-4.

4. Horizon 5-10

Etter en kontrakt med Sweetwater, og ferdigutbyggelsen av Horizon 1-4 for MSFT tror jeg Iren vil begynne byggingen av datasentre for de resterende 450MW i Childress. Der står det Bitcoin mining i dag, og tror selskapet vil kvitte seg med dette iløpet av 2026/starten av 2027. Dette er altså et datasenter som vil være 50% større enn Horizon 1-4 (300MW)(MSFT).

5. Annonsering av utvidelsen av Mackenzie fra 80-100MW, og Prince George fra 50-80MW.

Jeg forventer en annonsering av Iren i løpet av 2026 til å bekrefte utbyggelsen av datasentrene i Canada, og at dette vil begynne å påvirke regnskapet allerede i 2026. Har jo delt at dette er i gang, og tror 2026 blir året dette skjer.

6. Annonsering av GPU-er, og mulige kjøp fra andre Neoclouds

Det er mangel på GPU-er og er leveringene av brikkene har tatt lenger tid en antatt. Det er rykter om at Iren vil potensielt kjøpe GPU-er av konkurrenter som ikke har noen plasser å installere disse. Dette kan være med på å akselerere «Time to power», og påvirkning på regnskapet.

7. Annonsering av mer compute og økning av 2026 ARR guidance

Som nevnt i analysen tolket jeg det som at Iren potensielt hadde flere datasentre i batch 0 i Ercot. Dette betyr at det når som helst innen de neste 12 månedene kan bli lansert nye datasentre som er klare innen kort tid til å bli monetized/bygd. Samtidig tror jeg selskapet vil øke deres ARR guidance på $3,4B til nærmere $5-7B

8. NVIDIA tar mer del i Iren, og kjøper aksjer (kun spekulasjon)

Nvidia kjøpte nylig 2B i konkurrenten CRWV, og tror og mener at det samme vil skje med Iren, når de får mer tid til å bevise for markedet hva de er i stand til. Dette er kun spekulasjon men tror det vil skje innen 24 måneder.

9. Flytting av head quarters fra Sydney til USA

Tror at siden det blir mye kontraktforhandling og store deler av operasjonen drives i USA, at selskapet vil flytte deres headquarters over til statene. Dette i seg selv kan bety at selskapet kan bli inkludert i flere passive fond, og føre til mer eksponering og stabilitet i aksjen. Dette har vært diskutert av management tidligere, og tror at dette kan være en nyhet som kan påvirke aksjen i 2026 positivt (+5-10%). Dette er ikke bekreftet av Iren.

Fremtidige kursmål (bull case)

Har ved hjelp av @jiahanjimliu sin modell, laget en bull case på hvordan Iren kan ende opp med å se ut rundt 2030, hvis selskapet klarer å levere på kapasiteten de sitter på. Anbefaler alle å bruke litt tid på denne modellen for å legge inn deres antakelser og data, for å matche deres forventinger knyttet til bear, base og bull scenarioer:
https://jiahanliu.github.io/IREN-Community/

Det er mye antagelser, og det skal på ingen måte ses på som et finansielt råd eller anbefaling. Dette er etter min mening best case scenario:

Dette er bull casen til $IREN:

Skal prøve å gjøre dette så oversiktlig som mulig, så vil gå gjennom hvert enkelt datasenter. Mot slutten blir alt oppsummert, inkludert antakelsene jeg har gjort.

Canal Flats (Canada):

Størrelse: 30MW

Operasjonell: Ja

PUE (Power Usage Effectiviness): 1,1

ARR: $272,73M

EBIDTA marginer: 85%

Payback periode: 3,6 år

Resultat før skatt (Årlig): $99,12M

Mackenzie (Canada):

Operasjonell: Ja

Størrelse: 80MW

PUE (Power Usage Effectiviness): 1,1

ARR: $727,27M

EBIDTA marginer: 85%

Payback periode: 3,6 år

Resultat før skatt: $232,5M

Prince George (Canada):

Størrelse: 50MW

Operasjonell: Ja

PUE (Power Usage Effectiviness): 1,1

ARR: $500M

EBIDTA marginer: 85%

Payback periode: 2,7 år

Resultat før skatt: $232,5M

Horizon 1-4 (West Texas)

Størrelse: 300MW

Operasjonell: Snart (Kontrakt med $MSFT)

Konstruksjon: (ferdig rundt oktober 2026 med H4)

PUE (Power Usage Effectiviness): 1,5

ARR: $1,94B

EBIDTA marginer: 85%

Payback periode: 5,7 år

Resultat før skatt: $389,5M

Horizon 5-10 (West Texas)

Størrelse: 450MW

Operasjonell: Nei

Konstruksjon/kontrakt: H1 2027???

PUE (Power Usage Effectiviness): 1,5

ARR: $4,05B

EBIDTA marginer: 85%

Payback periode: 4,4 år

Resultat før skatt: $1313,02M

Sweetwater 1 (West Texas)

Størrelse: 1 400MW

Operasjonell: Nei (Energized april 2026)

Konstruksjon/kontrakt: Gjennom hele 2026, 2027 og sikkert 2028 også.

PUE (Power Usage Effectiviness): 1,5

ARR: $16,1B

EBIDTA marginer: 85%

Payback periode: 3,7 år

Resultat før skatt: $6 874,04M

Oklahoma

Størrelse: 1 600MW

Operasjonell: Nei

Konstruksjon/kontrakt: 2028/2029

PUE (Power Usage Effectiviness): 1,5

ARR: $14,4B

EBIDTA marginer: 85%

Payback periode: 4,3 år

Resultat før skatt: $4721,86M

Sweetwater 2 (West Texas)

Størrelse: 600MW

Operasjonell: Nei

Colocation: Ja

Konstruksjon/kontrakt: Mot slutten av 2027

PUE (Power Usage Effectiviness): 1,5

ARR: $0,73B

EBIDTA marginer: 85%

Payback periode: 8,2 år

Resultat før skatt: $432M

Oppsumering:

For å oppsummere vil totalt disse 8 datasentrene som Iren har godkjent kapasitet på genere:

Årlig omsetning: $38,73B

Årlig EBT: $14,33B

Årlig Net profitt: $13,74B

13,74B x 35 (PE)= 480B MCAP

Antar fremtidig dilution til å være 40%

480B / 572,8M (409m (dagens aksjer) x 1,4)

= Aksjekurs på $840

Antagelsene som er gjort:

• Antar operasjonell kapasitet på 4 500MW

• Alle datasentrene utenom SW2 vil være IAAS, altså at Iren kjøper opp databrikkene og leier datasentrene + GPU-ene. Dette er svært kapitalkrevende, og vil kreve mye fra management.

• Antar at alle datasentrene får kontrakter.

• Antar at restverdien av GPU-ene vil være 15% (base case?) etter 5 år. Altså at Iren kan selge brikkene til 15% av kjøps verdien om 5 år. (eks: Iren kjøpte 5,8B NVIDIA GPU-er fra Dell. Etter 5 år med avtalen til MSFT kan Iren selge dette til $870M.

• Antar at fremtidige kontrakter vil ha 20% prepayment som $MSFT kontrakten.

• Antar at Kapitalen som trengs å hentes for å finansieres GPU og Capex kjøp vil ha en rente på 6%. (Iren hentet dette til under 6% for $MSFT avtalen, men det gir mer mening at den økes ettersom Iren tar opp mer og mer gjeld.)

• Antar datasenter lifetime value på 20 år.

• Antar at SW1 vil ha en 20% bedre kontrakt enn $MSFT kontrakten.

• Antar at de kommende resterende datasentrene vil ha 15% bedre kontrakt enn $MSFT (Se på leieprisen på Hoppers modellen👀)

• Antar at kostnaden per MW holder seg rundt 15M/MW (Konservativt, vil trolig gå ned mot 12M/MW imo)

• Antar PUE på 1,5 på alle datasentrene utenom de i Canada (Konservativt imo).

• Antar utvanning på 40%, dette tallet kan være mye høyere, helt avhengig av execution og makro.

• Antar GPU levetid på 5 år.

• Antar at det ikke vil bli annonsert noen flere datasentre fra Iren, selvom det liggeri kortene at det vil komme flere.

• Antar ikke fremtidig ekspansjon (mai 2026) av Mackenzie (80MW ---> 100MW) eller andre datasentre.

• Antar null fremtidig Bitcoin mining

• Antar PE på 35

• Antar ingen reforhandling av MSFT avtalen til potensielt VR200.

Det er altså mye antagelser, og kan garantere at utfallet vil bli helt annerledes enn det jeg forventer. Det følger med svært stor risiko og investere i Iren, de må hente kapital til nærmere $100B for å monetize og drive IAAS over hele deres anlegg. Derfor er kjernen i tesen rundt Iren om tillit til management, og det har jeg.

Irens Konkurransefordel

• Fullintegrert

I motsetning til størsteparten av konkurrentene til Iren er de 100% fullintegrerte. Iren eier, og har bygd opp alt av deres datasenter. Dette gjør at ingen marginer blir spist opp av en 3-part. Konkurrenter som Nebius og Coreweave bruker hovedsakelig 3-parter for å leie kapasiteten fra dem, noe som spiser av marginene. Årsaken til at Iren har dette, er fordi de søkte/kjøpte kapasitet før denne ble så knapp. Iren skaffet dette før CHATGPT/LLM’s bølgen inntraff. Som Dan selv sier i en podcast: «Why put your destiny in the hands of a 3 party, control it yourself because people let you down. And if you want something done, your better of doing it yourself”. Så ikke nok med at du slipper å forholde deg til skuffelser og samtaler med 3-parter, men du beholder alle marginene selv, som er antatt til å være omtrent 20%.

• Compute

Iren sitter med 4 500MW med secured power. Dette i seg selv er en enorm fordel, som gjør at de slipper å slåss i køene som nå har oppstått i strømnettet. Som konkurrenten Nebius sa i deres earnings angående kontrakten med Meta: “The size of that contract was limited to what power they had available. If they had more, they could have sold more.” All kapasitet som er tilgjengelig blir solgt, og det er kunder som ønsker dette.

I Ercot var ventelisten på 63GW for et år siden, og har økt til 226GW (desember 2025), og har trolig økt siden det. Det er ikke nok kapasitet i forhold til etterspørselen, og Iren sitter på en enorm portefølje. OpenAI gjorde en avtale med Oracle, Softbank angående «The Stargate Project» som var på 10GW innen 2029 til en sum på opptil $500B. Og med Irens SW1 datasenter på 1 400MW som blir energized i april 2026, er dette noe som vil være attraktivt for kunder som Google, Meta, Anthropic osv.

• 100% renewable

Iren har siden dag 1 vært 100% kjørt av fornybar energi. Dette i seg selv er en kjemperessurs for kundene, som er avhengig av å følge klimamål osv. Dette i seg selv er attraktivt for kunder som ønsker å drive så grønt og bærekraftig som mulig for deres klimaregnskap og lignende.

• Allerede bygget datasentre for ekstrem ytelse

Som jeg har forklart i analysen har Iren allerede bygd datasenteret som har oppnådd det som er mulig i det ytterste laget (80kW rack density i Prince George), og bygger nå datasentre som skal klare de nyeste VR200 modellene. Dette i seg selv er svært imponerende og viser til at Iren i dag sitter med mye kunnskap gjennom utbyggingen, men ikke minst management teamet som jeg nevnte i starten av analysen, som har kunnskap med flere tiårs erfaring.

• Samfunnspåvirkning/plassering

Iren er kjent for å bygge opp lokalmiljøene de operer i, og hvis når de bygger opp datasentrene bruker de så mye som mulig av arbeidere og materialer fra lokalsamfunnet. Informasjonen jeg har hørt er at alle møblene som er plassert i datasentrene er kjøpt lokalt, og Iren bruker lokale hjelpeorganisasjoner og støtter disse.

Videoer og informasjon på dette:
youtube.com/watch?list=TLGGhuKHJCpLYq4wNDAzMjAyNg&v=DmXzL0nTn08&embeds_referring_euri=https%3A%2F%2Firen.com%2F&source_ve_path=Mjg2NjY (47 sekunder)

https://iren.com/company/community-grants (4 minutter 36 sekunder)

Iren har også deres datasentre langt borte fra lokale. Det er tilnærmet ingen som blir plaget av støy og lignende fra Iren sine datasentre. Vi har sett de siste månedene at det har kommet flere opptøyer blant sivile angående utbyggingen av datasentre med tanke på frykt for støy og høyere vann og strømpriser. Denne risikoen er mye mindre for Iren, ig i hvert fall når deres datasentre er hovedsakelig i Texas målt i størrelse.

Konkurrenter

De største konkurrentene til Iren vil jeg personlig si er Coreweave og Nebius. Dette er begge neoclouds som kjører en lik strategi som Iren. Det som skiller dem, er at Nebius sier de har en software del i deres AI satsing. Betydningen av dette er vanskelig å si, siden Nebius holder kortene veldig skjult rundt deres kontrakter.

Nebius har også en stor kontrakt med Microsoft på $17,4B over 5 år (300-350MW IT load). Informasjonen er veldig knap, og det eneste en har å gå på er at det har blitt rykter om at dette vil være over 100 000 GB300, og ingen CAPEX er annonsert knyttet til avtalen. Dette i seg selv gjør at jeg holder meg unna selskapet, selv om jeg personlig har eid aksjen tidligere ($25). At informasjon holdes skjult, gjør at jeg føler de har noe å skjule. Åpenhet er viktig for meg som investor, og er noe jeg verdsetter høyt med Iren. Nebius har selv guidet for over 3GW med kapasitet for 2026:

Coreweave er den mest etablerte aktøren i bransjen med 810MW of active power capacity. De har ekspandert total contracted power til 3,1 GW (3 100MW). Selskapet har i dag over 250 000 GPU-er, og brakte 100 000 GPU-er live i 2025. Selskapet har høy gjeldsgrad med DEBT/Equity på 8,9. Iren ligger på 1,5. Nebius ligger på 1,1. Disse tallene er hentet fra Fiscal.AI.

For å sammenligne revenue for Q4 hadde Coreweave $1,6B

Nebius hadde $228M for Q4

Iren hadde kun $16M.

Coreweave har altså 100x mer revenue enn Iren i dag, og jeg tror over de kommende årene at Iren vil bli den største og mest lønnsomme neocloud selskapet. Med dette forventer jeg også at Iren sin markedsverdi vil passere Coreweave og Nebius sin.

De største risikoene jeg ser med Iren:

Det er veldig mange risikoer knyttet til å investere i en så «hot» bransje og sektor (datasentre/Ai). Jeg skal nevne de viktigste risikoene jeg ser på som kan utfordre fremtiden til Iren. Dette er de største risikoene:

· Tilgang på kapital:

For at Iren skal klare å ekspandere i den farten som er ønskelig i denne sektoren er de avhengig av å hente så mye og billig kapital som mulig. Iren har hittil bevist at de klarer dette, men makrobildet kan endre på dette fort.

· Gjeld/Capex:

Den totale kostnaden for å bygge ut alle Iren sine datasentre, og potensielt kjøpe alle GPU-ene vil få en enorm prislapp på rundt $130B (antar samme capex som ved MSFT avtalen) (10x av dagens MCAP). Dette vil jo foregå over mange år, men dette medfører stor risiko for selskapet, og det er derfor viktig at en har tillit til management, og at de vil gjøre det som er riktig og nødvendig for den langsiktige «helsen» til selskapet.

· Etterspørselen synker

Om dette er realistisk de kommende årene er jeg usikker på, men det kan bli en realitet. Dette betyr for Iren at det blir vanskeligere å forhandle gode avtaler, og de vil miste marginer og bunnlinje på dette. Det vil også redusere verdien på datasentrene og GPU-ene som igjen vil redusere verdien til selskapet.

· Klarer ikke å levere

Hvis Iren ikke klarer å levere datasentre til kundene på tid og kvalitet. Dette vil ødelegge deres rykte, og forhindre fremtidig kunder og kontrakter, som vil påvirke veldig negativt for selskapet.

· Energiportefølje hovedsakelig bestående av Ercot og SPP

Strømleverandører består av Ercot i West texas (Childress, SW1, SW2) og SPP (Oklahoma). Hvis det skulle skje noe her, er konsentrasjonen stor i Texas. Dette kan være med på å påvirke negativt, men sliter med å forstå hva det skulle vært, fordi Iren har kontrakter på dette.

· Utvanning

For å kunne skalere videre er det å forvente at utvanning vil skje, management er tydelig på at de skal prøve å unngå det i størst mulig grad. Og kunne brukes når nødvendig. Vi har sett tidligere at Iren har utvannet aksjer på den laveste aksjekursen i fjor på 6$, dette er noe som mange synes var veldig negativt.

Avslutning:

For å oppsummere er dette et selskap som jeg er svært optimistisk til, og derfor selskapet i dag er over 50% av min personlig portefølje. Jeg tror de neste 36 månedene vil bli utrolig bra for selskapet og aksjekursen. Iren sitter på knappe ressurser som alle ønsker å ta del av. Hvis Iren klarer å kapitalisere på dette, kan dette selskapet vokse seg ekstremt stort ($B100’s). Denne analysen skal ikke ses på som et finansielt råd eller anbefaling, og er kun for underholdning. Mine synspunkter for selskapet er kun gjeldende fra skrivende tidspunkt, og synet mitt kan endre seg fort på selskapet.

Dette er en analyse jeg har brukt ekstremt mye tid på, og håper du som leser denne finne noe form for nytte fra denne analysen, og husk å gjøre egen analyse og research på selskapet.

Iren kurs på ferdigstillelsen av analysen (børstart) (04.03.2026) er på $38,85.

Vil gi en spesiell takk til Frans Baker (@FransBakker9812) (anbefaler å abonnere på hans X profil, mye informasjon å hente), Agrippa Investments (@Agrippa_Inv) Substracken hans (anbefales på det sterkeste for Iren investorer):

Agrippa Investments

Institutional-Grade Research & Analysis on Hyper-Growth Stocks.

By 𝐀𝐠𝐫𝐢𝐩𝐩𝐚

, og Jim Lui (@jiahanjimliu) for hans tanker og finansielle modell knyttet til Iren: https://jiahanliu.github.io/IREN-Community/.