IREN Limited (IREN) Earnings Call Transcript & Summary

November 5, 2026

NASDAQ US Information Technology Software earnings 65 min

Earnings Call Speaker Segments

Mike Power

executive
#1

Thanks very much, operator. Good afternoon, everyone, and welcome to Iren's presentation of results for the first quarter of fiscal 2027. My name is Mike Power, and I serve as Vice President of Investor Relations. Joining me on today's call are Daniel Roberts, our Co-Founder and Co-CEO; Anthony Lewis, our CFO; and Kent Draper, our Chief Commercial Officer. Before we get started, please be aware that this call is being webcast and comes with an accompanying presentation; if you are joining by phone, you may choose to pose a question through the moderator once our prepared remarks have concluded. I want to remind everybody that some statements we make on this call may count as forward-looking statements. Such statements rest on current expectations and assumptions and carry risks and uncertainties that might cause actual results to differ materially. For further information, please see slide two of the accompanying presentation in the Investor Relations section of our website, along with our SEC filings. On this call, we will present both GAAP and certain non-GAAP financial measures. Reconciliations of these measures to the most directly comparable GAAP measures appear in the appendix to today's earnings press release and in the appendix to the accompanying presentation, both of which are posted in the Investor Relations section of our website. With that, I will hand the call over to Dan Roberts.

Daniel Roberts

executive
#2

Thanks, Mike. Good afternoon, everyone, and I appreciate you joining us. In the quarter, we signed new multi-year agreements — among them one with a leading frontier AI lab — which lifted revenue backlog past $20 billion, with two-thirds of that coming from investment grade customers. We also closed GPU financing that, together with customer prepayments, covers more than 100% of the GPU capex for the Mackenzie deployments. We handed Horizon 1 over to Microsoft with NVIDIA Exemplar Cloud status. And finally, we pushed forward construction of new data center campuses around the globe, including our flagship 2 gigawatt Sweetwater hub. Versus the prior quarter, AI cloud revenue nearly tripled, reaching $200 million. Operating ARR doubled, reaching $1 billion, while adjusted EBITDA margin improved from 14% to 29%. There's one date that has moved, and I want to tackle that right away. In July, we indicated that over $4 billion in operating ARR was expected to be online by the close of the year; we're now in January. There are two factors behind this: first, one customer asked for a design change, which we accepted, and it goes beyond what was originally scoped; and second, supplier timing. The data centers are essentially complete, and we're awaiting the last equipment so commissioning can be finished. Should suppliers come through, we're aiming for January. If they run late, as suppliers have throughout the industry this year, it could push to the end of February. Doing this kind of build at scale is complicated—and it gets even harder when you control every piece yourself: the land and power, the data centers, the compute, and the team needed to pull it all together. We've put years into creating every part of it. Right now we operate three businesses, which are laid out on slide four of our presentation. The first is AI cloud: it has $1 billion of operating ARR today, with more than $4 billion contracted and fully funded, and that ramps up as the rest of the capacity comes online. The second is the data centers: we hold $6 billion of data center assets today, with a target of $10 billion by June next year—20-year assets capable of supporting multiple generations of compute, funded with cash so far and unencumbered. And third, the last point: land and power. We hold over 30,000 acres worldwide, which backs more than 5 gigawatts of secured power plus over 35 gigawatts still in development. A lot of those projects are early stage, and not every one will end up converting into grid connection agreements. From the very beginning, our approach has been to lock in land and power ahead of demand, and that's the reason the platform now reaches three continents. Most companies handle just one of those three; we handle all of them, which means we retain a larger share of the economics, gain additional avenues for funding growth, and keep more control over delivery. Compute demand has not slowed down. After the quarter closed, we signed deals worth $22 million in annualized revenue per megawatt of IT load on a five-year term, plus more than $40 million per megawatt on a contract of shorter duration. Notably, that five-year deal carries a customer prepayment that covers roughly 90% of the GPU capex tied to it. With that same top-tier AI frontier lab, we have expanded our AI cloud agreement further still, and we secured H100s at double the original price — and that pricing backdrop is what shapes the way we contract our Vera Rubin capacity. Over recent weeks, capital markets have grown pickier, and the discussion has clearly shifted away from the size of the contract toward whether it can actually be delivered, financed, and operated profitably. Those questions are the correct ones, and this quarter gives our answer to them. Horizon 1 moved from civils to an NVIDIA Exemplar-validated GB300 cluster in roughly 14 months. JLL estimates the worldwide average for constructing a 50 megawatt data center at 18 months, and that figure covers the building only. We completed both the building and the cluster within it four months ahead of that benchmark. I'll return later to what that final stage entails. Operating capacity climbed 127% during the quarter, reaching 0.1 gigawatts. Across the 12 months through September, we raised $19 billion in total, which includes $6.5 billion of GPU financing obtained since May at rates of 6 to 9.5%. Our AI cloud achieved an 86% gross margin, matching the prior quarter. This is precisely where Iren diverges from a typical neocloud. Rather than leasing another party's data center and carrying that liability on our balance sheet, we lock down the power, we develop the land, we construct the infrastructure, and we deploy the compute. Every deployment leaves us owning long-term assets. Because we own the power, the land, and the data centers, we also control the pace. We aim to build as quickly as demand permits, yet we aren't required to develop every site simultaneously. The power is secured and the data centers belong to us, so we don't sign contracts or raise capital in order to cover lease payments. Decisions on investment and contracting rest on risk-adjusted returns, financing conditions among them, and the economics support building. During the quarter, average revenue per megawatt across our three- and five-year contracts kept rising, now supporting GPU paybacks of roughly 18 months or even less, depending on how much customers prepay. At Mackenzie, customer prepayments combined with GPU financing covered over 100% of the GPU capex tied to it. Deals inked following the quarter’s end bolster that view: attractive rates plus clients covering a large portion of GPU capital spending. With more capacity coming online, we’re also improving how we roll out and run it. Working alongside NVIDIA, and with the Mirantis andCourted? AIteams, we’re boosting the way we provision, administer, and support these environments. Our installed base continues to renew and grow with us. Fresh logos show demand, while renewals show consistent execution; so now, let’s turn to execution on the ground. Slides 13 and 14 of the deck present a side-by-side view of progress on our planned buildout to 0.8 gigawatts by the close of 2027, and on data center growth beyond that horizon. The first 0.3 gigawatts is in its final stages, and construction of additional capacity is underway at six sites. That final stretch is where things get most complex. What the drone photos don't capture is this: GPUs installed in racks, extensive fiber laid, complete clusters wired up and validated, and the data center operation transitioning into the AI cloud operation. The know-how gained from those deployments is likewise informing the design of future AI factories. Under the leadership of John Grothe, our Chief Innovation Officer, we are collaborating intensively with NVIDIA on data center design and on faster construction methods. That effort extends to extracting greater capacity from existing sites, drawing on spare power capacity in current connections so we can run additional compute. Mackenzie illustrates this: 100 megawatts of power secured, and mass grading has now begun for new liquid-cooled data centers constructed to the same specification we have already delivered. Slide 17 lays out the pipeline supporting both the data center and AI cloud businesses I just covered. Sites keep being originated by our development team in advance of the coming growth phase. Those over 30,000 acres stretch across North America, Europe, and Asia-Pacific. As a result, we can decide where to serve customers and where to advance sites depending on what local conditions permit. Not every market has to progress at an identical rate. The question of where we build matters as much as the question of how we build. Our preference is rural areas with suitable setbacks from neighbors, and we talk with communities and regulators before moving grid applications forward so that local priorities are understood from the start. A strong illustration in Australia is the 900 megawatt Redbank site we announced recently. We recruit from the local area, source from local suppliers, and give priority to delivering water to site for our closed-loop cooling systems, which lessens dependence on local supplies. Our aim is for projects that deliver durable jobs and investment to these communities while honoring what residents there need. Growth is being approached with discipline: customer contracts that renew, returns sufficient to justify the capital, and financing tied to every deployment. With that, I'll pass it to Anthony, who will go through the figures and explain how the next phase is being funded.

Anthony Lewis

executive
#3

Thanks, Dan, and good evening, everyone — I appreciate all of you joining. Starting with our results, we delivered another quarter of strong growth in AI cloud revenue, supported by continued execution. AI cloud revenue was $200 million for the September quarter, up 183% quarter-over-quarter, which reflects the continued deployment and monetization of our data center and GPU capacity. Total revenue came to $228 million, up 66% quarter-over-quarter, while Bitcoin mining revenue fell to $28 million as we continued actively decommissioning Bitcoin mining hardware ahead of GPU installations. Our aim is to largely complete this transition by the end of this quarter. Gross margin before depreciation and amortization was 80%, versus 76% in the prior quarter, as AI cloud revenue accounted for a larger share of total revenue. AI cloud gross margin before depreciation and amortization was 86%, consistent with the prior quarter. Adjusted EBITDA was $66 million, a margin of about 29%, compared with 14% in the prior quarter. Operating expense, including SG&A and R&D, was $224 million, up $96 million quarter-over-quarter, reflecting a $33 million increase in cash operating expense, as we brought on the Mirantis operations from the middle of the quarter and continued significantly expanding head count to support our growth, and this was after stock-based compensation of $63 million. Once you factor in depreciation, amortization, financing costs, and other expenses and income along with other items, the quarter ended with a net loss of $263 million. What these results show is our ability to convert infrastructure investments and customer commitments into operating capacity that generates revenue. As Dan noted, our 0.3 gigawatt expansion is nearing completion, and the related ramp toward more than $4 billion of operating ARR should extend into early 2027, largely because of design changes customers have requested and supplier timing. We expect ARR to rise substantially beyond that level as additional capacity becomes available over the course of calendar year 2027. Even so, we're not setting a particular goal right now, because how much ARR we reach and when it lands both depend on which contracts get signed, when additional capacity becomes available, and when deployments receive acceptance; turning now specifically to the December quarter, we closed September with $1 billion in operating ARR, and we expect to book a quarter of that as revenue in the second quarter. We don't expect the added Horizon and Mackenzie capacity to contribute meaningful revenue, since most of it won't come until late December or the start of calendar 2027. Sequentially, for the December quarter we expect cash operating expense, SG&A and R&D included, to increase by roughly $50 million to $60 million. Before turning to the balance sheet and funding, I want to spend a few minutes on recent contracting activity and on how prepayments support our funding plan and produce better paybacks and returns on our investments. Demand for our AI cloud services remains solid, as Dan mentioned, and we're seeing attractive opportunities involving hyperscalers, frontier AI labs, and other customers. During the quarter, pricing for three-year terms came in at $19 million to $23 million per megawatt of IT load, while five-year pricing ran about $17 million. As Dan pointed out, once the quarter ended we signed contracts ranging from $22 million to $40 million per megawatt. That demand level gives us room to be selective about the contracts we accept and how we structure them. Customer prepayments are a key element of our model. Dan noted that some recent contracts have included customer prepayments reaching as high as 90% of the GPU capex associated with them. Prepayments reduce the upfront funding required, and they help align capital deployment with contracted demand. Combined with equipment financing, they can meaningfully lower the corporate capital a deployment needs and shorten the cash payback period on the overall investment. Slide nine in the presentation we've provided shows a hypothetical one-megawatt liquid-cooled deployment needing approximately $60 million in capex. One-third of that amount goes to the data center, two-thirds goes toward GPUs, and customer prepayments span 0 to 90% of GPU costs. With a five-year contract priced at $20 million per megawatt and a prepayment covering 50% of GPU capex, the GPU payback period works out to 18 months, and including the data center cost, payback is roughly three years. These figures are for illustration only—actual deal economics and payback timing differ based on the customer, the hardware, and how a contract is structured. What matters is that contracted revenue isn't our only focus; we care about how efficiently we can finance that revenue and what returns it produces. On balance sheet strength: as of September 30, total assets stood at $20.6 billion, including $12.3 billion in property, plant, and equipment. That includes $6 billion of data center assets that are unencumbered. Cash was $5.8 billion, of which $2.3 billion was restricted, set aside mainly to fund GPU deployments. Total debt was $10.2 billion, mostly low-coupon convertible notes, with maturities largely in 2032 and beyond. In the twelve months through September 30, we raised $19 billion of funding from multiple sources, including customer prepayments and secured GPU financing. Counting the portion of that financing not yet drawn, we have roughly $9 billion in cash and undrawn commitments available to support growth. On planned capex, about $5 billion was spent during the quarter. For the rest of the fiscal year, planned capex totals approximately $15 billion, below our earlier estimates, reflecting flexibility in how we pace data center and GPU capex. With $9 billion in cash and committed funding sources, $6 billion of additional planned capex remains to be funded. With this revised plan, we now anticipate that Childress air-cooled capacity deployments will be distributed throughout calendar year 2027, versus the more front-loaded timing we had previously expected. We'll direct more investment toward our longer-term capacity objectives, and we will continue updating the plan as conditions evolve — fundraising markets, contracting, customer requirements, and other factors. As Dan noted, this is exactly where our vertically integrated model's strength shows: it affords us maximum flexibility to adjust planning as circumstances change. Turning now to the funding plan. On slide 19, you'll see that by the end of FY27 we anticipate holding $10 billion of unencumbered data center assets plus $6 billion of unencumbered GPUs, prior to any further secured financing. Beyond that, we have cash flows backed by multi-year contracts with high-quality customers — Microsoft and NVIDIA, both investment grade counterparties, account for two-thirds of our backlog. That leaves us with a broad set of funding options, such as data center financing, GPU prepayments and financing, and other corporate-level initiatives — indicatively $12 billion to $15 billion available, against the $6 billion we need to fund, and that's on top of a range of roughly $50 million to $60 million in additional capacity we can draw on. Alongside other capital market sources like convertibles and equity, this puts us in a very strong position to fund our fiscal year 2027 growth and beyond. Let me go into more detail on some of what we're planning. On data center financing, near-term focus will be on Horizons 1 through 4 — the ones supporting the Microsoft contract, as you'd expect. Several initiatives are moving ahead in parallel, taking into account how far construction and delivery have progressed, the AAA-rated cash flows behind the Microsoft contract, and the long-term, high-quality character of the physical infrastructure. On the GPU side, prepayments will remain a core part of our overall funding strategy, as we've previously said. Across the portfolio, for GPU capex including prepayments, we're targeting total financing equal to 90% of the GPU capex involved. That's consistent with the investment grade and non-investment grade GPU financings we announced earlier — we achieved roughly 95% funding on the first and more than 100% on the second. As for near-term GPU financing priorities, we have work streams progressing for the Childress air-cooled deployments. NVIDIA and the leading lab contracts will serve as anchors, and the lab contract was expanded recently. Beyond data center and GPU financing, you should expect further corporate-level initiatives in the future, and as the platform grows we'll maintain the appropriate mix of debt and equity funding. In closing, this quarter showed how we convert infrastructure spending and customer commitments into operating capacity and high-margin revenue, and we believe we're strongly positioned to continue scaling the business while adapting our growth to external conditions. With that, I'll turn things back to Dan.

Daniel Roberts

executive
#4

Thanks, Anthony. Let me close by returning to my opening theme: execution. We're executing on three fronts across our company: acquiring land and power, constructing data centers, and activating AI cloud capacity. By June of next year, we anticipate holding $25 billion in data center and GPU assets. Over the 12 months through September, we raised $19 billion. The capacity supporting our more than $4 billion of ARR is completely funded. Our additional $6 billion in planned expansion capex is matched by $12 to $15 billion in sources before any converts or equity; we retain the flexibility to slow down or speed up that spending. We'll continue building wherever contracts, funding, and returns make it worthwhile. Our upcoming goal is surpassing $4 billion in operating ARR—a funded target set for January. Thanks for being with us; we're glad to take your questions now.

Paul Golding

analyst
#5

Many thanks for taking my question, and well done on all the advancement and the fresh leasing. I'd like to ask — there's been quite a bit of press coverage regarding the Microsoft capacity delivery, and I was wondering, Dan, whether you could offer any further detail on some of the dynamics you mentioned in the prepared remarks, perhaps breaking down how much of the supplier delay was tied to the change orders or whether they're unrelated, and just any other clarity around some of that commentary we've been hearing in the marketplace. Thanks very much.

Daniel Roberts

executive
#6

Thank you, Paul. Certainly. Back in August, we told everyone that Horizon 1 had been transferred to Microsoft. After that, they asked us to reconfigure and upgrade certain network infrastructure components. The goal was to make 200 megawatts of capacity available for the newest frontier models. This market shifts extremely fast, doesn't it? When the contract was signed in 2025, not every scenario had been considered. Well, we agreed. Sure, no issue at all. Our job is to tell customers yes rather than no. So there will be a minor delay; we're aiming for January. During the release, we mentioned that this hinges on supplier scheduling. Suppliers have had a difficult year when it comes to dependable delivery, so there's a real possibility it slips into late February. On construction, we're in solid shape. Horizon 2 is complete, power is live at the rack across all three halls, the GPUs are installed, the fiber is in place, and Dell is currently commissioning. One integrated test is underway there. At Horizon 3, every GPU has arrived on site and all three halls have power. Horizon 4's GPUs will be delivered over the next few weeks. We've seen the speculation, but the relationship is strong. I believe that, in certain respects, it speaks to the strength of the relationship and the business model that we could adapt and stay flexible regarding this network infrastructure. Let me give you a single illustration. Had we been a different kind of company—an asset-light neocloud working with a colocation partner—I doubt Microsoft would have put forward that same request, since as a cloud provider you would have to return to your colocation provider and request those modifications. Of course, 31 December—would we have been delighted to reach it? Without question, we came very close, but it now appears it will be January, possibly February, based on how suppliers perform.

Paul Golding

analyst
#7

Thanks a lot for that context, Dan. Oh, apologies. I just wanted to pick up on those points, if that's alright. So how ought we to think about capex per megawatt when it comes to some of this upgraded networking equipment? Does that math shift at all regarding that 14 to 16, or should we expect this to be more widespread across leasing that's already been completed in the form of change orders? How should we think about that, and the tenant's responsibility to cover any change orders of that kind? Thanks very much.

Daniel Roberts

executive
#8

The extra capex for that network upgrade is really negligible when you look at what we're spending on the cluster as a whole. It covers new cables, labor, and simply a different approach to how we implement it — I wouldn't focus on that. As for the benefit, it essentially brings the entire 200 megawatts up to speed for those newest, most advanced use cases. And one more thing I should have noted: we're now taking that learning and applying it across every other site we have under construction. I suppose that's an advantage of operating at the leading edge and delivering. Our long-standing view is that hands-on effort is how capability gets built — the more you practice, the more insight you gain, and those insights accumulate and feed into the wider platform, which is exactly what's happening right now.

Paul Golding

analyst
#9

Great. Thanks so much.

Brett Knoblauch

analyst
#10

Hi, guys. Thank you for taking my question. So maybe to start, there's been a lot of talk lately about modular data centers, and I'm curious whether you guys are looking into that at any of your sites around the globe. I know NVIDIA's coming out with their own modular products. Aivres, I think, signed a powered land lease for that product, if I had to guess. Is that something you guys are looking at, or do you guys want your own design and build, or are you flexible on that?

Daniel Roberts

executive
#11

We've actually been digging into this for a good while now, and we have dedicated teams in-house — formed specifically over the last nine months — that focus on construction methodologies and on making things faster and simpler. A couple of key ideas sit at the heart of that. The first is modularization, plus being able to plug and play more components on-site — having a standardized methodology and a standardized supply chain, so it really does become more plug and play, and shifting a lot of the construction off-site into factories, where it scales better. Modular design is something we've been examining for a good while. And as all of you are well aware, we have a very close relationship with NVIDIA. We're in discussions with them regarding various design options. Whether that particular DSX modular approach ends up being carried out in the form that's been shown is something we're still working through, but certainly the ideas and the broader North Star — attempting to make construction simpler, faster, and easier while cutting on-site labor needs — is definitely where both the industry and we are headed.

Brett Knoblauch

analyst
#12

Awesome. So then I suppose I could just ask a follow-up about how you folks are thinking regarding contract duration, maybe. I know you're signing with what you might call a barbell strategy — long, short pricing — and the economics differ somewhat between the two. Given this environment where funding is a concern, is having more IG-backed customers something you're emphasizing, or do you not see it all that differently when you're signing deals with customers who prepay the entirety of the capex in some cases? Thank you.

Daniel Roberts

executive
#13

Good question. The answer is both. Those two things aren't mutually exclusive, and you touched on it with the barbell approach. Right now we have a $20 billion backlog of customer contracts, and it's been constructed very deliberately, with two-thirds of that being investment grade; we're talking names like Microsoft, NVIDIA, $4 trillion credits standing behind that. Then, yes, there's a portion that is lower credit, but as you mentioned, those tend to go hand-in-hand with higher prepayments, and they're central to our broader strategic objective of servicing the broader ecosystem. We're very happy with the investment grade counterparties. We have a fantastic relationship with Microsoft, a really good relationship with NVIDIA, and we'd love to keep servicing their requirements. Equally, in parallel, we've said for quite some time that servicing the broader ecosystem is important. I feel like we've constructed that customer book deliberately, and in a position that leaves us in a good spot today, where we have forecast $25 billion in data center and GPU assets at 30 June next year, against that sort of credit profile, which dovetails into the financing and how we look to grow from here.

Brett Knoblauch

analyst
#14

Great. Thanks a lot.

Joseph Vafi

analyst
#15

Hi, everyone. Outstanding results. I appreciate you taking my question. I really appreciate what you said, Dan, regarding vertical integration and how you can ramp development and rollout up or down. So with that in mind, could you give us some insight into what you monitor in the market that might affect how fast you develop — factors that could lead you to slow down or speed up the pace of development, such as supply chain, financing, demand, your physical ability to build the infrastructure, or anything else? Thanks.

Daniel Roberts

executive
#16

Thanks, Joe. We've been talking since our IPO five years ago — almost exactly that date is coming up — and I think what we've witnessed during that period, plus the three years of our history before it, is that we've kept building this asset base with a solid long-term foundation. There's no short-term thinking; the company was founded eight years ago with a long-term vision focused on this digital exponential demand for compute, and we felt it was appropriate today to reveal a bit more about that development portfolio. We now have 30,000 acres of land secured worldwide, supporting 5 gigawatts of secured grid connection and a further 35 gigawatts of prospective grid connections. Each layer has been constructed upon the one before it. First comes land and power. After that, we have cash-funded every one of our data centers to date, with not an ounce of debt against them — it's cash-funded, $6 billion today, projected to reach $10 billion by 30 June — and then we purchase the GPUs on top of that. The advantage of this approach is that we aren't a WeWork-style business model, meaning we don't sit between an infrastructure provider and a compute user who leverages that infrastructure. Since we own and control our own destiny — and I know you've heard these words many times over the last five years — we are fully in control. We can flex up, we can flex down, and that relates to global diversification of projects, the ability to throttle up and down based on the local environment, it relates to capital availability, it relates to demand strength, and I think that's a great segue to comment on the market environment we're seeing. It's actually incredible when we consider how the markets are behaving, but then look at what's happening at the customer level and the real substance and fundamentals of our business model. The demand — I know we tend to say this every quarter, but I'll reiterate it again — has never been stronger. As we just announced, these are deals signed this week. This isn't old information. We just had a customer agree to fund 90% of the GPU capex. We signed a short-term deal at $40 million per megawatt. One of the top frontier AI labs — the very same one we told you about a couple of months back — has increased its capacity once more. And I think it's worth pointing out that we hold some older-generation H100s, and we've just signed those older units to new contracts at twice the original price. Demand for compute, along with the supply constraints everyone is witnessing, is generating a remarkable opportunity at the underlying business level, and the disconnect with financial markets right now is really striking. I suppose what benefits us with our model is that we're able to be patient, we can navigate these market dynamics, and we can be ready to scale up our growth once the financial markets correct again.

Joseph Vafi

analyst
#17

That's fantastic detail. I appreciate that, Dan. Many thanks.

Greg Lewis

analyst
#18

Hello there. Thanks, and good afternoon or good morning to you. I appreciate you taking my questions. I had one question — Daniel, I believe you touched on this and hinted at it regarding the flexibility on spending. As we look at where capex for fiscal year '27 sits right now, it has declined versus the prior quarter, which itself appears lower than the quarter before. So as we consider that flexibility around capex, what is largely behind it, and might you give a bit more detail on why it seems to be trending downward for this year?

Daniel Roberts

executive
#19

Fair question, and it certainly is trending down. We have intentionally reduced it, and that is not out of desire; nor is it because growth opportunities are lacking. Demand exists, customers exist, the sites, the 30,000 acres, the 5 gigawatts of secured power — every ingredient is in place. The reason is purely financial markets and the present situation, and capital must be respected in today's environment, and we won't pursue growth at any cost. We will locate that balanced point depending on funding availability and funding cost, to moderate that growth trajectory. As we pointed out in the deck and our prepared remarks, we don't execute 20-year colocation agreements carrying a 20-year liability. We have not executed $100 billion TCV contracts with AI labs. We have not committed to tens and tens of billions of GPU capex and built this business model in which we must raise money at any cost. In fact, that perception of a neocloud, in inverted commas, strikes me as the direct opposite of our core underlying business model, which was built literally from the ground up: the land, the power, the data centers funded with cash. The strength and resilience of that platform let us ride out volatility in financial markets while being positioned to capitalize when it returns. It's unfortunate, because the demand signals are present. As I said, all those data points are from this week, this is live. What is occurring at the business level is very disconnected from what the market is showing us.

Greg Lewis

analyst
#20

No, that's really useful. You mentioned how strong GPU pricing is, even for the H100s, but I'm curious — could you give any specifics about contract lengths, or perhaps some of the recontractings you're seeing with certain GPUs?

Kent Draper

executive
#21

Glad to add a bit more detail on that. Of course, we don't talk about particular contracts, but...

Greg Lewis

analyst
#22

No, no, a broad range works fine.

Kent Draper

executive
#23

Regarding what Dan mentioned about H100s doubling the price at which they had previously been leased, that additional pricing has now been locked in for a multi-year term. It isn't that we doubled capacity by selling on an ultra-short-term contract. This is genuine long-term contracting for those H100s that has still resulted in a doubling in price. As Dan alluded to, that simply indicates how strong the market is at the moment and the level of demand that exists.

Greg Lewis

analyst
#24

Okay, thanks so much. That was really useful.

Michael Donovan

analyst
#25

Hey, everyone. Appreciate you taking my question. I want to return to what you said about flexing. As you look across your sites, how do you think about ranking places like Oklahoma and Spain when financing is the constraint?

Daniel Roberts

executive
#26

Obviously, we have a portfolio of sites. Our approach and mindset has always been cheap call options—hope that doesn't come across the wrong way. It's really about creating optionality on top of a solid base. What you see is that along the value creation curve, from developing land and power all the way to purchasing the computers, there's an inverse relationship between time and how much money you must spend to hold onto that time to compute that we continually discuss. To get a little more granular on that, keeping projects moving forward costs little, and after that it begins to get pricier. An option on land is comparatively inexpensive. Acquiring the land is comparatively inexpensive. Putting money into a grid connection then costs a bit more. Civils, clearing land, pouring concrete foundations—you can see how capex begins to rise as your time to compute draws nearer. What we've done across our history is make real-time choices about capital allocation in which we maintain optionality around that. Right now construction work is underway at 6 sites, that figure will probably grow in the near term, and then it becomes a question of capital availability. As I've said a few times, we're aware of what's going on in financial markets, but we have plenty of alternatives on the capital side, even past what we've attempted to lay out in considerable detail in our slide deck today. Much of our background is in private markets, real assets, infrastructure funds. The overall depth of capital worldwide, listed or unlisted, I think the fundamentals and what customer demand is showing us will win out, and we're holding many conversations about funding that. My own belief is that this gap between the fundamentals and market sentiment right now is unlikely to last very long. We're ready if it does, but there are several options through which we can keep monetizing that 5 gigawatts, even without an equity market recovery.

Michael Donovan

analyst
#27

Thanks for that. You brought up the collaboration with NVIDIA aimed at speeding up site development. Are you also making use of NVIDIA software to squeeze out additional performance? And if that's the case, what can you share regarding efficiency improvements and possibly a greater number of GPUs being able to be deployed at a given site?

Kent Draper

executive
#28

I'm glad to jump in and offer a bit more color on that. A key reason we're collaborating so tightly with NVIDIA is that their focus isn't solely on the chip side of the business — they're viewing the broader data center as the unit of compute. You hear Jensen bring this up often. It's not just the chips or the servers anymore; it's about how that whole integrated data center unit operates to ultimately generate compute. There are several elements within that. Software is part of it, and we've been examining some NVIDIA software, which is called Max-QPS?, to enable more nameplate GPUs per megawatt than you'd get if you simply constrained the GPUs to the exact power envelope. The reason you can do that is that the vast majority of the time, GPUs are actually running below 100% of their nameplate power draw. Obviously, if you're only sizing to the nameplate power draw, that means you have stranded megawatts much of the time. What their software lets you do is actually oversubscribe the GPUs, and in the rare instances where they're all operating near nameplate capacity, the software will actually step in to make sure that you throttle the GPUs, so you're not exceeding nameplate capacity. That's one example where you can get additional GPUs per grid-connected megawatt. Another example is that NVIDIA is doing a lot of work around the temperatures at which you can operate, and to the extent you're able to operate at a higher temperature, it means you can direct fewer of your grid-connected megawatts towards cooling infrastructure and more of your grid-connected megawatts to the actual underlying IT compute. Those are a couple of examples, but there are numerous other elements that we continue to refine and work on with NVIDIA.

Daniel Roberts

executive
#29

To build on those points, there's a physical infrastructure edge too — the fact that the grid connections and the data centers are ours, plus the chance to optimize past what the software layer allows instead of sitting in the middle between an infrastructure provider and the end customer. That's largely why we've been asked to upgrade clusters: we're able to do it, and we have that execution capability. It's also much of the reason we're securing expansion contracts from leading AI frontier labs and seeing so much customer traction — they trust our ability to execute, since we own and control the entire stack. That data point in the deck captures it well: we delivered Horizon 1 data centers and GPUs — our first ever liquid-cooled data center — in 14 months, against an 18-month industry standard for the data centers alone. The value of that physical infrastructure is really hard to overstate, and I know we've discussed it for a while, but I think this market — particularly at least on the customer side — is really beginning to appreciate it.

Michael Donovan

analyst
#30

That's useful context. Many thanks.

Mike Colonnese

analyst
#31

Hello, Dan and team. I appreciate you taking my question. I'm just wondering, how much of the 500 megawatts of planned incremental data center capacity for '27 is contracted at this point following the recent deals you signed here? I'm curious how we should think about upcoming contract signings relative to phased data center deployments next year as you work to fill the remaining capacity, and how you weigh that against capex back up? Would love your thoughts.

Daniel Roberts

executive
#32

Regarding the 0.5 gigawatts, we haven't given specific guidance on that, the precise makeup, but inside it, it's evident NVIDIA accounts for a portion of that capacity; it's evident another solid credit with one of these leading AI frontier labs is another credit behind that. We have the air-cooled at Childress rolling out in 2027, and then we have VRs arriving later in the year, the latest generation liquid for both Childress and Sweetwater 1. That's where we're especially excited, I think. We haven't brought it up on this call, but the supply of VRs next year is extraordinarily scarce, and we're seeing considerable interest, and that's something that genuinely excites us regarding the perspective economics on it. Most of the near-term activity we're seeing is on the air-cooled at Childress. All right.

Mike Colonnese

analyst
#33

Appreciate the color there, Dan, and good luck with the rest of the deployments, sir.

Michael Ng

analyst
#34

Hey there, good afternoon, and thanks for the question. If I may, I have just two. First, regarding the revised capex outlook, could you speak a bit about the flexibility you referenced, and why doesn't that shift in the capex outlook alter the expectation for data center capacity targeted for year-end 2027? Second, I wanted to ask whether, if some form of community opposition leads to a delay, customers could exit any leases or receive any of their GPU deposits back in the event something caused a substantial delay? Thank you.

Daniel Roberts

executive
#35

Sorry, could you repeat the first half of that question? I caught the 30 community opposition.

Michael Ng

analyst
#36

First, on this year's capex reduction: how is it that capex comes down while planned capacity more or less holds steady? Thanks.

Daniel Roberts

executive
#37

The simplest way to put it is that this reflects the investment behind the 35 gigawatts of development we're pursuing. We aren't doing this for 200 megawatts of Horizon 1, and we aren't even doing it for the 5 gigawatts of secured power; we have a vision and we intend to chase it, and we can flex capital spend up or down, but it comes back to my earlier point about long lead items and staying ahead of the curve. If you were starting from scratch today and hunting for large-scale power, assuming you could find any, you'd comfortably be looking at the 2030s. Keeping that 35 gigawatts moving forward matters. At the same time, in the presentation, where capex carries more weight, long-term development isn't especially material, and you can see we've begun construction across a great many sites. We haven't given guidance for 2028, beyond saying we're building capacity to be available by then, and there's room to flex the cadence of capex spend and the pace of growth up or down across all those sites. When the market shifts the way it has over the last few months, the first thing we do is simply pace that construction and pace that capex spend, holding it back and waiting for market conditions to improve. That's the flexibility of the model when you own and control everything, and we aren't building into 20-year colocation liabilities where we construct assets, and we get to choose when and how to build on top of those assets instead of being forced to double down when markets aren't conducive. On community opposition, I think when we say opposition, our view is that what we're seeing is entirely sensible. AI is an unknown, and I think data centers are the physical manifestation of AI and this big thing happening in the world. I think whether it's Texas, whether it's Australia, whether it's Spain, everywhere, all we're seeing is people wanting to ensure that data centers pay their own way and do the right thing; it feels completely reasonable; we hire locally, we work with local suppliers, we prioritize bringing in the limited water we need for our closed-loop systems from off-site. I think everyone wants these projects to bring lasting jobs and investment to these communities and respect the needs of the people that live there. My view is that we're simply moving through a stretch in which the end-market opportunity for AI, in terms of its size and scale — what we're observing politically and at the community level, I'd say, signals how large that market is and the growth we're moving into, and it's a reasonable wish to ensure that gets handled the right way.

Michael Ng

analyst
#38

Great. Thank you, Dan.

Kent Draper

executive
#39

One more thing I would mention on that point is that we factor these concerns in right from the very start of development, including where the projects themselves are placed. If you look at our whole fleet of projects, you won't find data centers sitting in urban environments where they'd have specific near-term effects on residential areas — construction noise and operational noise. Our data centers sit in regional areas, where the extra economic development these projects bring in is hugely beneficial to those local communities; we cooperate very closely with those local communities throughout — as Dan mentioned, from local hiring and the use of local contractors to community grants programs, and also, when it comes to new developments, community consultation and education. This is something we literally incorporate into our development process from day one.

Michael Ng

analyst
#40

Thanks for that color, Kent.

Jon Petersen

analyst
#41

Oh, great. Thanks. Perhaps for the first question, maybe we'll stay on that same subject. I'm curious whether you could remind us about Sweetwater — I know you have conditional base load approval. Do you still expect to receive final interconnection approval in December? And then, are there any other permits still outstanding, or anything else we should be thinking about when it comes to bringing Sweetwater online?

Kent Draper

executive
#42

To be clear on Sweetwater, Sweetwater 1, our 1.4 gigawatt project, is already connected to the grid. I'm not quite certain what December date you were referring to, but that project is already grid-connected. As we announced a couple of months ago, Sweetwater 1 and Sweetwater 2 were both included in the base load of the batching process. That means all of the capacity — the 2 gigawatts across those sites — is secured; on top of that, we had a number of other projects that also qualified for the batch zero process. As we said at the time, we only announce projects once they actually sign binding grid interconnection agreements, so we plan to announce those projects in due course; we do have additional projects beyond Sweetwater that are included as well. I'd also note that for both Sweetwater 1 and Sweetwater 2, those projects are already well under construction, as Dan pointed out during the presentation. You can see very clearly the status of those, but there is a significant amount of construction work already underway on-site, and we obtain the relevant permits throughout the construction process. Hopefully that gives you a bit more color on the status of each of those projects.

Jon Petersen

analyst
#43

That's excellent. My apologies for getting the dates confused. Here's another question. Could you discuss what it costs to buy new GPUs and what direction that's heading? Do you find GPUs are obtainable whenever you're able to get data center infrastructure up and running? And how do you anticipate those GPU purchase costs will move over the next several quarters? Put differently, what I'm really asking is whether the price of buying GPUs will eventually match today's leasing rates, or whether you think these quite short payback periods will continue for some time.

Kent Draper

executive
#44

It's widely reported, I believe, that supply difficulties persist across the industry—SSDs and RAM especially, with various memory types having been tight for a good while. Other components remain constrained as well; given that situation, we'd anticipate GPU prices continuing to rise as time goes on. Naturally, end-user customer demand is what drives that. Once more, it's fascinating to observe these dynamics in which customer demand is ultimately so robust that it's the very thing pushing hardware prices upward. On the pricing side, what we observe is that any price hikes pass straight through into our hourly rates under contracts. Payback periods are at minimum holding steady, and in fact we've seen some of them shrink during the last six months as that market demand keeps flowing in. To put it briefly in answer to your question: yes, payback periods essentially stay the same as hardware price increases are passed along.

Daniel Roberts

executive
#45

Perhaps to build on that point too—if we sort of jump ahead and consider our strategy along with capital allocation, that's partly why we've intentionally not bought 100% of the GPUs for next year. We view that as right-way risk. If you map out scenarios for where the market stands right now, if financial markets stay weak on sentiment, and if construction and execution keep being difficult throughout the industry, then it's quite possible GPUs will drop in price simply to reflect the shortage of power points. In that case we think it's right-way risk, because if markets bounce back, they begin charging, and in the worst case it's correlated and perhaps we have to pay a modest premium for the GPUs, perhaps we have to wait two or three months, but it returns to this core theme of it's time to compute, it's time for those power points. Putting priority on the physical infrastructure, starting construction at all these sites globally and maintaining that optionality—if markets are strong, we'll be in a position to afford the GPUs, we'll be in a position to afford waiting an additional couple of months on supply, but if markets stay muted, we're going to be very happy that we haven't overcommitted on the GPU capex and that we're keeping platform growth moving along in the background.

Jon Petersen

analyst
#46

Good. I appreciate that extra detail. Thank you.

Daniel Roberts

executive
#47

Thank you, everybody. As we wrap up: Horizon 1 was delivered to Microsoft and carries NVIDIA Exemplar status, achieved 14 months from civils versus an industry benchmark of 18 months. Horizons 2 through 4 are close to finished now. Contracted ARR of more than $4 billion is funded and is now targeted for January, though it may slip into the end of February. During this quarter, we take real pride in AI cloud revenue nearly tripling, and our operating ARR hit $1 billion. Backlog stands above $20 billion, and two-thirds of it comes from investment grade counterparties. As we've said repeatedly on this call, demand has honestly never been stronger. Over the past few days, we've entered into an agreement with a top frontier AI lab to grow the capacity they already have with us. We finalized a new contract running five years at roughly $22 million per megawatt, and that customer is prepaying 90% of the GPU capex. We also signed a short-term deal at $40 million a megawatt, and importantly, on older generation H100s, we just recontracted at twice the original contract price. We own the land, the power, and the data centers beneath all of it, which is what allows us to set the pace. Thank you all for your time. Our next milestone is the 0.3 gigawatts targeted for January. Thank you.

Operator

operator
#48

That concludes today's conference call. Thank you for your participation. You may now disconnect.

Read the full transcript via the API

You're viewing the first half of this call. Get the complete IREN Limited transcript — plus 132,000+ transcripts from 7,000+ companies, speaker segments, AI summaries and full-text search — through the EarningsCalls.dev API.

Get the API View API docs →

This call discussed

For developers and AI pipelines

Programmatic access to IREN Limited earnings transcripts and 132,000+ others is available through the EarningsCalls.dev REST API. Plans from $39.99/month — full transcripts, speaker segments, full-text search, and the recently-added /api/v1/transcripts/recent polling endpoint for ETL pipelines.

More tools for stock research

OptionsBell: an unusual options activity scanner that tracks options flow across all US stocks and emails you when large trades appear in IREN Limited or any ticker on your watchlist.

StockMarketScan: a stock screener with alerts: 24 pre-built screens refreshed every 5 minutes, watchlists and email notifications. Both offer API and MCP access and fair monthly pricing.