IREN Ltd (IREN) September 2026 Conference Transcript

Great. Good afternoon, everybody. Welcome to the IREN Fireside Chat at the Goldman Sachs Communicopia and Technology Conference. My name is Mike Ng, and I cover IREN and data centers here at Goldman. And I have the privilege of introducing Dan Roberts, who's the co-CEO of IREN, and then Kent Draper, who's the chief commercial officer. Dan, Kent, thank you so much for coming out to the conference. It's a pleasure and a privilege to have you out here.

No, thanks for having me. Sorry I'm a bit late. I think the schedule you've put together.

You know, to start off, I was just wondering if you could talk about the vision and some of the strategic priorities for the company. IREN, obviously an AI cloud that owns both the data center, LAN PowerShell, as well as the IT hardware, right, servers, networking, GPUs. The company has signed $4 billion of ARR with a range of marquee customers, Microsoft, NVIDIA, Coher, Prometheus, Propoxity, and Figure AI, to name a few. But to start, we'd love to just hear about your strategic priorities and the vision for the company.

I think if you had to boil it down, we think the real world's really going to struggle to keep pace with what's happening in the digital world. The real world is harder the bigger the scale gets. We're already starting to see delays come through, and I think the next 12 to 18 months, this industry is going to be defined by execution and delays. Whereas if you look at what's happening on the demand side for compute, it's only going up. So you've got this wedge of kind of growing exponential demand, but the real world's getting really hard, whether it's supply chains, whether it's political, social, regulatory pressure. And I think the heart of our business model is capitalised on that growing wedge. And yes, that was the thesis that started the business eight years ago and I think we bought our ticket to this industry by getting ahead of that curve and locking up a lot of land and power so again to recap for people we've got five gigawatts or over that five gigawatts of secured land and power globally so this is signed binding grid interconnection agreements the lands done we own it all we've got multiples of that under development but we're big believers coming from a development background that it's a one and zero you can talk about all these made up megawatts in a presentation but if you don't get a signed interconnection agreement it's it's a donut and we've seen that time and time again um and what that's done is really given us the ticket to play in this kind of time locked window where you can build on top of that you can build the data centers you can buy the compute you can start doing software on top of the compute and really have a crack at becoming a very large provider of compute to the world and i think that mentality and that philosophy has underpinned all of the decisions that we've made today and continues to be the North Star. Great.

Execution is critically important, not only to prove out the ability to execute against your pipeline, but I think it also helps with attracting customers. And the Microsoft deal and what you've done with Childress, I think, is a good example of that execution. So I was just wondering if you could talk a little bit about that 200-megawatt IT deal, $1.94 billion of ARR, a five-year contract. The first phase, Verizon 1, was delivered and then accepted by Microsoft last month. I think the three remaining phases are set to be delivered in the calendar fourth quarter. So would you just talk a little bit about the execution on this Microsoft deal and talk about the deliveries and learnings of Verizon 1 through 4?

Yeah, I'm happy to maybe start on that. So, yes, we handed over the first phase to them. Phases two to four are exact replicas of the first phase. So, from that perspective, as the market looks at it, a significant de-risking on the execution front. But the reality on our side is we've been building our high-performance computing data centres for eight years continuously now. We have a very strong track record of execution across our entire history as a business. And certainly since we listed, we put out a number of targets and continued to hit that. So it's nice to have that official proof point there now in the market that people can see. But, you know, we're very proud of our execution track record over a long period of time now. And as Dan mentioned, I think it is going to be a major differentiator within this industry. You know, we are still very early in the cycle where you haven't really been able to determine yet who can execute and who can't. And that is going to be a huge differentiator as we move forward. So, yeah, it's, again, nice to get the early runs on the board there as it relates to Microsoft, but it's just a continuation of what we've been doing for a long time now.

Maybe we can just stay on Childress, which is where Microsoft is. In the 2027 outlook, I think the expectation is for Childress to expand to include 250 megawatts of gross air-cooled capacity and 150 megawatts of gross liquid-cooled capacity with Horizons 5 through 6. First, do you have GPU leases for capacity at either of those opportunities, sites? I think the NVIDIA 45-megawatt IT deal is using some Childress air-cooled retrofit capacity, but I just wanted to confirm that.

Yeah, that's right. So in terms of overall delivery of new capacity in 27, 500 megawatts of IT load, there or thereabouts, of which the NVIDIA contract represents just under 10%, and the remainder is uncontracted as of today.

And then second, could you talk a little bit about the liquid-cooled capacity timeline for 2027? We've had some questions from investors about why the liquid-cooling deployments are shifted more towards the back end of the calendar year. Is that due to the timing of servers and GPUs? Is it related to data center construction? Just any thoughts there would be helpful.

Yeah, largely related to data center construction. So the air-cooled capacity that we have, it's a very well-proven model now for us to be able to retrofit that for AI purposes and we can bring online that capacity very quickly and that is very advantageous to us to be able to offer capacity throughout the back half of this year and the first half of next year. And then in terms of the liquid-cooled build-out, it is a longer build-out timeline relative to retrofitting the existing data centers. So the second-half capacity that will be coming online that's targeted at Vera Rubens, that is just when the data centers will be ready to accommodate the GPUs.

I would love to just talk a little bit about everything that's happening in Texas. You know, I think over 90, maybe 95% of the capacity that's coming online for IRN is in Texas. So could you just talk a little bit and give us an overview about ERCOT's batch zero process? And then if you could just share, you know, the company's perspectives on the governor's mandate for the data center audit, you know, what implications that may have on, you know, Texas data center energization, if any. and IRN Sweetwater's campus was successfully included in the baseload. I think that was announced this week, so maybe you can touch on that milestone as well.

Yeah, as you mentioned, we put out a release yesterday morning confirming that we were formally notified that Sweetwater's two gigawatts was included in batch zero baseload. That's a position that we've had a high degree of confidence in and we've expressed that over a number of earnings calls recently. So not a surprise to us, but nice to have that formal acknowledgement. I think what's probably more exciting for us is we had a number of additional projects and additional capacity that were included in the batch zero process. As Dan mentioned, we don't announce capacity to the market until we sign binding grid connection agreements. So we expect to announce that capacity over time as we sign those agreements, but we did have substantial amount of capacity included in batch zero over and above the two gigawatts at Sweetwater. In terms of Governor Abbott's directive, and for those who aren't aware, he's asked ERCOT, which is the grid operator, to look at the various projects that are included in batch zero and audit them from a perspective of a number of key criteria, including transparency, impact on local community, resource usage. And this is something that we actually welcome in terms of the overall approach. We have always developed these assets with social licence to operate in mind. We think that is incredibly important. And across the various metrics that they'll be looking at, we think our projects will measure up very strongly. We don't operate in urban environments. We don't have residential neighbours at our sites. We won't be impacting them from a noise or construction perspective. We use very minimal water in our data centres, given our very efficient closed-loop cooling system. We utilise 100% renewable energy. And as it relates to the transparency question, I mean, one of the reasons why he actually came up with this directive was ERCOT had asked data centre participants about their resource usage and most didn't respond to the survey. We have responded because we have nothing to hide. And so, you know, overall, we welcome the context of it and think that our projects are very well positioned.

I know you mentioned that you guys always had a high degree of confidence that the Sweetwater campuses would be included in batch zero baseload. I was just wondering if you could just share whether all your prospective customers have the same level of confidence and if being included as base load designation, does that help your leasing prospects from here on out? Or was there never really a doubt in customers' minds?

Look, it's probably one element that gets taken off the table a little bit in terms of customer due diligence, but Sweetwater One is already grid-connected. We connected a number of months ago to the grid. We've already begun the data center build out there, so we had a high degree of confidence, enough to actually start physical construction there. And I think customers largely have taken a lead off of that in the historical discussion. So no impact in terms of the commercial conversations that we're having.

I think the commercial conversations are more dictated by the universe of customers that are available at every point in time before a cluster is commissioned. So if you think about data centres that are 12, 18 months away from being commissioned, the universe of prospective customers is really small, right? You've got the hyperscalers instead of some of the AI labs that have the demand outlook, the balance sheets to be able to forward contract. They're also the same people that are able to self-build or enter co-location agreements. So this is why every second day you're seeing $20, $30, $40 billion deals being announced. Those guys are able to do that because they just take a view, well it's cheap compute if the counterparty delivers happy days but the reality is every month closer to get to commissioning that circle of prospective customers gets bigger and wider and if you look strategically again back to our north star where we want to be we don't want to wake up in five years and have six customers we want to have 60 or 600 and the ladder's not off the table right with the enterprise pursuit that we're going through with mirantis going down to that layer of software. And we just believe that's going to deliver a far more valuable and resilient business model over time. Great.

I mean, that's a great segue. Maybe we can talk a little bit about the Mirantis acquisition, you know, how it opens up a new customer class and I think enables more on-demand compute. You know, what's the timeline for Mirantis really opening up that funnel, or maybe it's happened already, and has Mirantis, you know, helped in any of your current negotiations for new prospective customers?

Short answer to the last one is, yep, absolutely. But maybe just take a quick step back and give more context for everyone. So we view this business as three layers. So again, the ticket to the sector that we outlined at the start of this chat is getting that land power, the data centers. Now, once you've got that, you've got a decision point. You can become a REIT or a data centre provider earning a real estate return and start arbitraging disc cap rates, etc. And that's a good business for some. Or you can buy the computers and plug the computers in and now start servicing a much deeper revenue pool and customer base. So instead of earning $1.5 million a megawatt, we're earning $20, $25 million a megawatt and not even having to pay for the CapEx difference, right? We financed our latest GPU deployment, 90% gear in plus 50% prepayment, so 140% of the GPU costs we actually got up front. Anyway, the context is the final layer on top of that is software and becoming even more vertically integrated into the end markets for compute, right? And this is where Mirantis pops up. Software's the easiest thing to solve at the end. What you can't solve is five gigawatts of powered land and data centres. If you're starting from scratch today, and you want to get 500 megawatts worth of data centres, see you in 2032, if you can find it. Like, it's real. Like, ERCOT just won't take more applications. Like, everything is chockers. So we've got this time-locked structural advantage. So every decision along the way is how do you compound that into a greater profitability over time? And this is where we get really excited by Mirantis and their capabilities. So first of all, they've brought a lot of capability to just operation of the bare metal clusters. Tick. Helpful. But really strategically, they've built their whole business off the last couple of decades servicing enterprise private CPU clouds. So they service 1,500 enterprise clients like Visa, NASA, all the biggest names, servicing their requirements. So by buying them, we've just bought 15 years of track record, credibility, the security. and part of the justification or the rationalisation from their perspective to coming to us was their clients want private GPU clouds. So with the emergence of open-way and open-source models and all the proprietary data that corporates and enterprises retain internally, the ability to navigate AI use of all that data within a wall-secured garden is becoming really important to them. So Miranda's like, well, how do we get security of GPUs? Like, it's hard to get GPUs. The strategic rationale for merging with us was to get access to the GPUs. Strategic rationale for us to go in there is to increase our profitability. And one last point on the profitability thing, we are actively selling to intermediaries at the moment that overlay software and provide inference capabilities to end customers, where they are selling our compute for multiples of what they're paying us. So the last announced deal we had was $23 million a megawatt as a revenue line. they're selling our compute for multiples of that to their end customers. So that's the opportunity, and that's why we're so excited. And when Kent says 90% of our 2027 capacity is uncontracted, the whole challenge for us is how long can we wait? Like, can we hold out? Because every month that goes by, the profitability, the buyer pool gets bigger and bigger on that uncontracted portion, and the opportunity gets bigger.

And that kind of class of customers that you're referencing that are reselling the compute, that's presumably like a Fireworks or like a Together, and now you get to do what they're doing enabled by Mirantis. Is that fair?

It will take time, but exactly that's directly where we're going.

Super interesting. I guess on the pricing point, you know, lots of focus on that, I think, by the market as a whole. You know, at the earnings call, you know, you talked about 2027 capacity being actively discussed at something closer to $25 million per megawatt, certainly well above your previously signed deals. So could you just talk a little bit about the pricing dynamics that you're seeing in the market? You know, how important is duration when people are quoting these prices in terms of, you know, near-term versus long-term, right? Like, you know, how important of a nuance is that? And then, you know, how much of a difference is there when you think about pricing across, you know, chip families, right? Is every megawatt created equally?

Yeah, so the $20 million to $25 million that we referenced, that is for three-year deals. So I think you have very, very strong economics in the context of still signing what are relatively long-term contracts. Since we signed our last deal at a $23 million per megawatt basis, we've had discussions in excess of that for three-year contracts, as well as a number of discussions looking at shorter-term contracts for even further on than $25 million, and it just depends how short you go. If you're looking at one-year or six-month contracts, you can be earning $40 million a megawatt or upwards from there. And then there's all the way down to, again, what Mirantis unlocks for us is ability to offer a true on-demand type service where you're selling at multiples to what the bare metal megawatts are selling at. So I think extremely strong contracting position for any capacity that we bring to market. There is still multiples in terms of the demand that we're seeing for that capacity. And that has allowed us to be in a fortunate position where we're able to be strategic about the customers that we choose. to Dan's point to be able to build up a future customer base. And I think the recent set of customers that we've signed up is a testament for that. I mean, it really is some of the leading names in the industry. Foundational AI Lab, we can't name them, unfortunately, but, you know, it's one of two, so people can have a guess. Prometheus Figure AI that live in the mechanical AI world, robotics, manufacturing, other model builders, Perplexity, Cohere, these are all leading names in the industry that we are expanding our customer base and getting very attractive long-term economics at the same time. In terms of the differences that we're seeing between GPU class, I mean, we kind of see those three-year deals coalescing around a two-year payback revenue period. And actually, the most recent contracts we've signed have been lower than two years. So, you know, that's been a significant improvement overall where we were above two years, close to it, but above two years previously. And we're now seeing less than two years. So, you know, we are continuing to see improvement in economics over time. Great.

Super interesting. You know, just taking everything that was discussed and just thinking about, you know, what the potential 2027 ARR could be, you know, is it reasonable to think, hey, we've got a rough $25 million per megawatt. I think you've got, you know, an incremental, you know, 500 megawatts IT coming online in 2027. That would imply, you know, $12.5 billion of additional ARR. Does that make sense? You know, how are you thinking about the range of outcomes and maybe the risks upside and downside on pricing and also capacity coming online next year?

So directionally, you're right. But the challenge with providing guidance is if we sign six-year contracts for all of that capacity, the number's going to be lower than what you just outlined. But if we signed 18-month contracts at commissioning, it's probably going to be higher. So how that contracting strategy evolves will determine the guidance that we're able to give at the right time. But the other moving part in all of this is capital markets and financing because that is evolving at the speed of light. And we all know that value creation is sources and uses. You have the best business in the world generating the best returns on capital, but if you can't finance the capital, it's useless. So you look at rewinding 12 months ago when we signed the Microsoft contract. It was great because it was a five-year investment grade offtake. we were able to secure 6% money, debt, against that for 96% of the GPU costs, including their prepayments. What a cracker. But three weeks ago, we announced a GPU financing that had no investment grade offtake. That was unheard of 12 months ago. So again, we raised $2.8 billion, 9% interest rate against just AI lab credit. And that's 90% of the GPU, plus we got 50% prepayments from those same AI labs. So if we're sitting here and we say, look, investment grade's the only way we're going to get the capital, then of course you're going to sign long-term off-take agreements with investment grade counterparties. But the way the sector's moving, the availability of capital, particularly in the credit markets, gives us the ability to pursue a more profitable, probably positive NPV business. And yes, we had to issue a lot of shares post-IPO as we accumulated this land and power base and the five gigawatts and built the data centres. But that's throttling off. Like, if you look at the last 12 months, we raised $19 billion in capital, $2.8 billion with equity. Like, the capital markets are coming, and we still haven't securitised any of the data centre infrastructure that we own yet. So that capital market flywheel, I think, is going to really start supporting our growth and optimisation of the customer strategy.

And I would love to talk about CapEx and funding, and maybe just starting out with CapEx. The company is guided to $25 to $30 billion of CapEx in the next fiscal year, fiscal 27. That's a fairly meaningful step up from the, you know, greater than $4 billion in fiscal 26. You know, first, could you just talk a little bit about, you know, how much of the CapEx is related to, you know, 2026 capacity build out versus, beyond and naturally that'll require discussion around how much is GPUs versus data center land power shell.

Yeah, so some of that CapEx does relate to future year build out. We're obviously in an industry that has long development cycles and so some of the CapEx that we're putting out of the door today will relate to 28, 29 capacity but what we've picked up in that number is our are CapEx both on the data centre side and GPUs through to fiscal year-end 27. In terms of the sources of capital against that, so you mentioned $25 to $30 billion in terms of uses. The sources are $14 billion of cash on the balance sheet and committed debt financing today. So that's done. We conservatively forecast $8 billion of additional debt against GPU CapEx in the future. And I say conservative because, as Dan mentioned, we've recently gone through this process and we were able to obtain 140% of the GPU CapEx, but we're assuming a much more conservative level in that $8 billion. It could well be in excess of that. And then the remainder, which is a few billion up to $8 billion in terms of the differential, there are a variety of different sources available to us. So we have used the convertible notes market in the past. As Dan mentioned, we don't have any debt against our data centre assets today, and that is billions of dollars of tangible infrastructure assets in the ground. So that's an area of financing we can draw on. There's other forms of corporate secured and unsecured debt that we can draw on. So, you know, we feel like we have great optionality as to how we approach the funding side of the equation. And as Dan mentioned again, you know, the equity portion is becoming a smaller and smaller piece over time as we're able to recycle capital and get that flywheel going.

It's probably worth just clarifying as well that we are unique as a business. if you wanted to benchmark us to other businesses where our listed company essentially owns two separate businesses. One is Iron Data Centres, which builds, own and operates data centre infrastructure and enters into long-term co-location agreements with Iron Cloud, which is essentially the NeoCloud business. So we are vertically integrated. We've never had third-party leases. We've never dealt with anyone else. We own and control the entire stack from the dirt, the concrete, the steel, the racks, the servers, absolutely everything. And the way Kent's outlining the financing is the data centre company has been equity funded for 100% to date. But it has these co-location agreements with a neocloud, iron cloud, over here. So you think about any other data centre business financing against a co-location agreement with a neocloud, they would raise financing against it. So for us, that just hasn't been the priority because it hasn't been the bigger sums of money, but that's a real unlock for us in the near term.

I guess, like, how aggressively will you pursue that, you know, data-centered, secure financing? And then maybe just a related question, do you expect to utilize the remainder at the market agreement? I think there's $3.5 billion left on that.

We're pursuing the data-centered financing. Spectrum of what's aggressive. like we're pursuing it and I expect to get it I mean we are pretty risk adverse Will and I as co-founders a lot of our money is in this we're not interested in gambling on leverage like we know we're on a sure thing and a great bet so you've got to be sensible about how you do it but I feel confident there's a good product and we'll be able to do some issuance in the near future on that front in terms of the ATM I mean it's really a function of growth and de-risking right because we've got five gigawatts of secured power that's three and a half gigawatts of IT load multiplied by your 20 25 million per megawatt so you can get the calculators out you can you can see there's a big potential for this so a lot of it is is the market going to back us are they going to reward us and I think we're in that inflection point at the moment where people are going where's iron at we did 71 million in AI revenue last quarter it's like are they real and you're telling us you're at a billion ARR now and you're going to be $4 billion in a few months, gee, this is going to be interesting to see how it plays out. The credit markets are absolutely supporting us now as we're seeing through these instruments. So I think the ATM is there, equity is there. We'll almost certainly raise equity at some point in the future, but are we reliant on it for a growth trajectory? Arguably not anymore.

I was wondering if I could ask about the personnel, the executive team. You talked about hiring five new C-suite executives. Your headcount has tripled. So maybe strategically, where is the most personnel and headcount investment happening?

And how should we think about the growth directory beyond next quarter or the quarter after and whether or not you see opportunities to continue to grow the company from a personnel perspective yeah absolutely i mean you've got to hire ahead of the curve and you see the curve right the four billion exit revenue in a few months time you've extrapolated the numbers for the end of next year being multiples of that you know whether you spend a couple hundred million bucks on headcount like it's kind of becomes immaterial in the scheme of where you're trying to get and the reality is we're in a really good place now where people see that we're delivering and we seem to be a destination choice for people wanting to join and they're coming out of hyperscalers, they're coming out of other Neo clouds, they're bringing their entire teams across. At the end of the day, I don't know everything. I know very little, but I know that I don't know much. And you're bringing in people that understand this business, have been doing this stuff for the last 10 to 20 years, bring the systems, processes, policies, the automation, because we are turning into a really, really big company and we need people that have done it before and know all the tricks. So, yeah, headcount tripled over the last 12 months. It'll probably triple again over the next 12 months, but the key is to keep hiring the right people in the right roles and set them up for success.

I mean, in closing, maybe you could just talk a little bit more about priorities, key areas of execution and milestone that you're focused on over the next year.

Yeah, I mean, execution itself is the big one. As we've spoken about here, the financing side looks great. the customer side, the demand picture looks incredibly strong. We've railroaded the land and power side of things, got a very large secured portfolio of land and power. So now it's down to execution. We've had that first proof point with delivery of the initial phase of the Microsoft contract, and we'll look to continue our strong track record of execution from here. But as we do that, you know, you can do that back of the envelope math and sort of get to the numbers that you're seeing very easily because it is that execution piece that's key. And as we bring capacity online, there absolutely will be a revenue line for it. Great.

Well, Dan, Kent, thank you so much for participating in our conference, and it's been an absolute privilege to have you on stage here.

Thanks for having us. Thank you, everyone.