All right, we are continuing today at the sixth annual Canaccord Digital Asset Symposium. Once again, I'm Joe Vav, the equity research analyst here at the firm. And so we are wrapping up our day one with a very important fireside chat, at least relative to my research coverage. We have with us Dan Roberts, co-founder and co-CEO of IREN. So, as many of you know, IREN is an energy infrastructure company founded in 2018 and has been on an exciting journey ever since its founding. First, the company was at the tip of the spear in acquiring a large power portfolio across British Columbia and then West Texas, after which it stood up one of the largest Bitcoin miners. And then, as the Chinese proverb goes, luck favors the prepared mind. IREN is now embarking on an exciting next chapter, pivoting its large power portfolio from Bitcoin mining toward the broader AI opportunity. To date, the company's focusing on building its own GPU cloud services, having inked a landmark large deal already with Microsoft and continuing to expand its GPU acquisitions from here, targeting installation of approximately 150,000 GPUs by the end of 2026.
That GPU fleet could generate annualized run rate revenue of almost $4 billion. The company has over 2 gigawatts of uncommitted power at its West Texas facility known as Sweetwater. And so the next year should be filled with exciting updates, just as we've seen in the last year as the company has evolved. The company's got 810 megawatts of operating capacity and over four and a half gigawatts of secured grid-connected power, and IREN has one of the largest power portfolios out there with a number of options to monetize it. So with that not-so-short intro, we are delighted to have Dan Roberts with us today. So thanks for being with us, Dan.
Thanks for having us, Joe. Pleasure.
So maybe I think a lot of people know the IREN story, but let's start with a high-level question. Where are we in IREN's journey right now?
Sure. Joe, at a high level, we're building one of the world's largest AI cloud platforms. We're scaling towards roughly 150,000 GPUs across our data center portfolio this year, as you mentioned earlier. But what makes IREN different is that we're vertically integrated. So we control the full stack: land, grid connections, substations, data centers, and all the way down to the GPUs. And importantly, that gives us speed of delivery. It gives us control over costs and construction, and gives us a lot of flexibility for customers. The platform more broadly is underpinned by what is now more than four and a half gigawatts of grid-connected secured power across the US and Canada. So these are power connections backed with binding connection agreements rather than development sites. We've got plenty of them. And that platform is again underpinned by in-house engineering, procurement, construction and operations team. So when Will and I started this company in 2018, we started it around a very simple thesis, which is the digital world would grow faster than the physical infrastructure needed to support it. And what we're seeing today with AI is that thesis playing out in real time.
It's growing much faster. There's no doubt about it. You know, we're seeing that almost on a monthly basis, it feels like. So I know you mentioned, Dan, you know, your vertical integration. Clearly not everybody is doing that. You know, other cloud providers are, you know, are taking leases with, you know, data center providers. They're, you know, they're basically, you know, they may be somewhat vertically integrated, but not fully vertically integrated. Can you just drill down a little bit on what owning the full stack means and, you know, what kind of competitive advantage it provides, maybe not just in the marketplace, but, you know, ultimately in financial performance coming out of IREN?
Sure. So I guess if you're an AI cloud provider reliant on colocation without the security of land and data centers, there's a number of challenges, and it's not insurmountable by any stretch, but it's just additional moving parts that you need to deal with. So first and foremost, in order to matchmake customer revenue contracts with compute, you need to also bring in line in parallel the access to the data center. So colocation contracts, and that can be easy sometimes in a market that's well supplied, but in today's market, which is demand constrained, the ability to secure colocation contracts on timelines that make sense, on a risk profile that makes sense, because we are starting to see execution blowouts across the sector, that causes challenges. So for us, having full control over the whole stack, keeping our destiny in our hands, is really important. It's really important to Will and I. We don't even outsource construction to a head contractor. We manage it all in-house. Yes, it's a bit more complex. Yes, we've got a few more moving parts, but to us, it's much easier to have full control over the whole stack rather than, you know, at an extreme level, putting your head in the sand, signing a contract, and hoping that it appears one day because this sector is just moving so fast, so quick, the stakes are so high that we just want that absolute control. And I think over the journey over the last five, six years, Joe, that we've been on together, you've seen that we have been able to deliver under that model on a lot more predictable basis than perhaps the sector generally.
Sure. And asset values may not show up in the P&L, but, you know, those vertical integrations clearly have value that is enduring. So that's great.
Well, and I think maybe sorry to jump in, Joe, but I think that's also important because you think about the structure of this industry over time. So in five years, these cloud contracts roll off. You know, people used to talk about GPU cycles of three to four years. It's probably well in excess of five now, potentially six, potentially seven, eight. But the point is, at the end of that cycle, if you're sitting there with ownership of the underlying asset, you've got an asset there that you can monetize again into another cloud contract. You can monetize it into colocation. You've got a lot of flexibility, but the point is you've got an asset at that point in time. If instead you've got a colocation contract, you've got a liability. So the pressure is really on to utilize that lease that you've signed up for and find a use case for it. So it sets us up for a very different future as this sector evolves.
Sure. Absolutely. So you are taking a little bit of a different path than some of your miner peers that are making the pivot to AI. Most are mostly focused on colocation. Like you said, you're taking the full vertical strategy all the way up into owning your GPU clouds. Maybe you could kind of give us, you know, a little more insight as to why you're not stopping on colo like the other guys are.
Yeah, look, in some ways it'd be a lot easier, right? Just build a powered shell and hand over the keys and see you in 15 years, but we just don't feel like it's the best value-creating opportunity for us and our shareholders. I mean, power is the scarce resource in this industry. Access to compute is the scarce resource. So for us, the key question becomes how do you maximize the value of each incremental megawatt that you can bring online? AI cloud is harder, undoubtedly, it sits much higher up the value chain. So in our view, it can generate roughly $10 million a megawatt in prospective revenue versus around one and a half to two million a megawatt in colocation. So to illustrate, if we'd signed a 200-megawatt colocation contract with Microsoft, it might generate around 300, 350 million a year in annualized revenue, but instead it's closer to two billion per year. So the advantage of owning the infrastructure is optionality. We can run cloud today, we can shift to colocation tomorrow, or we can blend the two depending on what creates the most value. But for us, the idea of signing a lease agreement, essentially selling your site for 15 plus years, and then looking to arbitrage, you know, cost of capital with REITs and, you know, sign a deal at 8%, sell it off at 6%. For us, honestly, it feels like picking up pennies in front of one of the biggest generational steam trains in terms of industrial buildouts. And for us, we just want to compound those returns and deliver the world with compute.
Sure. Fair enough. A lot of opportunity, a little more work. Not sitting back and just clipping a coupon for 10 years. That makes sense. Your landmark deal with Microsoft, signed, I think last November, was it? Just late last year. Could you give us an update on progress there, how things are going, maybe some of the things you've learned along the way so far? And then maybe what to expect from it during the rest of 2026.
Sure. So just to recap, the Microsoft deal is a 200-megawatt IT load, tier three data center deployment with tens of thousands of next-generation GPUs, where we essentially recover the majority, if not all, of our capital over the first five years. So over the life of the contract, we expect to generate close to $10 billion of revenue with really strong EBITDA margins. Construction is progressing well. The data center buildout continues to remain on schedule. In terms of final commissioning dates for each cluster, a lot of that is driven by GPU delivery timelines rather than facility readiness, but yeah, we expect it all to come online through the course of this year.
Great. So it sounds like it's on schedule. And then you've just recently announced further buildout of your GPU fleet beyond just the Microsoft deal and beyond some of the other GPU cloud services you are already providing. Can you give us, you know, maybe an early sense of, you know, what prompted you to go ahead and make the commitment to continue to expand your GPU fleet? Clearly, you must be seeing, you know, good demand in the marketplace to support this, and maybe if you build it, they will come. Just kind of some thoughts on, you know, staying pedal to the metal on GPU acquisition.
Look, I think that's right. I had this conversation the other day on a Twitter space. I'm not aware of any GPUs that are available in a data center today that are sitting idle anywhere in the world. Like the whole reason we are on this call, the whole reason this sector is being talked about, because we've got this crazy demand-supply equation where you just can't bring online cloud capacity fast enough. And the key factor that we're seeing across the industry everywhere is this concept of time to compute. And I know a lot of people have been talking about time to power for quite some time, but for us it's all about being time to data centers, time to compute. You know, customers increasingly care about how quickly capacity can be deployed once an agreement is reached and what the risk profile is to actually deploying that capacity. Because as we mentioned earlier, there are a lot of contracts being signed, but it's very different between signing a contract and actually generating revenue from that contract. There's a lot of things that go wrong. So, look, for us, it's just about derisking everything by securing the GPUs early. We reduce supply chain risk. We reduce cost exposure, which we are seeing on the GPU side, particularly memory. We're accelerating deployment timelines. We're also seeing strong interest in air-cooled GPU deployments, which can be brought online faster, which plays directly into our strengths. Like we operate 31 air-cooled data centers right now across Childress and Mackenzie, which we can scale into using air-cooled GPUs. So it's exciting. The final point is capital efficiency. So those particular GPUs have a 30-day net payment term, which means that payment isn't actually due until 30 days post shipping. So there's no cash outlay until later this year. So look, the decision was driven by a combination of, you know, the demand, the speed to deployment, but also a well-structured procurement process where there's no cash down until later this year.
That's great. Yeah. Maybe we talk about the financing on the GPU side. It's clearly evolved pretty fast. Financing is a key part of this buildout. It doesn't happen without, you know, some maturing markets around it. Can you give us an update on some of the financing markets, leasing, other financing options in the GPU market? And, you know, where do you think it, you know, are we going to hit a capacity constraint on the, you know, on the financing side? You know, people talk about constraints on power, but maybe the constraints on finance, I don't know.
