IREN 8.3.26
Analyst 2
I think so. I did do a lot of work on CoreWeave because we owned Core Scientific. I came to the same conclusion. I thought they should have completed that deal. They should have agreed to a different price with Core Scientific and gotten a lot of their underlying power. But when they didn't, it wasn't a company I was very interested in owning because, in the end, it will be more difficult to be competitive if you have someone else taking margin.
And the other bullish thing about IREN is if you just kind of understand at least where their original assets were. Obviously they're expanding in Australia today, and they've announced Spain and Oklahoma. But a lot of where they are in West Texas, that stranded power typically has very low rates because there's a lack of transmission to get that renewable power from where they are to the rest of the grid useably. In addition, before the complete parabolic turn up in demand from the hyperscalers, if you spoke with AWS or Azure or particularly GCP, they really did want renewable-backed power, and IREN has that. They've got solar-based power. So it's got to be the most attractive power portfolio in a good place. Before, everybody wanted to be in Northern Virginia. Now it pretty much seems like West Texas is going to be, if not the largest, the second largest data center site in the continental United States. So it seems like where their power is, and the type of their power, is very advantageous.
And so the hyperscalers should actually be willing to pay up a little bit for that. That should make their bare metal deals somewhat more attractive than others. So that was another point on the power side. I completely agree with you. It's one of the reasons I own the stock.
Analyst 1
One thing which was interesting to me was around how aggressively they're trying to increase their power portfolio. This time last year, they didn't have the Oklahoma site. Even before that, they were at 3, 3.5 gigawatts compared to almost half a gigawatt of actually contracted power. Do you have a hypothesis of why they're trying to chase to get to as big a power portfolio as they can right now, where they're not constrained by power to grow their business?
Analyst 2
I mean, it is a little funny, right? Because clearly they can't finance this, and so you do kind of wonder, why do you go ahead with it? I think they truly believe the bottleneck is power. Therefore, if they're able to create this portfolio, they'll get paid for it. I do think they'll do some colocation deals. They talk about the different risk-weighted IRRs and cost-benefit analyses and think that owning their own cloud is better. But clearly, if they're going to secure this much land and this much power, they're not going to be able to finance it all, even if they get Microsoft, Azure, and GCP on the other side. So it is interesting. I just assume they're going to use some of it for colocation deals, and they think it's a better way to finance growth in their own cloud as well.
Analyst 1
The colocation is very interesting because now Nebius has also come out with their own proposal, essentially clarifying that they're interested in colocation deals as well, and just not bare-metal or full-stack type of deals where they own GPUs. Historically, I can't say anything historically because it's been like a 2-year history of this entire sector. But we're seeing that Nebius is laying the groundwork where CoreWeave has sort of laid the groundwork in terms of what you can do with financing. Nebius has done that with what you can do with the software stack and the business model. It's very well possible for IREN to go that route as well. I think that can be a much better way to monetize that because, if you're thinking about the bottleneck and people are willing to pay up to solve that bottleneck, it makes more sense for someone to do a colocation deal with IREN and get their own GPU and train their own things. Now there are companies like Swarm, etc., which are also doing something even before colocation where they're just providing the land and a shell, and the hyperscaler is expected to get their own electrical equipment and build the data center as well.
Analyst 2
But I think it's almost like you sign a colocation deal, you raise high-yield or secure debt based on that contract, amortizing it down. Then at the end of that contract, or near the end of that contract, you could take out debt on that asset that's backed by that data center, which frees up cash elsewhere on the corporate balance sheet because cash is fungible. These guys clearly come from Macquarie (MQG.AX); they're thinking about the cheapest and best way to finance themselves, which is why they're doing the leases, why they're going to do some contract-based lending, whether they'll do some off of the actual data center and just bankruptcy remote project financing. I think that's part of it. I'm guessing that Nebius has figured that out as well, and that's why they're also open to doing some of those colocation deals, even if both of them in the end really want to be a cloud operator.
Analyst 1
The cloud operator is where I don't know what the vision for IREN is. Do they actually want to be a cloud operator? I don't think that's the case. I'd be surprised if they actually want to be more power operators, compete with the legacy cloud providers. I think they want to play in the power space and just try to extract the maximum value that they can for their power assets.
Analyst 2
Although, I mean, I think they know that if they only have power and they only have colocation, they're not going to be able to extract as much revenue per megawatt as they otherwise could. Otherwise, why would they have done the Mirantis acquisition? And when you talk to them, today they don't have spot pricing, right? It's all contracted. But they believe that with Mirantis they should over time be able to offer spot pricing. They've been asked in the past by some of their counterparties why they contract everything out. You see the spot rates are so much more attractive than contracted rates, and they can retort that it's in theory safer. It's also more financially for them, obviously. It's safe, but if your underlying customer who signed a 5-year contract goes bankrupt or has to downsize, it doesn't really matter if you have a contract or not, right? So if you're able to capture some of the on-demand spot market, there could be a real reason to do that. At least when I speak with them, and with other people who know the company, it seems like they do want to have at least some sort of software solution. It doesn't make them a full AWS cloud or GCP cloud. I don't think they ever have those ambitions, but they want to be able to serve AI startups. They want to be able to serve AI enterprises. They don't want just to do the bare metal, obviously.
Analyst 1
So yes, that's a good signaling mechanism that they did with their Mirantis acquisition, but I don't know if there's anything more than signaling that they want to do more of a full-stack cloud operation. When we're thinking about that, the Nebius software assets are across all things, going more towards inference, agentic search, etc., as well. At the same time, the Mirantis one is more focused on infrastructure as a service and the orchestration layer, which helps the front end train better and make it more effective while the hyperscaler is training their models. Nebius is more geared toward full-stack, developing or supporting AI startups as well.
Analyst 2
Is your reason for owning IREN over Nebius purely at this point valuation-based?
Analyst 1
No, we still have a position in Nebius. We sold half our position, and we then allocated into these 3 stocks. It's not that we own IREN over Nebius because of valuation. It's that when we were thinking about where to deploy a 5% stake, the answer turned out to be IREN, Hut 8, and TeraWulf as the best risk-reward options, which came out ahead of the other companies. These companies also have a lot of power in their portfolio, which they can monetize. It's almost like the free call option that you get by having so much access to power. If Nebius today had a 5-gigawatt power portfolio, I can easily underwrite a $300 share price for Nebius. Similarly, for IREN, if they're able to actually monetize and have their execution monetized well, it's not difficult to see IREN as a $100+ stock. So the upside I think is much bigger if they're able to convert that call option into actual revenues. In the case of IREN, the only risk is how diluted I will be before I get that upside.
Analyst 2
I mean, compared to a Hut 8 or a TeraWulf or a Cipher, your upside is obviously much larger, but your downside if they don't finance this correctly is also -- we don't even know exactly what it is, right? It's not great.
Analyst 1
Exactly. It's almost like I'm thinking of IREN as how we were thinking of Nebius when it was a $20, $30 stock. At that point, my models were probably underwriting that Nebius would be going towards 1 gigawatt of data center capacity by 2027, 2028. But now we're thinking about 4 or 5 gigawatts. Obviously, that's leaps and bounds ahead of what I was modeling back then. The share price has reflected that. It's almost similar right now, right? Essentially, the current share price of IREN is not even fully reflecting the value of the signed contract or the contracted revenues so far. So everything else beyond that, which is like 10x, is a free call option in the stock. As the probabilities of that go from zero in the stock price to anything positive, the upside can be pretty meaningful, is how I'm thinking about IREN.
Analyst 2
It has been slightly confusing. If you look at their NVIDIA deal, clearly the price for an IT megawatt was significantly higher than it was for the Microsoft deal. And their old guidance, I guess they've updated slightly, but it was basically they would only sign at the same dollar per megawatt as their Microsoft deal for their remaining uncommitted capacity, which, if you just pay attention to the market, makes no sense, right? Obviously prices are going up, availability is declining. When I did the math, you could do it different ways because you have to assume different margins and different residual values at the end of the day. I mean, if they thought Microsoft was low double digits, and I didn't always agree with them, but you're at least mid-teens unlevered on NVIDIA, at least the math I did. And then your levered is significantly greater than that. If you get to run it out for 8 years, or if you can refresh it, it seems like the contracts they have in place are quite valuable.
Analyst 1
Yeah, exactly. And they're quite valuable, and the share price is currently not reflecting that. I think a lot of that also has to do with what is happening with Situational Awareness and other funds, which were big owners of IREN.
Analyst 2
How do you guys think about the residual value of at least the data centers in year 5 when Microsoft or NVIDIA contracts end?
Analyst 1
What we think is essentially you start with almost a philosophical question: will the AI compute demand keep on increasing or will it stagnate at some point? We think that over the next 5 years, there's very little probability that AI compute demand will stagnate. So this will continue to remain a market where there's a lot of demand and supply is essentially bottlenecked by how fast you can grow capacity. Obviously, there are a lot of factors, including power and everything from the government side, New York banning data center construction, etc. This will be a situation where demand should continue to exceed supply meaningfully over the next 5 to 10 years.
From that perspective, I see that right now a lot of the way that the financing has been done is that you would be able to recover the cost of your GPUs over this period of time. After that, there is definitely a good residual value in the data center. The open question mark that we have is: do you actually need to replace the GPUs? It's related to a two-part question. You may have to replace the GPUs because the new GPUs that are coming out are just exponentially better than the previous generations. Someone like Microsoft or NVIDIA may not be willing to work with the older GPUs at that point. The second thing is what happens to the pricing of the older generation GPU. The reason why these stocks are still doing well is that you have the H100 and H200 still doing pretty well on the spot prices today, which is essentially giving people a reasonable amount of certainty that these things have value in the long term.
But it's the first question which is really the question mark, because if your main business model is that you're dealing with hyperscalers and they want the latest GPUs no matter what, then you automatically get into a situation where you would necessarily need to replace those GPUs at those sites. So what can happen realistically is that you would have to take the older GPUs, and we think that you'll have some residual value, and they will be using it for a different sort of customer cohort, which will be paying them some price. You will be able to get some residual value for the next 5, 10 years, as AI demand is greater than the supply of tokens. But for the hyperscaler contracts, if you want to continue with them, you would have to replace the GPUs and go through the depreciation cycle again. That's how we're thinking, but curious to hear how you're thinking about it.
Analyst 2
It's interesting because you can still get A100s from the hyperscalers.
Analyst 1
But that's for the regular cloud, right? That's not the AI.
Analyst 2
The truth is, in 5 years, the hyperscalers might have enough of their own GB300 or B200s where they could ask these sites to be upgraded to new GPUs. But that's why I think it's important that they're trying to get away from just pure bare metal. Because you're right, maybe they need to relocate some of these GPUs, but maybe they could also take the 5-year-old product, use it for inference if they're serving some of the AI labs, as opposed to just serving the 3 or 4 hyperscalers. So there's that.
I mean, the other way I think about it is, and I actually just lost where my numbers were, but I went through and thought about what percentage of the DC do I think is reusable? Because that's the other thing. If you have to replace the GPUs, who knows what liquid cooling mechanism we're going to use in 5 years versus what we're using today. But there still is some equipment, right? Like the power, the transformers, the inverters, switchgear, the shell of the site, the site grading. There are still a lot of things that have value. How much you ascribe to it, you can probably come up with ranges, but I would think it's at least half. I'm trying to find my number, but I don't know if you guys have gone through that exercise before.