$iren Sunday Night 6.7.26

Hosted by @₿itcoin ₿utcher 🥩 🐑 🐷 · 2026-06-07 · Tags: IREN

TLDR

Participants assessed IREN’s expanding powered-land portfolio, GPU deployment schedule, construction cadence, and potential AI-cloud contract economics. The discussion was broadly bullish on long-term demand and the strategic value of grid-connected capacity, while acknowledging delayed revenue, execution risks, uncertain contract timing, and potentially inflated SpaceX/XAI deal comparisons.

Speakers

Notable quotes

Transcript

₿itcoin ₿utcher: Christophers, and Christophe.

Speaker 2: Stop. . Hey, everyone, that was Gavin Rossdale and Bush. If you're not familiar with the song, It's been a while, but we got to shake things up here. And I'm going to invite Mark on stage. But if you came here to feel sorry for yourself and the price action, I suggest you log off right now because we're not going to do that tonight. I'm going to talk about a few things I saw this week. I think Franz is going to join us later. He was dropping his son off at school. So I'm going to invite Mark up right now and we'll get this thing going. But thanks for joining me on a Sunday night. And yeah, we're not going to do therapy tonight. It's just not my style. I feel for everyone, but **** that right now. We don't need that. Let's talk about where we're going, not what's happened. So what I noticed this past week from iron specifically. was the new 800 megawatt site that's roughly 80 miles from Adelaide. And that's cool on a few different fronts. The historical context, given that both Dan and Will are from Australia and many of the seed investors, so the same reason why I'm proud to be an American and a Detroiter and. There's that sense of civic pride that those guys have. So for them to be able to bring a project back home. And we've speculated before on what that means as it relates to the money. I believe it was $25 billion about to be invested by Microsoft into Australia. And certainly everyone's favorite company in the world to speculate about Anthropic. So there's a lot of opportunity there. And I can dig a little deeper, maybe Franz can speak to it later, but I have a hard time believing there's that many 800 megawatt sites in the APAC region. Those of you unfamiliar, APAC just refers to Asia-Pacific. So Australia, given its proximity to Singapore, Japan, South Korea presents a opportunity where there's such a shortage right now for compute and the related power that it takes to plug in the GPUs that the same people that were complaining about latency in the past are beginning to realize that the latency itself is kind of secondary and that even for inference, that people are willing to risk a few milliseconds for that capacity just so that they can process tokens. So that's number one. The news that we got on Friday that kind of shook things up. I've made a post to the nest as it relates to this week, I believe. I know it's. Check my calendar. Yep. Friday. June 12th, I'm almost positive it is, June 12th will be the SpaceX IPO.

₿itcoin ₿utcher: And.

Speaker 2: We've talked previously about Coreweave and Nevius is kind of the two competitors that Iron kind of should be watching. But now it appears after recent deals with Anthropic and Google that Elon and XAI might be the third NeoCloud per se. And what struck me with that, and I've posted about it, is they being Elon and XAI have two data centers, Colossus One and Colossus Two. Colossus One is a bit smaller. You can look them up later, but it's filled with H200s, and H100, so it's an air-cooled data center that runs on gas, diesel generation, so it's behind the meter. And then they have Colossus 2, which is larger in size, which runs GB200s and GB300s, so it's liquid cool.

₿itcoin ₿utcher: It's unclear to me.

Speaker 2: Related to that deal, what was leased out. But from the update to the IPO disclosure, I believe the form's S-1, it might have be S-2, so forgive me, but essentially before the IPO, they have to disclose certain material facts so that investors are fully informed. And 110,000 GPUs were leased out from XAI to Google, and it's an interesting contract where Google has the right to cancel. But so I'm just running through the numbers that I threw together, some napkin math, as I like to call it, but $920 million monthly over 12 months comes out to $11 billion annual recurring revenue. It was for 110,000 GPUs. So that comes out to 100,000, just over $100,000 per year per GPU. And if you divide that by 365 days and 24 hours in a day, that comes out to $11.46 per compute hour. If I had to guess, it's Colossus 2, and it's probably GV200s or GV300s liquid cooled, just given the rates, if you're comparing it to the recent price updates that were published by Navius. But having said that, like let's say they're GV300s, Iron's leasing those same GPUs out to Microsoft at Horizon one through four for under $3 an hour. And now we're seeing a new competitor emerge in the space that is leasing it out for more than three times the amount. Now, I also want to caution people to not get too bullish on this because I personally, the more I dug into it, Alphabet, which is Google's parent company, has a 6% interest in SpaceX. So what I think's going on personally is between the Anthropic deal and the Google deal, those were two new Business transactions that were published right before the IPO that helped drop the price to sales ratio from over 90 to around 40 now.

₿itcoin ₿utcher: So that makes me kind of suspicious of the intent of it.

Speaker 2: I know Jamie Diamond's on the road with Elon kind of trying to push the stock and then I'm seeing. the brokerage accounts, the rates, I should say, the account balances required to participate in the IPO are dropping. To me, that sounds like the market's kind of balking at the initial valuation, which should be in excess of a trillion dollars. And that's not to say that SpaceX and XAI can't accomplish great things, but how far ahead are they in their evaluation?

₿itcoin ₿utcher: I personally, I can't make an educated comment on it right now.

Speaker 2: But I would say there's not many trillion dollar companies in the world. It's not to say they can't get there yet, but as an investor, I personally would have to take more time and get comfortable with it as opposed to chasing this on Friday. And my fear is there's a lot of people out there that know how talented Elon is. And then they see a guy like Ron Baron on TV talking about how it's going to be a $30 trillion company.

₿itcoin ₿utcher: And assume that's enough for due diligence.

Speaker 2: And personally, that's just not how I invest. So a lot to be seen from that. I think as it relates to Iron, something I've been harping on for the better part of a few months now is The big elephant in the room is we have GPUs from three months ago that are still not contracted that were to be delivered to McKenzie and Childress. So that 50,000 GPU batch was ordered, I believe, March 4th. And those are in our guidance for approximately $1.3 billion, which comes out to just under $3 per GPU hour. But now, granted, there are B300s, not potentially GB300s, but the direction of pricing and the premium on compute that's readily available right now. I hate saying this phrase, but it is relevant. Time to compute right now is relevant. It's just we haven't been plugging in fast enough as a company, and I'm sure everyone already feels the same way about that, including management. But this 50,000 is an opportunity where we're guiding for $3 and after the NVIDIA deal for most likely B300s, given the pricing, those were roughly 77,000, which would imply around an 11, 10 to 11% increase in GPU pricing. since they announced the Nvidia deal, it's most likely we know that it's an air cooled data center for 60 megawatts. So it's fair to say it's probably B300s. And if that's the case, then 377, which includes managed services from Marantis, I believe appears to be the floor. And given that these clients that we're most likely going to sign, with the exception of maybe Anthropic, I can't see anyone with the negotiating leverage of Nvidia. And if anything, Anthropic has less leverage than Nvidia because they don't have control of the GPUs that we desire. So my anticipation, that's a long way of saying that $296 an hour and guidance is ******** and we're going to most likely see $4 an hour. I would put money on that. That's how I'm positioned. And if that's the case and you have roughly a 30% increase in guidance for $1.3 million, roughly a third of 1.3 is just over $400 million. So I think that 1.3 looks more like 1.7 and maybe as high as 1.8. And why that matters right now is the market says that we have 33 million in quarterly revenue from AI and that we're just a powered shell company that doesn't know what they're doing and can't plug in long enough. And I think it's important that they knock the **** out of the cover of the ball on McKenzie and Childress, because right now we're guiding for 4.4, but that includes some. revenue that won't be delivered until next year, which is the Nvidia is 700 million or 680 million that rounds up to 700, the .7. So if you can get 4.4 to look something more like 4.7 or 4.8 or maybe even 4.9, that extra four or $500 million means a lot more to us right now than it's going to mean next year. And that's kind of where my attention has been right now. Certainly Sweetwater's on the forefront, and I think Mark wanted to speak a little bit about some of the advantages of partnering with NVIDIA, and I'll give them the floor in the second. But my thing is right now, and maybe the other piece that I'm that I had in the nest is the new partnership. Forgot which company. Yeah, so they're partnering for with a high bandwidth memory maker, which will help with high bandwidth memory being important for not only GPUs, but also the CPUs that are all in a server. So that also reinforces, and I think Mark will speak to this more, why it's important that we are partnered with NVIDIA if they're able to lock up. the supply chain and we know that there's certainty in delivery and now we're higher priority on their list given the partnership. I think there's going to be more certainty in us being able to purchase GPUs and subsequently lease them out and then ultimately install them and then that flows to the income statement. So those are some of my opening thoughts. Certainly Mark's going to come up, bring some energy, talk about What he had sent me a pretty cool graphic that I mark. I apologize. I just it's my dad's birthday this weekend and was that celebrating the old man's birthday and got back and but I'm sure you can speak to it. And if you want to post it, feel free to post it so people can look at it. But. basically reinforcing what we were speaking to last week in a little more detail on some of the operational efficiencies and what we can look forward to from working with NVIDIA. So I hope that's a good enough intro for you, Mark. It's the floor is yours. Welcome tonight.

Mark: Hey, thanks for having me again. It's great to be with you. Don't worry about celebrating your dad's birthday. That's important. That's the most important thing.

Speaker 2: Yeah, I tend to agree. The only thing, family above iron, unfortunate. Well, I'm fortunate. I have a great wife, child on the way, and some great parents. So those are the only things that come up in religion, but that's for another time. But iron's right up there below those. So take it away, Mark.

Mark: Okay, well, I'd love it if you stayed open, because I like the back and forth, like you mentioned last week. I kind of like the the like the radio show kind of host dynamic. But that's up to you. There's I mean, you've already talked about so much, so my head is like is swirling because I want to talk about some of the things that you already touched on. I, you know, with the with the dollars per megawatts and this most recent contract, Google and SpaceX, Interestingly enough, I've been looking into a lot of resources when it comes to these contracts and things, and you're not some of the things you're not going to find with a lot of proof behind it. These deals have been announced as as is, as we all know. But I don't know if anyone's really thought about this and this might be a little off topic, but as far as As far as contracts go with Anthropic, just in summary, I think everyone will remember Nebius and of course our favorite Iron, we both contracted with Microsoft, right, back in the fall of 2025. Two different kinds of contracts, but those are both to support Microsoft Compute, right? What Microsoft needs to do with their business. Now, you won't find proof of this anywhere, but the timing of it is interesting. I'm going to try and keep this sounding like I actually know what I'm talking about. But yes, you know, Elon Musk did originally say that this deal with Anthropic directly with SpaceX, XAI for the two Colossus locations. And then you had Google coming in and all of a sudden, you know, having this, you know, amazing, never before seen contract. But Musk said that this was going to be a month to month thing with Anthropic. And the reason that I find that interesting is because I do expect that to fall off. And I do expect Microsoft to pick up a lot of the heavy lifting when it comes to the compute. And that was the whole reason behind why Microsoft signed the deals with Nebius and with Iron, because they had to back up the compute needs they were going to need for their own business and also to support to support Anthropic. So I know a lot of people in the Iron universe, especially here on X, they're always talking about when deal, when deal, when deal. The resources I did look into believe that it's likely that 30% of the compute that Microsoft is going to be buying from Nebius and Iron is going to be going to support Anthropic. Have you have you thought about that? Is that is this old news to people or has has anyone looked into that?

₿itcoin ₿utcher: I think the angle of it sounds like you're suggesting that given the 90-day clause with XAI that it's almost, you know, in construction, you have like a bridge loan or this is almost bridge capacity that gets to potentially Microsoft. Where I would push back on it is I think Anthropic's primary need right now is most likely more related to inference. Oh, yeah. So with Iron specifically, Horizon one through four, you have the Supercluster, which my understanding is more for training. And I thought looking back, I can pull the tweet up. It's going to Dan had commented on a tweet from Satya, the CEO of Microsoft, referring to. Yeah, Nadella, Nadella, Microsoft AI or M AI. And previously, my understanding of it was Microsoft put up the first billion for OpenAI, and then they were using OpenAI essentially for as the foundation. And again, tech people come up and correct me, but essentially as the foundation of Copilot, whereas now they're going in their own direction and creating their own frontier model to compete with OpenAI, which is why they had kind of a subsequent breakup and why now there's all these incestuous relationships where Amazon and Google and everyone's investing in both OpenAI and Anthropic will meanwhile putting up their own frontier models, which also is another value proposition for Iron right now with Anthropic is having a landlord who doesn't necessarily want to compete with you on a frontier level model. So I think that's another reason why eventually they'll find, just from sheer numbers, they need the compute, but also the idea that If I'm Dario and I don't, you know, if I can go in a room with Dan and be like, hey, you're not going to try and screw me later with what we're using and the intimate details that we're sharing, which kind of felt like what was going on with Microsoft, where they could almost use open AI and they probably felt entitled because they had an equity interest. But meanwhile, they're trying to take it for themselves for Microsoft AI, and that's a whole other conversation, but. So those were kind of, that's why I only pushed back a little on the Microsoft piece, Mark. I can't speak to what Nabius was trying to do for Microsoft. The only thing, I mean, it was a large top line number and our frustration in trying to analyze the deals side by side is we were having to essentially do forensic accounting and kind of guess, guess how many GPUs were in a cluster and how many they were packing based on PUE and just the location being in New Jersey. And now there's a whole other element to it where their Vineland site was previously trying to use these steamboat engines. I believe they were, but they couldn't get air emission approval. So then that's why Bloom Energy comes in. So they're going behind the meter and using fusion technology and just that whole site is kind of, I don't think people are looking at that close enough. Not to say they won't deliver on time, but even if they deliver on time and they're spending an extra between the co-location fee and then the Bloom.

Speaker 2: Energy piece of it, if you're spending a few extra million dollars a megawatt.

₿itcoin ₿utcher: Are you going to be that much more profitable than iron on that deal? And I would argue no.

Speaker 2: So no, the numbers.

₿itcoin ₿utcher: So those are some of those thoughts related to what you brought up.

Mark: Oh, yeah, The numbers say that they won't be. I believe it was 2.6 billion, I believe, right?

₿itcoin ₿utcher: Yeah, I'd have to pull it up like over 10 years, I think, an extra. Yeah. I remember it being 2.6 billion and then just for horizon one through 4. for 200 megawatts critical IT at 15 million per, that's 3 billion. So we get a full data center for 3 billion and for 2.6 billion, they're looking at. a power solution that is going to depreciate over time and not have a useful life probably beyond those 10 years or maybe 15 years. I am speaking out of turn and don't know this to be certain, but I'm willing to bet at first glance that the data center being constructed in Childress will last longer than the Bloom Energy technology will. So yeah, yeah.

Mark: Back to the, but and just to finish off that other thought there, but Microsoft and Nvidia, between them, we're going to invest up to 15 billion in Anthropic, and that goes back to November 18th of 2025. So up to is, obviously different than will do, but But the point you made is, yes, I think Anthropic is going to land and find that Iron will be their best partner because of because of the the the lesser co-mingling of service services that they're providing compared with everybody else that's in the basket, you know, for providing training and inference. So, but Anthropic is certainly right now the leading model and who everyone is looking at for the moment. But I think, and we were talking about this a little last week, I think Anthropic will certainly end up being involved in the Sweetwater campus. there's two, there's two, there's two, of course, 1.4 gigawatts and 600 megawatts. And we're working on the first 1.4 right now. And I believe the plan is for them to, you know, for iron to, along with Nvidia, to get the equipment in there in 50 megawatt tranches and have 50 megawatts going in intervals. So they'll be adding 50 megawatts at a time up to that 200 megawatts that are supposed to be commissioned and revenue operating mid 2027, you know, mid to late 2027. So there's going to be new annual recurring revenue for IRN to report once they have that capacity contracted. And of course, the the that that'll be for both revenue and ARR. So but revenue and ARR are both important. They, there are some it's a sum of the latter and the former. But nevertheless, it's going to happen and it's going to be very exciting. And this goes back to last week we were talking about the Nvidia penetration plan of AC. So, you know, the cloud the industry, the industrials and the enterprises. So the AI cloud, the industrials and the enterprises, that's what Nvidia is going after. If you saw the breakout, they had matching revenue this past quarter for a blend of just hyperscalers versus the AC category. And they're about even, but the AC category is growing 30 plus percent quarter over quarter. And this is the magic piece that Iron is going to fill for Nvidia. And this is where Iron is going to really blast off with revenue producing assets. Nvidia is going to be Nvidia is partnered with, you know, original equipment manufacturers, most popularly Dell and Lenovo. They are capturing this enterprise segment by bringing AI to the premises. So Jensen's main focus right now is all these enterprises and all these industrial companies across the globe. And he wants to bring AI on-premises. He wants to bring AI there. And how Iron fits in here is Iron is going to support the on-premises AI hardware with the heavier supporting AI factory super data center is what it's going to be really. And that's what we were talking about a little bit last week. And that's why when we are when we're also modeling some of these revenue, this revenue scaling, we're going to be seeing revenue scaling for Iron in the four, 5, 6 billion, you know, 7 billion per year ARR numbers. So over the course of a five year contract, if any of them would be that, but then, now we're starting to get into the 25 to 30 to $35 billion contract range. And the aggressive scaling of business that we're seeing, I think these, I think the Google, you know, and Anthropic payments are outliers still at this point, but it's a glimpse of the future. And you're going to be looking at 10 to 15% easily year over year as the tranches build not only at Sweetwater, but they also build in these other locations that Iron now has as in Oklahoma. And now we have the new one in Australia. And of course, There's the model, there's the location in Spain, but obviously there's 5.8 gigawatts, and it's still in infancy. I always find it interesting that people always try to quantify where we are in this process, and that gets a little old. I mean, it's still first inning. This is decades. This is not something you just roll out and we're like, oh, two to three years, though, we're done. We're there. No, The infrastructure is beyond what anybody can quantify right now. Decades is the best way to put it. Decades. Inference is going to be a huge deal, obviously, but decades and decades?

Speaker 2: Let's take a breather there, Mark. We got no, I love it.

₿itcoin ₿utcher: I love the enthusiasm.

Speaker 2: We just I got Franz on stage, wanted to welcome him back from hiatus and.

₿itcoin ₿utcher: Give you a break there for a second, see if he had any opening remarks or any other topics.

Speaker 2: But I want to keep you up here, Mark, because everyone loves your energy. And I know you've been passionate in the in some of my comments and that's why last week meant a lot to finally get you up there and this week. So we want to keep you up here.

₿itcoin ₿utcher: I just want to.

Frans Bakker: No worries.

₿itcoin ₿utcher: Give a hand off to Franz for a second, welcome him back, some opening thoughts or if there were any other topics he wanted to discuss.

Speaker 2: I'm not sure how late he jumped on, but Franz, I touched on the new Australia site and the X AI deal with Google and talked through some of the numbers related to that. So welcome back.

Frans Bakker: Hey, thank you, Butcher. Yeah, I literally just jumped on when Mark was making his final sentences. So nothing, I didn't hear anything so far. Thanks for hosting, by the way. Yeah, I've been a little bit distracted from hosting spaces in the last two weeks. So thanks for, you know, filling in for me. I guess we can most likely run one together next week again. This current week should be a lot better for me. I'm just going to respond sort of to one of the posts you planned in the Nest, just to give some food for thought to start the space off on my part. I see SK Telecom plans gigawatt scale AI cloud in South Korea. And I think that we've seen plans gigawatt scale, and we heard this sentence or part of the sentence, how many times in the last six months? Way too many times, right? Everyone and their dog is planning gigawatt scale facilities all over the place. And the emphasis should be on plan, planning, they're planning things. And I think that we as a, sort of early iron investors, if I can call it like that, have a very different look at this. Because every time when we see a news, a release from an end user or a cloud provider that doesn't build or own their own data centers or doesn't develop powered land, who is planning gigawatt scale facilities, automatically you start to think, oh, where are they going to get the power from? Right? I mean, obviously that's the first thing I always think about. And that's why last year we've been thinking, oh, who's Iron going to partner with? Because guess what? They have a gigawatt scale site. So So far, it seems that Sweetwater is both the hottest thing in the world and the coldest thing in the world because for some reason, everyone is planning capacity far out for gigawatts and falling to behind the meter and getting fuel cells to their sites because there is no power. We've heard the OpenAI CFO or I think it was the CFO, right, on the All In podcast talking about there is no power in 2026 or in 2027. And we keep asking ourselves, so what about Sweetwater? And I think that we should just consider the following, and going back to this post in the Nest, planning gigawatt scale AI cloud facilities or sites or campuses is something you can announce in 2026. But guess what? If you're announcing a gigawatt scale facility planned in Korea and you announce it in 2026, when do you think realistically this is going to be deployed? You're probably talking about the first phase in 2027. And then, okay, I just read the second part of it, where the first AI factory is set to come online in 2027. So, you know, this is a phased rollout. And obviously, Iron will do the same thing with Sweetwater. But to all the people that are worried about, will Iron be able to find a customer? I mean, I think that we should go back a couple of years into you know, back into the time where Dan and Will, and I don't want to rehash their me and my brother eight years ago story, but basically it comes down to that because why is everyone announcing plans to run a gigawatt scale facility in 2026? That's because Iron, well, it's not because of that, but it's Iron has a gigawatt site that is not commercialized yet, not monetized yet, but they planned for this years ago, right? So just because Iron didn't announce it when they first filed for the interconnection queue or when they started to buy land, which we already knew about, and now I think we're now two years out, right, since we found out about Sweetwater. I think it was right around August of 2024. So Iron already started to look into the Sweetwater gigawatt scale facility, most likely between 2020 and 2022. So just to put this into perspective, internally, Iron already planned for a gigawatt scale facility, and it doesn't matter if it's for AI or for Bitcoin mining, they planned A1 gigawatt plus site because that's how much power they ask for from the utility and that's how much they got from the interconnection agreement. So I just think people should slow down a little bit when they see these announcements and they worry like, oh, it's again, it's not with Iron. You know, why are all these companies, you remember when when there was a poolside who announced a gigawatts scale facility called Horizon with Coreweave or something last year. You remember the plans of this gigawatt scale facility that never really happened. I mean, it's not to say there's not going to be anything there, but first of all, it's not a grid connected site. They are going for behind the meter and they have to fall back on a different customer now, the last we heard, I guess, was they were talking to Google or something, I don't know. But just to put things in perspective, Iron has been actively searching for gigawatt scale sites since the beginning of this decade. And the latest we know internally, Iron's only looking at gigawatt scale sites since last year. I mean, in 2024, you could go to Iron and say, I have 300 megawatts, and they would pick up the phone. And in 2025, if you didn't have at least 500 megawatt, they were basically not interested. And by the end of last year, it's turned completely to GW scale sites. So Iron already has these sites, and there's some only France people in here that probably know there are more 1 GW plus sites that we know of that haven't been disclosed yet. But just to remember that right now it may look like Iran's only got land and power and no customers, but you can bet your *** that if all these companies that are the end users or, facilitating the compute to the like end user. These are customers that are also interested in sites that Iron has. And if there's not a deal being signed today or in the last couple of months, it is not a sign of a lack of demand. It's not a sign of Iron's taking too much land and power and they don't know what to do with it. with it. I think, strategically not going with Morgan Stanley's site selling, you know, originally was the original plan. I think that was a very good decision. And just because they haven't figured out the exact use case and customer or haven't announced it right now doesn't mean that there is anything to worry about. Actually, We should just consider the following. Is there any other powered site in 2026 of more than a GW that's grid connected? I think that's something we forget. Iron has like a bunch of primary substations and a self-owned bulk substation that everything is tier 3 redundant, it's connected to the grid and it's energized. You know, this is something people seem to gloss over and just put it on the bulk, it's just put it on the big pile of power that they have. But A1 gigawatt plus site is still very attractive for multiple reasons compared to a 300 or a 500 megawatt site. So I think the reason why people are announcing gigawatt scale plus facilities, and I think even Nabius has been starting to announce gigawatt scale facilities, right? If I'm not mistaken, I think Missouri is ultimately going to be a gigawatt or 1.2 gigawatt. So I think it is, you know, just to pause here, but I just want to remind people that We have the goodies, I think Iron has this site already powered and they've been looking into gigawatts plus scale sites for a long time now. And well, I won't announce every site that they have. Some of the people that are in my Discord, they know about it, but there is definitely more in the pipeline. with regards to this. So just to touch on this post in the nest, those are just my first thoughts when I'm seeing another announcement of a gigawatt scale site. So I'll pause there.

₿itcoin ₿utcher: I'll take the L on that. I saw the SK and picked the wrong tweet to put in the nest, but I was trying to emphasize the partnership with SK, INEX and NVIDIA regarding the memory piece for both Vera Rubin and Vera CPU and tie that back to how that supply certainty for NVIDIA bodes well for Iron going forward as we use up those GPUs and the new relationship that we consummated during the prior earnings call. So my bad on that, but I would say There was a small cap. I don't know if he's in here, but small cap had a post today on an interview from, I think his name's Asher, the CEO of Hutt, speaking to their eight gigawatt pipeline. And let's make one thing clear before I make this analogy. I'm happily married and I'm not seeking any relationships. But back in the day before I met my wife, I had plenty developments out there, plenty of prospects, but that doesn't necessarily mean that you close the deal with all of them. So when Asher's talking about eight gigawatts, like that's cool, but how much of it are you going to actually secure and when is it? And I do know that Hutt specifically is trying to secure grid connections in Texas, and I guarantee they weren't doing it as early as Will and Dan. So whether that works out for them is a whole other scenario. And I don't think the market fully appreciates that yet. And right now they look at 5.8 gigawatts and they see that it's empty or that it's not, you know, it's not completely empty. We're certainly have a 3.1 contracted right now. But what they fail to see is they'll bring up the bloom energies or the behind the meter. I just don't think it's going to take some time until it hits our financials for people to really appreciate. And the same people that are saying that power isn't scarce because we keep announcing it as like you're missing the point. Like they've been working on this for four or five years and this is all coming to fruition now. All these other sites like Franz is speaking to one gigawatt here or one gigawatt there. First, you don't even know if it's going to get approved. Second, it might be behind the meter. Third, it's probably going to be 2028 or 2029. So a lot of this is just it's it takes a little more than reading a headline to appreciate it for what it is. And I think real people can see through it. And unfortunately, it sounds nice when you go on CNBC and everyone says, wow, they just signed Anthropic to 250 megawatts. And if you multiply that by four, that's one gigawatt. And he just said they had eight gigawatts. Oh my gosh, Hut's a home run. And it's like, this is just an example too with Hut. It's not, I don't have a personal grievance with them, but it's just flawed thinking if people actually understood how grid connections work and certainly behind the meter, like behind the meter, no one is doing that by choice. They're doing it because they have to. They're doing it as a hedge for the grid connections that they may not be able to secure, or they're doing it to bring more power online. But the same hyperscalers that are aggressively.

Speaker 2: Negotiating with Dan right now, they may pay up a little bit, but if it costs that much more to produce a data center, how much margin really is going to be left specifically for a co-location deal? So those were just some of the additional thoughts I wanted to add to what you were saying, Franz, and I had touched on earlier what we were talking about this morning, the 50,000 GPUs that we do have secured that are supposed to be delivered in second-half of this year, how that's our biggest opportunity with the exception of Sweetwater, I think, to.

₿itcoin ₿utcher: Re-rate if we're able to deliver those for closer to $4 an hour. And by contracting those and announcing those deals, whether they're separate press releases or at earnings, announcing those contract terms.

Speaker 2: I think our 4.4 will turn into 4.7 or 4.8 billion. But yeah, my bad on the gigawatt site in Korea. We all know that the 800 megawatt site that we just signed for Australia is going to be probably supplying the compute for Korea for that site that may or may not exist. And I'll hand it back to you or Mark if you guys have any comments.

Frans Bakker: So I had something before you mentioned Australia. So I don't really mind that you posted the wrong tweet because that's still very relevant for iron investors, right, to see that another announcement of a GW scale facility, which is basically going to be our bread and butter. So Yeah, to your point about announcing power and announcing deals, Iron is just not going to play the same game as some of the other companies in this space. And unfortunately, that is being very harshly punished by the market. The market wants to see backlogs and Iron is not giving them. at least not to the tune of what some of the other players are doing, including a Nebius with multiple billions with Meta. So, you know, this is something that looks like we are behind, and technically we are, but this is not a period where there's nothing happening. And that is the awkward period we are in now. Iron is running north of 2.5 K bodies to children's every day. And they are ramping up the Sweetwater team as well. So they are building out these sites and we have no idea what's next. But we have an idea what's next, we just can't time it. And I think it's very, important to remember that Dan and the team, they look at this from a whole different perspective. They are only signing deals when they know that they are going to get, you know, the GPUs online in the time frame that they're promising. And in, you know, to put that in perspective, We are only talking about how bad is the Microsoft deal and why is it so delayed and why is Horizon One not online yet or handed over to Microsoft? Why is there no customer for the 50,000 B300s? But if you zoom out and you just look at the very technical reality, Iron is deploying in 2026, 76,000 GB300s, 50,000 B300s, and then everything in Prince George that they didn't deploy in 2025. So a quick math will bring you to close to 150,000 GPUs. And it's probably more closer to 135, 140,000. I don't know exactly how many GPUs they deployed in the quarter four of last year. But, they are deploying over 100,000 GPUs this year. And I know that the economics are not all that great of the Microsoft deal and the time to revenue recognition is not what we want it to be because, we've already known about most of these GPUs for, close to a year now. okay, not close to a year, but Microsoft contract was announced in November. I mean, it's taken a while, right? So investors are impatient, but Iron is an infrastructure company and they are deploying over 125,000 GPUs in 2026. I think that's a great achievement. if you compare it to 2025, where they deployed maybe less than 10,000 GPUs. So there is a lot of nuance, obviously, to this way of looking at the business, but it is a big growth phase where they're in right now. So instead of only deploying GPUs in one size in 2025, they have now switched to deploying GPUs across two sites. So there is another growth in absolute site deployments. So what we're seeing happening in front of our eyes is a company that is seemingly not accelerating, right? It looks like they are not signing customers. It looks like they're not increasing their revenue by meaningful amounts. But what they're really doing is they're setting the layer, the ground layer for the next level of, pace acceleration. And in a period right now, we're like basically entering the summer now and it looks like nothing is really happening, right? Nothing has really happened yet. But that's not the case. And these things seem to go much slower than some of our competitors or peers. But there is an explanation for that, obviously. I will touch on that a little bit later. But just focusing on iron right now, I think the year over year growth is going to look like absolutely phenomenal. And 2025 to 2026, if we just look at a bunch of metrics that are relevant for an infrastructure company, I think it's, you know, basically insane if they are going to go from 10,000 to 125,000 GPUs deployed year over year. And we know that they signed this agreement with Nvidia to deploy 600,000 GPUs over five years. Well, quick math will bring you to 120,000 GPUs, right? So If they will be able to do what they promised us, and that means deploying Horizon 1 to 4 in this calendar year, install the 50,000 GPUs, and finish off whatever they bought and did with Prince George, that means that they are able to deploy 125,000 GPUs, or let's just call it 120,000 GPUs in a year. So that means that they, they've proven to NVIDIA, look, you guys, we signed this agreement with you and look what we've done. And I know there's some spillover from 2025. I know they started building Horizon somewhere in July, I think, of last year, but that was not exactly the Horizon that we know today. I mean, it is Horizon One is the same buildings, but it's, you know, So there's also some nuance to that. And of course, Prince George is part of, is partly retrofit. But the reality is that, you know, Iron's been behind on many metrics and most notably on the revenue and the ARR. But 2026 is a show, you know, show me you can do it year. And I think if we just take a little pause and you fast forward to the end of this year, it's going to look a lot different. And to go back to why I think our peers have done so much better is because, yeah, well, I don't want to make only the comparison to Nabius. I think probably Core Weave is also a very good example, but it's just that I know a bit more about what Nebius have done so far. I've really never been interested in Core Weave for some reason. But last year, Nebius announced a lot of GPU purchases. I don't know if you remember, they were talking about Blackwell's last year and it was all, we are the best friend of Nvidia and we get the GPUs first. And yeah, well, that's really sort of true. And they did get a lot of stock of GPUs and they already had a data center running. And that's also something that, Nibius investors like to, talk about, we have, we've had this self-built, self-owned data center of 75 megawatts for so many years. That's because, that's why they are able to deploy these GPUs in such a fast manner, right? And they did a little co-location on the side for, I think it was in the UK. So This is a very different scenario where you already have a data center and you have GPUs in a larger quantity earlier on. So that's why we are lagging this company, for example, with Irons is lagging NEBUs in revenue for this specific reason that we didn't have all those GPUs in that quantity last year and we didn't have a data center that was already ready for that many GPUs. I mean, they've been fixing Prince George in the first quarter of this year in terms of permitting for UPS buildings and generators. So Prince George has also had their own, you know, development struggles. Let me just call it like that. So yes, Iron is lagging. in many metrics, but what's really going on is they are preparing for, I've always called this a mass data center construction cadence. And I put a post in a nest of an unknown banker. He says, my biggest concern in general for iron is the cadence of construction that they had until now. Well, he means until now. And if we compare what they've been doing so far to the Bitcoin mining era, where they were able to squeeze out, I don't know, 30 exahash in what, nine months or something, and we could see that they were really fast at building all these data centers in Childress. And that's when I started to track it with the satellite imagery. But just because they have not replicated that so far, doesn't mean that that's not where we're going to, because I can tell you this much, in Sweetwater, they have deployed 3 primary substations at the same time almost. And that is just unprecedented for Iron. If you go back to Childress, they've always done one by one. So they build one primary substation, they build 6 data centers, then they go on to the next primary substation, they build 6 data centers. Now what they've done, they have built a bulk substation and three primary substations at the same time on different places on the side. That means when they ramp up the team towards the end of this year and the beginning of 2027, they will be able to construct 3 separate construction sites, basically. I know it's the same Sweetwater site, but they will be able to go three times as fast if they wanted to, because they have three separate places where they can follow the same cadence. So it will be not just staggered on a data center level, but it can also be staggered on a primary substation level, if you catch my drift. So I think It is mostly optics, why Iron is seemingly slow with all the metrics if we compare it to, for example, Core Weave and Nebius. But the truth is, there is a very good reason why they are behind on this. It's because they are just taking a step back to prepare to launch a massive, mass-scale data center construction operation, which is going to kick off with Mirantis on board, with NVIDIA and Dell as partners. We basically are locked in for our entire supply chain, and we have an agreement with NVIDIA for them to also supply us with GPUs for this exact capacity. So you have to look at this from above, and it's like a puzzle. And the puzzle pieces that are on the board right now, they don't fit together. So it looks like nothing is happening, but they are sort of close together. And as soon as you get one or two pieces that connect a bunch of those puzzle pieces together, amongst them is, for example, a commercial deal for a certain amount of megawatts, that's when all these puzzle pieces start to fit together and it starts to look like, wait a minute, these guys were preparing for this all along. And I think that this is the sentence you're going to most likely hear. more often in the coming 6 to 12 months. So I just wanted to put that in perspective. So to go back to this post I put in the nest, I don't think that's a concern. I think you have to look at the strategy that they're pulling off. They're intentionally becoming a late mover to capture the highest value. of the GPU generations that are coming out in the period when they are bringing these data centers online. And also a final remark from me, and then I'll pause, is the 50,000 B300s that they have announced earlier this year. It's not strange to me that they haven't been contracted yet, because Aaron wants to contract them as late as possible. at least for a most part. I'm not going to say that they won't pre-contract any of those GPUs, but I think we should get used to the delay between an announced GPU purchase and an announced deal or a commercial agreement or whatever, because GPUs carry so much more value when you can tell the customer, yeah, I will be able to finish burning them in two weeks. So how much do you want to have? I mean, if you go back to the OpenAI CFO saying, we cannot find any power for 2027, imagine you're a company that says we have GPUs in 2026. I mean, and we can provide them to you in 10 days, right? I mean, that's insane. Aaron is telling us that their ARR is going to look like ****. And then what they're really doing is they're going to surprise the market with much better rates than what they've currently announced. It's not technically, I think they're not doing this to sandbag. I think they just don't want to tell the market things that they cannot 100% guarantee. Because as we have seen, They were sure they were going to get $225 million ARR in Q4, and they didn't do that. They said they would get $500 million ARR by Q1, and they didn't deliver on that either. They are extremely sick and tired of telling the market things, like we expect this to happen, and then it doesn't happen. I mean, if you do that two times in a row and your earnings are **** people are going to question what you're really doing. And they are, they're like scratching their head, they're like pulling out their hairs, like, What the ****? We've done everything right, and why can't we get, well, why can't we perform what, you know, what we do, what we said we would do? So I think 2026 is like a year where the gears are shifting a little bit. Like they're going to under promise a little bit more, but they're going to over deliver a little bit more than that. So relatively. So, I can rant on this for rant on about this for a long time, but I think it will just, take a little while to really show up in the financials. But I think we are in a very good place and I think we should really, appreciate the puzzle pieces that they did put in place so far. And once they, you know, get the connecting pieces to make it look like a beautiful puzzle, you will remember this rant. I'm very confident about that. And, you know, I guess I'll just pause there.

₿itcoin ₿utcher: Hey, speaking of puzzle pieces, Metaman, your graphic reminds me of a puzzle piece. Metaman, how are you tonight? Metaman, going once, going twice. Stock Analyst Pro, your turn.

Stock Analyst Pro: I'm very excited with the SpaceX IPO this week. Why? Because if you look at their overall valuations, right, I mean, when the people start the, the generic analyst who starts writing about the SpaceX and its valuation and its revenue streams. So almost 60% of their revenue stream right now for this year, the ARR comes from the AI leasing, which is around like 26 billion out of the 44 billion. That would be really, really interesting and it can actually spin off, I know, spin up a lot more focus into the AI leasing and the data center space. That's what I'm excited for. And it's very interesting where 1.75 evaluation until last week, people were very skeptical, but now Musk was able to pull off the Google deal, which was very expensive looking at the numbers. And that can really, really put more focus on the the new clouds, especially Iron and NBS and COVID. So what are your thoughts?

₿itcoin ₿utcher: So can you better explain, do you have a position currently or were you pursuing the SpaceX IPO? Because I think you and I had a somewhat fundamental disagreement on, I thought, I do think there's something to be said about what Franz was saying, having burnt-in GPUs available in two weeks and being able to contract those for a higher rate. But I also think paying an excess of $11 an hour for what most likely are GB300s might be some window dressing for the IPO that's taking place on Friday. So can you remind me of what you're, you had some pushback on that. I just think everyone else should hear it and we can hash it out.

Stock Analyst Pro: You know, I don't have any position into SpaceX, but I have position into ASTS and Rocket Lab and Iron. ASTS competes with Starlink, Rocket Lab competes with the SpaceX launch services, and Iron is our You know, the data center playing, right? So I personally feel because there's so much freaking demand for them to, you know, give the leasing of the GPUs right now, whoever has it can charge it a higher rates. And we all know that, OK, even Google who adds so much infra play and even they are struggling to get the compute because the demand is so high, right? And whenever. I agree with what Franz was saying, where I think Dan might be delaying the announcements of the projects because he knows the demand would be even more higher in H2 and things how shape up, how it's shaping up. My whole discussion in that online chat was because they have the GPU right now, because they built their own data center, thinking that there will be a lot of demand for Grok, but Grok is not going anywhere except when people who are on XAI just use it. So they have additional compute, they have stuff. So I think they are in much better position to negotiate. And whenever Iron, you know, completes, you know, getting some of the GPUs online and gets things done, so they'll also be in a much better position to demand higher rates. I agree with Frank on that.

Speaker 2: All right, thanks for coming up.

₿itcoin ₿utcher: And if you want to chime in, you're more than welcome. Let's go to Jonathan next. Jonathan, how are you from Canada?

Jonathan: Doing pretty good, Bitcoin. How are you doing from Detroit?

₿itcoin ₿utcher: You know, we have on a off topic here, but our captain, Dylan Larkin, who's actually from Waterford, Michigan, asked for a trade from the Red Wing.

Speaker 2: So it's kind of a sad time in hockey town right now. But but iron's good. So why don't we we'll redirect it back to that. But I I saw you at the I believe it's the Leafs in your avatar, but maybe you guys will trade us Matthews for Larkin. We'll see what how that plays out. But what can we do for you tonight?

Jonathan: Yeah, I'm definitely familiar with that whole drama and good luck to you. But yeah, in regards to Iron, I don't know, like now Franz has got me a little bit nervous because I do understand what he's saying about maybe we're going to be waiting a little bit longer for the deals because imagine if you can go to the customer and say, you know, we've got the GPUs ready as well as the power. But like now I'm thinking, well, like when are the GPUs going to be ready for Sweetwater and then like man like how how later are we talking about a Sweetwater deal then in that case and then it's like I mean like I'm willing to be patient but I don't know if I'm willing to be that patient so maybe France can ease my concerns a little bit go ahead France.

Frans Bakker: So I guess I failed to mention that most of this RAND was in particular to the 50,000 P300s that are announced. Obviously, Iron's not going to go 100% only contract capacity when the GPUs are arriving or have been installed to start to burn in or whatever. It is going to be a mix, obviously, between sort of backlogs and this tactic of waiting. And for Sweetwater in particular, I don't expect them to do the same strategy as they are doing for the B-300s in Childers and Mackenzie. I think that it's going to be the entire opposite of this, you know, of this strategy. They're, I think that they are going to sign a backlog a sort of backlog, basically. Obviously, we're talking about 2027 capacity. So we are in the middle of 2026. Is it really a backlog? It is technically, I guess. But so just to ease your mind, I think for Sweetwater, it's not going to go the same way as I think they're going to pull off the 50,000 B300. So I think that's just going to be a commercial deal that they will sign this year. for 200 megawatt of IT load, all basically according to the rendering that we saw at the at the earnings presentation. So yeah, I hope that answers your question.

₿itcoin ₿utcher: Hey, Jonathan. Yeah, if you don't mind.

Speaker 2: I would just add when you think of the NVIDIA deal, like the GPUs haven't been procured yet. That's for 2027. So.

₿itcoin ₿utcher: That kind of goes against.

Speaker 2: That thought process, as well as Horizon 1 through 4 that was announced in November, and those GPUs.

₿itcoin ₿utcher: Weren't delivered until more recently, and the contract terms were agreed upon.

Speaker 2: So I tend to believe that's what's going to happen at Sweetwater.

₿itcoin ₿utcher: Sorry for cutting you off.

Jonathan: No, that's what I kind of expected that answer. I just, I wanted to make sure 100%. So I appreciate the clarity, guys. Thanks.

Frans Bakker: Yeah, and just to add to that, obviously SweetWater is going to be a Vera Rubin deployment. And I think, you know, Iron and other cloud providers are probably negotiating or discussing rates with Nvidia now for acquiring these GPUs. I guess Nvidia is starting to take on, you know, orders for the production of these GPUs as it's going to start ramping up in the second-half of this year. Jim knows a little bit more about that. I think he said that the real mass production is starting, you know, next year, but I guess it would make sense if Iron gets a sense of how much they're going to pay for these GPUs. I mean, that's the beauty of this cooperation with Dell and Nvidia. Well, it's technically, I think it's more with Nvidia, but we know Dell is in the mix here as an OEM. I think they're going to get better insights into when and how much they are going to pay for these new GPUs. So if you are still a year away from deploying the GPUs, because first of all, you have to build a data center, so you have to wait for the production of the chips, but you do know the rates. I think that is the big difference with those 50,000 B300s. You're not looking at four to six months to deployment. like with the B300s, you're looking at probably 10 to 12 months for when you start seeing these GPUs come online. So I don't think Aaron's going to wait until, you know, the middle of next year until they sell these GPUs. First of all, that is a massive amount of CapEx that are going to, you know, invest into these data centers. So the mix that I was talking about I think 50,000 B300s are pretty easy for Iron to just order and deploy in their already existing data centers with a minimal retrofit cost. But for Vera Rubin in Sweetwater, that's a whole different story. So I think in order to finance this operation, they're going to have to sign a backlog for that. And I think that's exactly what they're currently negotiating. So I don't think we have to be patient until next year for a deal with Sweetwater. That's my point. And I think it is by deduction you can get to a logical conclusion that, you know, agrees with that. So I think that that's how it's going to go. You know, last minute monetizing of the B300s and a backlog for Vera Rubens in Sweetwater.

₿itcoin ₿utcher: Franz, I had a question, high level. I you might have to kind of do some more research on your end if I'm wrong, but what's your understanding just for everyone as a refresher on where we stand at Childress? And after you, I'll have a few follow up questions, but if you could just kind of run us through your understanding of where Childress stands right now with respect to Horizon one through four, that would be a good starting point. And then I can kind of use that information for a follow up question.

Frans Bakker: Yeah, so let me just, you know, think how I'm going to explain this without making people worried again. Basically, whenever I say something that is sounds worse than what people expect, it is automatically means irons is in trouble. So High level, I think that they're going to deliver Horizon 1 to 4 in 2026 or maybe some spillover into quarter one of next year. But that is pretty much my bear case. Horizon 1 is, you know, I expect the GPUs that they have in the first building. So Horizon 1 is 3 buildings. I think the first building will most likely be starting to bill in, if it's separate billing in July, so next month. But if there is no separate billing, then I think that the combined Horizon 1, 3 buildings will be done and billing by the latest by the end of August. So I don't think this is a September story. And I know that September is Q3. So I just want to say that it may be the middle of Q3 is what I am considering now. There are already GPUs in the second building. That's also what I heard. I don't know if how many, and I don't know 100% in what stage they are. I know that a big delay, or well, a part of the delay that they had with Horizon One was the liquid cooling plant and dry cooler installation. There was something that they didn't like. or didn't work the way they thought it would. I don't, I mean, obviously I'm not an insider, so I don't exactly know. But if we look at the following iterations of their liquid cooling plant and dry cooler installation, they changed to a different version. I mean, you can see it in the picture. There is something that they changed. So they didn't change this because they liked to have the fins on top, that they needed that. So, I also heard that over the long run, they might consider to exchange the liquid cooling plant for the first horizon to match the others. So that will probably give you some kind of idea of I know why they were a little bit slower than we expected with Horizon One. But the real, the reality is that they have already figured out a solution for that. So this is the good part. They found a bridge solution in order for them to start burning in these GPUs. GPUs in building one have already been humming for a while now. So I think they are in, they're being burned in. but it's still an ongoing process. And so, I'm mostly focused on Horizon One now in this story, because Horizon One is the problem child, and Horizon Two, Three, and Four are going to be, you know, the more smooth deployments with a different style of liquid cooling plants. Different dry cooler installations and a bunch of other things that they changed they stopped using roof fans, for example. So, my take on this deployment is... that if they are able to deliver the first horizon in Q3, that means that the biggest concern of me personally is out of the way. And Microsoft and Dell and Iron are all very working together on this. And they're so there's not like a delay penalty or anything like that. There's no, from what I heard, there's no such thing being discussed even because Microsoft wants a perfect deployment. And from an infrastructure standpoint, that just means they have to perfectionize a lot of things on the fly. And so this deployment is also sort of a R&D a project, if I can call it like that. And that's also why I keep telling people, don't focus too much on the economics of this deal, because this is a first deployment within a partnership, right? It's, yeah, it's expensive and it's, some people call it, you know, shareholder value destruction, but that's only if you zoom in on the economics of the deal. But this is part of a much bigger story that you can't see yet if you only look at the Horizon deployment. So Horizon 5 and 6 was no surprise to me because within a partnership, you're going to expand. And I think the cadence of delivering Horizon 5 to 6 compared to Horizon 1 from start to finish is probably going to show a massive improvement in time to revenue. So just to sum it up, I see Horizon One being delivered in August. I think they're well ahead with the first building to, you know, start delivering some compute to Microsoft. I just don't know if they are being built, building by building, or only as a cluster of three. And I have 0 concerns about the following three horizons, because outside of late delivery of GPUs or certain liquid cooling components, I am very confident that Iron got it now, if you catch my drift. So I think it's going to roll out like a hockey stick up towards the end of the year where you're going to see, you start to see signs of that Bitcoin mining era where they were able to deliver building by building in such a quick fashion. And to that point, they already finished the last building structure of Horizon 4. So now there are 12 shells with a roof. And it's basically comes down to deploying all the external components, the pipes and the liquid cooling installations. So, they are starting to get into this cadence where deployments are going more rapidly and they are professionalizing the last little bits of the deployment in the first building. But as soon as they can sign off on that and say, okay, this is how we're going to do it, then they're just going to copy paste that across all the 12 buildings. So have no fear. They are really getting there. And I see the 75% likely to deliver the entire thing this year and 25% with a spillover in the first quarter. So It's just my own personal estimates. That's not financial or investment advice. It's just my, that's what I'm going off from. So, yeah.

₿itcoin ₿utcher: Okay, so as a follow-up question, I'm looking at the breakdown of the 730 megawatts that they're scheduling to build next year. And I'll just read it off for everyone. There's the 30 at Canal Flats, the 300 at Sweetwater 1, 150 for Horizon 5 and 6, 60 for Nvidia, also at Childress, and then another 190 at Childress. So I think my question to you, Franz, is there's two ways we can go with this. First, the problem child that was Prince George, what's a realistic timeframe for, I'm counting this right now, with Nvidia and the other air-cooled capacity, that's 250 megawatts. Do you have any estimate as to how much faster they're able to construct, or I should say retrofit air-cooled DCs versus liquid-cool? And then related to that as a second question, I think where I'm getting at is you spoke about how they changed their design at Sweetwater One to accommodate, or I should say, install simultaneously three distinct substations, whereas previous iron iterations might have only had one substation at a time. So when you say something like that, and correct me if I misunderstood you, I guess I'm looking at it as a potential bull case for next year if they really figured their **** out and can maximize their performance here going into year end and finish horizon two through four on time. Certainly there's overlap of labor between the two sites. So yes, you would have to construct horizon five and six and keep some people in Childers to do that. But It appears there might be an opportunity to build more than just 300 megawatts at Sweetwater next year if they've changed their approach and how they're kind of installing electrical infrastructure substations first, and if they're able to speed up that process given these growing pains. You know, they're only guiding for 300 megawatts, but if they can work on three sites in parallel, that might be an upside bull case. So I just want, I know I threw a lot at you right now, but I think the first question was specifically how the air cooled cadence, what's a realistic time frame on those? And then the second piece would be. your thesis on Sweetwater being able to be developed more quickly given all three substations are already laid out, yet they're only guiding 300 megawatts next year.

Frans Bakker: Yeah, so the air cooled GPUs in Childress and maybe McKenzie as well. Let's just focus on Childress first because that is a 2027 deployment. Iron has repeatedly said that it's basically the same thing as they've done in Prince George. So it's just taking out the Bitcoin mining racks and rolling in the GPU racks. I mean, that's what I'm seeing right now. They've literally put containers around the mining buildings and they are, you know, separating metal and non-metals and stuff like that. It's really being retrofitting right now just means displacing or removing things from the DCs. And, that brings the question, when are the GPUs expected to arrive? Because outside of electrical redundancy, which for the most part happens outside of the DC, what are they really What do they really need to change internally in these data centers? I'm just going to go off by their own guidance and that is, a matter of weeks, right? I think what they said is it takes like what, four to six weeks or 6 to 8 weeks. So I think the deployment cadence of B300s in Childress and, you know, to The same thing applies, I guess, for McKenzie, even though McKenzie specifically has a bit more of a, the generator and the UPS buildings outside of the DCs most likely need permits as well. So there's probably something there which makes it a little different than Childress, but I think it all falls back on the arrival times of the GPUs. So That makes me question why they are retrofitting more DCs than they have announced. Maybe that's more interesting to talk about for me, but the 119 megawatts for 2026 are just a, you know, one-two months retrofit a time. So The only thing I would say what makes it questionable is the primary substations in Childress that they use for the Bitcoin mining operations. There is 2 substations, two primary substations that they're going to use for Horizon 5 and 6. And those are redundant with two transformers. And then there is the Block 3 on the far right of the site where they're going to deploy the 17,000 B3 hundreds from the 50,000 batch. And, you know, in that picture they they sent in the they showed in the in the earnings presentation, I think it was only two data centers. But so what we've seen happening right now is that they're retrofitting four data centers in that corner. So that's starts I started to wonder, why are they retrofitting 4 data centers or they only announced 2? And the hash rate is also plummeting. So it looks like they are really moving forward with retrofitting 2027 capacity already in the middle of 2026, which is kind of weird, right? Because there are no GPUs yet coming into the site for at least, well, six months. So, the speed to deployment, I think it will depend if they can just use the primary substations from block one and two, which only have one transformer. Maybe the whole delay for 2026 and seven for those, I mean, for the 190 megawatt in 2027 comes down to maybe primary substation redundancy or maybe generator availability. I don't know what it is. Maybe it's just the GPUs, but retrofitting doesn't take a lot of time if you have everything you need on site, right? So it is going to be depending on when these components come in, when those components arrive. And that's why, yeah, I don't want to... sell my own story again, but that's what we do in only France, right? These $800 satellite images, you can literally count the components that are on site. So for example, this is not very material, but just to give you an example, Horizon 1 to 4 has every building has an MER building, you can just consider that a UPS building. And on top, they put three air conditioners. And those air conditioners are from a certain brand. And we found a model and we can literally count them on the site, scattered across 3 laydown yards. And, you know, so we know that they have the air conditioners for the Mayor buildings. for all the horizons already on site. So that's how you can estimate, like we can see they have this much of this component. So that means that this will not delay them once they start to construct or start to retrofit. So it's just going to be something to keep an eye on, right? So there is nothing materially holding them back outside of components arriving. and that includes GPUs, it includes generators. But the retrofit time itself is not something that, it's not hard to do that. So, yeah. And I guess to the point of Sweetwater, I think this goes back to my previous point I made. I think Iron is only announcing capacity that they know for 100% sure that they can deploy. So The 300 megawatt in Sweetwater is a gross capacity, obviously. And it's a day budget with 1.5 PUE for 200 megawatt of IT load. But I guess it is, we should focus on the 200 megawatt. But can they build more next year? I think you're going to see partially builds for future deployments. So most likely you're going to see more buildings being constructed next year, or even, you're going to start. I mean, we already see foundations of much more than 300 megawatt now on the site, to give you an example. I don't have this picture here in front of me now, but if you remember the rendering from the presentation, it was one long building. on the right side of the site. And then on the left, you saw another building, but that was not green. That was like sort of grayed out, like that's not our 2027 deployment. But that building is still being prepared today. So they already have the groundworks being done for the building that is not part of the 2027 capacity. And they've already started that groundwork in 2026. So I think there is a possibility that they will build more than the announced 300 megawatts in 2027 at SweetWater. But I just think Iron is only announcing what they know for 100% sure that's going to happen. So if you remember with Bitcoin mining, they also said we are going to do 30 megawatts in 2025. They said this in the beginning of 2024 or something. And then they said, no, actually, we're going to do 50 exo hash. Oh, actually, we're going to do 52. Oh, actually, we're going to do 57. Oh, actually, we're going to go back to 52. Oh, sorry, guys, we're only going to do 50. So this is not like, oh, we said it once. So it's only going to follow this this plan, plans change. And right now, plans being changed is more of a luxury position because of the very bare minimum of 300 megawatts next year on a site of 1.4 gigawatt that's energized in April 2026, you know, they are obviously only announcing things that they can know for certain that they can deliver. So maybe later this year or early next year, they say, actually, we found out that we were able to, and we're actually going to build 400 or 500 megawatts. I mean, I'm not saying that this is going to happen, but I wouldn't just go by their announced guidance as a fixed amount of capacity. Just because they said it once doesn't mean it's going to, stay. It's only going to be that amount. I think it's just going to depend on, their ability to procure everything for more capacity, I guess. But, is it really so bad? I mean, if you see the year-over-year comparison between 2026 deployments 2027 capacity. I mean, it's still a lot of retrofit, but the self-build is going to go up massively, right? I mean, 150 megawatt in Childress and 300 in Sweetwater is 450 megawatts of self-build capacity in a single year. I mean, that's showing some really strong growth, right? So I guess it depends. One final thing I'll say about Sweetwater, that's the thing I can't really understand. They have another data center area that is next to the bulk substation, and they have not talked about that, and they have not shown it in any presentation. But there is a primary substation there, and I think they could deploy like 150 or 200 megawatts of compute there. And it's not part of 2026 or 2027 for some reason. So this is the dark horse that I am anticipating. That's also why we continue to look at this site, because If we see any sign of construction there, then we know that there is more planned than they have announced. So yeah, we'll keep an eye on it.

₿itcoin ₿utcher: Sounds good. Mark had a question. Mark, what do you got?

Mark: Yeah, hi, Franz. I heard you mention 1.5 PUE. Was that in reference to Sweetwater One? Because Nvidia's internal documentation, they're looking at under 1.2 for the VR200 AI factory. What were your thoughts on that?

Frans Bakker: Yeah, so when I say PUE, I don't really mean PUE as a metric. It just means Iron sells IT load and then they budget gross power around that. just to have sufficient power and never ever oversubscribe a single megawatt. So for Microsoft, Kent has stated that for the most part, this will run below 1.2, but in peak summer, it will be 1.32 or something or 1.3. So you get to a blended average of 1.23 or 1.24. I think that's something Maybe I'm off with the numbers, but that is the actual PUE when you go from the IT load and then you calculate back to the gross power that they needed to facilitate that. So you cannot look at a deployment from a gross first perspective, but that's seemingly what Iron's doing when they say we it is 300 megawatts for 200 megawatt IT load. if you do the math, it sounds like it's a PUE 1.5, but it's just their budget for the contract. And the same thing happens in Sweetwater. They are not going to use 300 megawatt gross power to deliver 200 megawatt of IT load. They are just budgeting 300. out of the 1400 to, for this 200 megawatt deployment. That means that they have a lot of leeway inside of the 300 megawatt to always have enough power to, supply this contract. So If you are very constrained in your megawatts, you would never do that because you would probably oversubscribe your megawatts. That would mean if you have, let's say you have 300 megawatts of gross power and that's your whole site, and then you sell a contract for 250 megawatt of IT load, or let's say 200 megawatt of IT load, and that leaves you still a lot of power between the gross and the actual IT load usage. So there are companies that oversubscribe their megawatts and they will lease out the gap between the actual usage of their contract and the budgeted megawatts to, for example, to on-demand or something like that. So then you get Only in peak summer, when the original contract is taking up too much of the power, that means that they need to, throttle GPUs that they have used for on-demand or something like that. I mean, I can probably word it better if I would take a little bit more time, but the point I'm making here, it's not actual PUE, it's just Iron says that they will, budget 300 megawatt to deliver a 200 megawatt IT contract. That's what it comes down to. So that's also why I don't really like to talk about gross power because it makes the deployment look so much worse. You know, so maybe that answers your question.

Mark: Absolutely. No, thank you for your insight on that. I was just I'm always looking at the specs that Nvidia is expecting with, you know, their with their system. So I was just wanting to get a little clarification on your thoughts there.

Frans Bakker: Yeah, the one final thing I'll say, that's my that's that is really where my gripe with Nibi started. is when they announced that their Microsoft top line, people started to look at the facility and then they said, it is a 300 megawatts IT load, a 300 megawatt gross deployment. But GPU contracts don't care about gross megawatts. That is something for the operator to budget. First and foremost is about GPU hours. And secondly, that can be calculated towards IT load. And obviously, the gross power is what is using the electricity. But yeah. the vagueness around gross and IT load across all these operators is very annoying. But if you don't know, how much power is being actually deployed and you don't know the GPU count, you cannot say a single thing about profitability or how good a contract is. And For that reason, I am very happy that Iron was able to tell us that it was 60 megawatt of gross power for NVIDIA because, you know, if you apply a 1.25 PUE on top of that, then you start to see that this is the actual usage, right? And then on an IT load basis, these contract just looks phenomenal, right? You're looking at $15 million a megawatts per IT load. And that's basically unprecedented in AI Cloud. So yeah, I think we should be really proud of that. And I think it's only the first deal and many more will follow.

₿itcoin ₿utcher: Frans, I had a follow-up question related to what Mark said. If you do some quick math, They budget 300 gross for 200 at Sweetwater One, but you're suggesting in actuality it's most likely going to be one and a quarter. So my question to you is if it comes out to one and a quarter, one and a quarter times the 200 would come out to That's 250 gross, if that's the actual PUE. So you have 50 extra megawatts of buffer or slack, if you will. My question is, when they're budgeting that, that extra 50, is that potential upside for their ARR, like they're able to fit in? more Vera Rubens potentially, or they're able to contract those out to someone else. How do you think of that? Because it used to be maybe they were going to plug minors in for spare capacity during downtime, but this is kind of a little different where they're almost being overly conservative with what they're guiding. Like 300 gross for 200 critical seems overly conservative. based off of what you're saying. How do you think of that or like that extra 50 gross potentially? Because I mean, like, yes, iron has no shortage of power, but at the same time, like they still want to monetize their megawatts to most efficiently, like those guys are too **** to just leave 50 megawatts, let it be, and just be like, well, we only needed 250 here. There's 50 extras. So how does that kind of reconcile itself? Because Horizon 1 through 4 was a little different where they fit in a particular number of GB300s. specifically because the racks were flexible and were able to scale up to Vera Rubin if they were to substitute. So that was almost intentional because of Rubin and how many racks they wanted to fit in the data all. But Sweetwater one is designed specifically for Rubin and there won't be any GB 300. So I just can't help but ask out loud, like if Jensen's trying to create this flagship property, I don't think they're going to squeeze every single Vera Rubin they can per megawatt, but I also don't think they're going to underutilize space. Like what's the healthy medium there and how do you think they reconcile that?

Frans Bakker: Well, I just know that for the Microsoft contracts specifically, Iron has said that they will not oversubscribe megawatts. And basically what they're doing is undersubscribing megawatts, right? So I wouldn't call it underutilized, but it just means that on a project level that they have reserved 300 megawatts of gross power for it. But that's not to say that on a site level, at some point when the whole site is deployed and they start to see that, okay, we ran horizon one to four now for a year, it's never going over 250 megawatts of gross power usage. Maybe we can start to monetize that gap. But that's currently just not what they're doing. And I don't know if that is from a principal standpoint or if it's from a Vera Rubin standpoint. So I don't exactly know how it's going to change if there would be a substitution clause. I think that the rack amount is fixed. So that would mean if my calculations are right and every one building has 11 racks, sorry, 11 groups of eight racks, so 88 racks, times 72 GPUs is 6,336 GPUs per building, times 12 buildings is 76,032 GPUs, which is basically exactly perfect. But my internal research has also indicated that there is a small chance that there is actually room for 12 racks in every building. So if you put 12 sorry, not 12 racks, 12 clusters or pods, let's call them pods, 12 pods of eight racks. Then you get to 96 racks in a building, times 72 GPUs is 6,912 GPUs in a building, times 12 buildings, so you get to 82,944, which is a lot more than that they have disclosed, that they have bought, et cetera, et cetera. So that doesn't look to be happening, but Let's just entertain the idea that Iron is going for Vera Rubin in Horizon 3 and 4. Could that also mean that they could unlock, you know, the extra rack and sorry, the extra part of eight racks in those buildings, which means they can up the GPU count and they can up the IT load and that still falls within the 300 megawatt. You know what I mean? Maybe the 300 megawatt is not just a budget for the current deployment. Maybe it's also headroom for a full deployment of Fierra Rubens in these shells. But I think this is maybe a little extreme. I mean, maybe this is not realistic because I don't know if a single building can actually run that much IT load. to the tune of 12 times 96 NVL VR 200. So that's something I don't know, but to your, to answer your question, for Microsoft, they are not, they have said that they're not going to oversubscribe megawatts. I don't know if they're going to do it in Sweetwater. I think they just want to be on the safe side. You know, they want to rule out any kind of overdrawing from the grid. they don't want to violate their interconnection agreement even by a single MW. They just want to be a good grid citizen, and they don't, even if they have an excess of power on the site, I don't think that they're going to, risk on either side, the contract side with Microsoft or the grid side, with the utility. So I think they're acting from a position of a wealth of megawatts and they're under subscribing their power to have a budget. And within that budget, they have a lot of, headroom, but also flexibility. So, I know what you're saying and it make it doesn't make sense if you think about Jensen and the token per megawatts and everyone is scrambling for power and Iron is just over budgeting or under budgeting their No, under-utilizing and over-budgeting their gross power, right? I mean, it doesn't seem to make sense because why would you have why would you budget 1.5 PUE when you say when you tell the market is 1.32 at peak, right? Just it's it seems excessive to leave that much headroom, even if you cannot predict the weather or whatever. It just doesn't seem efficient, right? So, Yeah, I don't know. what the internal idea is behind this, but I would argue that maybe at some point on a side level they could say, look guys, we have a 1.4 gigawatt interconnection agreement and we continue to draw only 1.25 because of this. You know, maybe they will squeeze out another 100 megawatt when all things are said and done and the dust is settled, I just don't know how they are looking at this thing, Maybe 100 or 150 megawatts are immaterial to iron on a side level like Sweetwater, right? And they'd rather be safe than sorry. And yeah, I don't know. I guess that's some considerations. I don't know really the answer, but I just know they're being conservative. for now.

₿itcoin ₿utcher: I personally view it as an upside scenario that most of the market's probably not paying attention to right now. But to Mark's point, if you're going to team up with Jensen and have the most efficient design and architecture and then the most efficient hardware being built to or I should say servers to run these GPUs. I just have a hard time believing that the flagship site for an Nvidia AI factory is going to underutilize megawatts by a material amount. And to me, as a shareholder, hearing that if they're running at 125, like I can understand year one, but if they're just simply throwing away 50 megawatts of capacity out of overabundance of conservatism. I view that as money being left on the table. Certainly I'm oversimplifying it and I don't do it for a living, but I think it's at least worth the question of those guys next time we have a chance to talk to them, how they think of that and where does that, you know, how just I would be interested in hearing the question I posed to you, like how Dan thinks about that or how Kent thinks about that. Because ultimately, the other piece of it is, all things being equal, if we're waiting for more capacity to come online and let's say Australia in 2028 and we're trying to squeeze out more earnings or revenue growth and we are underutilizing our existing portfolio, it seems rational to be searching for additional megawatts within those sites so that they can have more GPUs and create more revenue and make everyone in here a little more money. But Franz, you wanted to add something to that.

Frans Bakker: Let's just assume that Iron is going to sell SweetWater in parts, compartments of 300 megawatts, right? It grows. That would mean that they can sell 4 and then they only have 200 megawatt left. But if there is a POE of 1.25, you could probably do 5 deployments of 200 megawatt. And then you would get to 12150. So I guess it will be somewhere in the middle, right? So if you only factor 1.5, you would get to 1500 megawatts for five phases of 300 megawatts. If you would only factor 1.25, you would get to 1250 megawatts. So what's between 1250 and 1500? I think 1400 megawatt is probably a very good, you know, direction. So I think ultimately they would, it would be likely if I look at it like this, that they would go for a 1400 megawatt gross to deliver 5 times 200 megawatt of IT load. So I think on a side level, it would make sense that it would be, it would come out to be a GW of IT load. And that also you know, in line is in line with what Dan has been repeating many times in his last interviews, a gigawatt of AI computer, gigawatt of computer gigawatts. You know, I think they're going to sell five times 200 megawatt I.T. load at Sweetwater. So yes, it is not squeezing out the maximum, but it's also not a budgeting 300 megawatt for every 200 megawatt of IT loads. So that's my anticipation and I think it also fits the circumstantial evidence that we have. So I guess we should just consider. They will do what's in the best interest of the portfolio. I don't think that we should just take the budget of Horizon 1 to 4 as the way forward from now, you know? maybe things will start to, I think the NVIDIA deployment is a much better way to look at it, right? That's 1.25 POE and they've budgeted 1.25. So yeah.

₿itcoin ₿utcher: We're pushing 2 hours here. I will go back to the panel. Any closing thoughts, Franz, stock analysts or Mark? Thanks for joining me tonight, guys.

Frans Bakker: Yeah, I have talked too much. So I will, I thought about going to talk about this post that someone sent me a post that explained what happened on Friday, which I really subscribed to, the entire perfect storm that we had with the markets. But I guess we can leave that for another time. The problem with me on these iron spaces is as soon as we get into an area where I can talk about, I talk too long and too much and we get very, it starts to become very detailed. So I'll try to keep it a bit more high level next time. But thanks for hosting and I hope to host one next week.

₿itcoin ₿utcher: Yeah, maybe just repost or you can send along and we can send to people who are looking through the tea leaves. But I certainly think a long story short, the option market, you know, exasperates, moves up and moves down and it's a two way street is my thought on that. Mark, thanks again. Two weeks in a row. You're a nice injection of energy and stock analyst. You're always welcome as well. We'll talk to everyone soon. Have a great week.