← Blockspace: AI & Bitcoin
Blockspace: AI & Bitcoin
Strategy Opens Door to Sell $1.25B BTC, SpaceX Eyes Mesh Optical, Understanding GPU Financing
01:22:46
Loading summary
Transcript133 lines
- [00:53]
A
Mother.
- [01:00]
B
All right, what's up y'?
- [01:01]
C
All?
- [01:01]
B
Welcome back to Block Space Live. Running this one more time, looks like Psyop anime might have a little bit of competition in the bitcoin ecosystem. Yes, that's right. We will be talking about Strategy and Stretch today, specifically Strategy's new five point plan for capital management. Guess what? If you listen to Block Space Live last week one of our guests actually said this is precisely what they need to start doing. So make sure to tune into us to stay ahead of the alpha. That is not our lead story though. We will actually start off with a stock market update as bitcoin miners turned AI stocks are bleeding today. They've started rallying actually as we started recording, but many names are down over the opening of trading on Monday. But for our most interesting and lead story today, SpaceX or specifically Elon Musk, wants to acquire an optics company. But it's not fiber optics, as my co host Charlie Spears will explain. It's actually laser optics and it's the latest in SpaceX's attempt to become a full stack or as much of a full stack AI company as they can be. Following that, we've got a slew of interviews today. We have Francis Corvino of Lygos Finance on to talk about Strategy's new capital plan. Following that, we have Ethan Vera of Luxor on to discuss AI financing and GPU financing specifically. We will then move into our second segment for news today, which is Bit Deer announcing an AI tenant, or rather saying they've signed an agreement for an AI tenant at their Tidal Norway facility. And we will wrap up with Aydin Kill, president of Hive to talk about their recent acquisition of their Bowdoin Sweden campus, their convertible note to revamp that campus and and what the timeline for that will be now that they have an LOI for a tenant at that facility.
- [02:49]
A
That's right. BlockSpace goes live every weekday at 1pm Eastern featuring quick hits on AI data centers, some bitcoin, bitcoin tech markets and emerging technology. Make sure to like and subscribe. Hit the notification bell to get the little push notification on your phone whenever we go live on YouTube and chat in the comments. Sometimes if we see something interesting, we'll highlight it and we'll, we'll bring your comment up onto the main stage. If you like the pod, if you like the live stream, it'll turn to a podcast shortly thereafter. Anywhere podcasts are stream, Spotify, Apple, etc. And if you like the podcast, you'll love the newsletter, newsletter, dot blogspacemedia, dot com the this show is brought to you by CleanSpark. Nasdaq listed ticker CLSK. More on CleanSpark later on in the show. Colin I love that we don't cover crypto and bitcoin anymore because that means I can stop looking at the markets on the weekends.
- [03:53]
B
Yeah, and you can stop being depressed when you look at everything else rallying and bitcoin getting crushed. Thanks Michael Sailor thank you. That being said, we will be looking at some bitcoin mining turned AI stocks today. Right before we get to everyone's favorite least worst trillionaire or erstwhile trillionaire only trillionaire so far your favorite worst nightmare, Elon Musk, according to your political leanings or your savior. But enough on that. We will get to that in a second. I just wanted to throw up tradingview really quickly and actually the names are not looking as bad as they did upon open. All of these stocks just cratered as the stock market open today. I've got the usual suspects up here. I've got Keel, Galaxy, Wolf, Cipher, Hut, Riot, Marahive, Bitsier, Core Weave, Nebius and Iron, but they've actually started recovering kind of as we started the show. So thank you. Your bags can thank us for going live to show how the market is reacting on Monday. Seems that actually us bringing it up to show them reverting is now leading into a kind of V shaped recovery. But the only one that's currently up on the day is Bit Deer. And I wonder, I assume that's related to the announcement they had today of signing an agreement with an AI tenant for their Tidal Norway facility. But if you looked at this, they were actually trading down as the wider market was actually trading up on hopes that this actual peace deal in Iran will actually stick. Right now though, most of the names lagging, Nevius and Core Weave just popped up into 0.5% return on the day territory. Mara Flat and the rest of them are still in the negative. From what I saw from chatter on social media, there wasn't necessarily any huge news. I wonder if some of these are just trading down because they're proxies to bitcoin a little bit. Still when bitcoin sells off, even though they are moving to AI, they still get a little bit of a hit as a result. But so far it looks like kind of just a minor correction, nothing crazy going on.
- [05:55]
A
Yeah, and if we look at the diametrics dashboard we can see most categories have kind of been flat or slightly down in the past day or week. So we have powershells up 2%, neoclouds down 8.2% quite a bit. Kind of across the board hyperscalers. That's an interesting1. Down 16.1% on the month, mostly flat on the day. So markets interesting again it's just Monday. A lot can happen in a week. As for what's happening today, Colin, I think we jump into SpaceX or really this one is an Elon story. So for anyone who didn't catch this, Data Center Dynamics reports that Elon Musk gets FTC nod to acquire datacenter optics business Mesh Optical. What is this deal? First of all note this is Elon himself personally. This is not SpaceX doing the acquisition. This is a company that specializes in what are called data center optical transceivers. Mesh Optical technology is co founded by Travis Beshears and Cameron Ramos. They previously did a lot of work for the Starlink. So the Starlink satellites communicate with basically lasers and use these transceivers. Between the Starlink satellites and the guys who headed up that project and department, this is their company recently raised $50 million. So Colin, why is this a big deal in the context of like data centers and Elon's full stack data center business? With SpaceX we have to get into a little bit of the technical side of how a data center runs and is built. First of all, you have all the GPUs, all the compute in the data centers, but all that compute has to talk to each other. Traditionally and largely we use copper wires to connect the GPUs and then and the things the GPUs are coordinated with together these copper wires are a considerable amount of the overall electricity consumption of the entire site and they spin off a lot of heat. So if you're thinking about what this looks like in space or just how data centers operate more efficiently in general, the fact that copper wires are such a huge heat and power source is an issue. So if you can replace that with optical communication, mostly produced by these optical transceivers, you can save a ton on power cost and on heat. Some numbers for you GPUs in an average cluster, again it kind of depends will consume up to 40 to 50% of the overall power of that site, which is different than bitcoin mining where your Compute itself, the ASICS are 99.5% of all the power power drop in a. In an AI cluster the GPUs are half or slightly less than half. The rest of the cluster is servers, memory, storage, networking. Networking alone takes up 10 to 50% of the power draw of a site. So if you're looking at 100 megawatt site for AI training or inference you'd be looking at 10 of those megawatts just ballpark hand wavy being just networking. And then we have about 20 to 30% which is cooling. So if we can reduce both those numbers with optical transceivers, that is huge. I'm going to sound like a big SpaceX bull. And the thing is, candidly, I only really started looking into optical networking in the past 24 hours, so I'm not an expert yet. But you can see the play and trajectory that Elon is taking here. Colin, I'll throw it to you for some comments.
- [09:57]
B
It's the Tony Starkification of the AI space, no, but in all seriousness, to me it seems to be an indication of just two things. One, a lot of the tech stacks that Elon Musk's companies deal with are really ahead of the curve on number of verticals. And this also to me shows that the company is very interested in trying to become as fully integrated in the data center space as possible. And then specifically within the AI space, you know, their cursor acquisition recently shown that they want to kind of own part of the AI software stack. They run their own model and with Clock or sorry, excuse me, with Groq, they also rent out GPU space. They run the actual data centers. They're now looking at actually becoming a gas company so that they can make sure that they have enough power to power these data centers. And then when you look at some of the far flung aspirations, they are now acquiring a company that will help with the tech stack for actually operating the kind of grand vision of XAI operating data centers in space. The last thing that is interesting to note, the regulatory pathway for this in terms of why Elon Musk is deciding to purchase the company rather than do it through SpaceX SpaceX. As I understand it, this actually makes it a little bit easier to get this, this acquisition approved, assuming it goes through. And I just going back to the Tony Stark comment, it is just wild to see all of these really cutting edge technologies operating under one umbrella. And I think that's part of the reason also why Elon Musk is such a contentious figure with a lot of people, because most people are probably not equipped enough to evaluate all of these different tech stacks and their potential.
- [11:40]
A
I mean I only just, I was just looking at the optics by selfless boarding. So like even I didn't have the, the tools to really.
- [11:48]
B
Yeah, and you know, frankly, most people don't. And the fact of the matter is you either are going to discount or probably you're either going to discount or value that maybe too much, depending on where you fall on the spectrum. And so you, it's very easy to say, oh well, none of these things will end up working. This is all very, you know, high flung and this is, these are kind of pie in the sky technologies. Or you think that Elon Musk literally is Tony Stark and so you think betting against them is insane, which has been largely true. Although as we've seen recently, the market is really in this liminal space in terms of where it can value or where it thinks to value. SpaceX. Right now the float on the stock is still like just under 5% and we'll see a bunch of share unlocks later this year. But their bonds are having trouble. The bond issuance that they actually just issued last week, we covered that on Friday. How some of those are already underwater and the stock is obviously down from its all time high shortly after ipo. And which I think anyone who had enough brain cells to rub together probably expected.
- [12:49]
A
But I mean, you make the Tony Stark comment and I'll, I'll say that Tony Stark, prior to his capture and invention of the Iron man suit was actually kind of a villain. He was a hyper capitalist weapons manufacturer, if you remember, I guess. Well, we've, we've got Francis Corvino in the wings. We're going to talk about strategy in just a moment. I think the one last thing on this SpaceX or this Elon acquisition is that in the broader context of compute and GPU scaling, a lot of folks think it's just a problem of getting as many GPUs onto site as possible. In my recent, again fairly recent research on networking, apparently communication between clusters is just as much of a bottleneck as actual physical scaling of the dies to actual energy production and routing through these GPUs. Apparently how they talk to each other is just as much of a constraint. So as the AI workloads get bigger and bigger and bigger, and your 100 megawatts baseline now becomes the 1 gigawatt baseline up in order of magnitude, then the communication between these AI clusters, these clusters becomes a huge challenge. So I think that probably wraps it up. We've got Francis in the wings. We're talking about stretch and strategy. After a word from our sponsor, CleanSpark.
- [14:29]
B
We are CleanSpark, America's Bitcoin miner, a publicly traded company with the largest operating
- [14:36]
D
hash Rate powered entirely by self operated infrastructure across four states.
- [14:45]
B
This is our proof of work and
- [14:48]
D
we are setting the standard for what's next. Learn more about the intersection of energy and bitcoin@cleanspark.com.
- [15:01]
B
If Bitcoin's actually the best money and
- [15:04]
D
it's the thing that people should accumulate
- [15:05]
B
and it's the best risk adjusted asset,
- [15:07]
D
I lose zero sleep about whether or
- [15:09]
E
not that's gonna happen.
- [15:09]
B
I just ask the question of when. It's literally matrix math that you're running
- [15:16]
D
on large pieces of data. The bitcoin miners can absorb that energy.
- [15:20]
A
And in many ways this feels like
- [15:22]
B
a second bite at the apple to build a new Internet.
- [15:34]
A
All right, we have France in the wings. Fingers crossed. We have audio set up. Francis, welcome to the show.
- [15:42]
C
Hello, sir, how are you?
- [15:43]
A
Hey, good to see you. Okay, so Sailor and Fong Lee were very busy over the weekend, it seems, because everybody was talking about stretch. It got low in the 70s, where I believe it still is. And we have this Morning Sailor and MicroStrategy dropping the five point plan and simultaneously finger in the wind saying that they are cleared to sell $1.25 billion of Bitcoin, as I understand. So I'm gonna throw this at you. Which of these do you want to talk first? Do you want to talk about did this five point plan and, and what you think about it?
- [16:24]
C
I think the, the right thing to talk about maybe is like posturing versus okay, action. Right. So ultimately, he and MicroStrategy has opened the door to allow themselves to sell bitcoin, whether it be today, tomorrow in a bear market or in a bull market. They've fully given themselves the ability to do that. And then I think also vitally, they've given themselves two years of Runway on the stretch, essentially the dividend obligations. And I think the market has taken to that somewhat positively. That being said, ultimately what they did was not sell bitcoin. They opened the door to sell bitcoin. What they did was something that was massively dilutive, right? They raised, you know, a billion and a half dollars in order to get to that two year mark with just pure sales of equity. And as a result, MNAV of MSTR has, has dipped below 1 for the first time. So now Saylor and MicroStrategy are fully beginning to feel the end of their ability to sell pure equity from MicroStrategy. And they've sort of bottomed that. And now they have opened the door to beginning to sell Bitcoin because shares of MicroStrategy are worth less per dollar, essentially, than the bitcoin that underlies them. So it makes more sense to sell that bitcoin now than it does to sell shares, because now we're. We've tipped. We're no longer at 1.2 m. Nav. We're at sub one.
- [17:52]
B
Yeah. And they just bought bitcoin last week. It was only $35 million. But the most material buy they've had as of late was over a month ago on May 18th for 2 billion. So at 80,000, basically $81,000. So this, you know, if they were just at 1.25 billion, you know, they're selling almost roughly half of what they've bought in the last month and a half, with a few other purchases over $100 million. I mean, it seems like trends as the market is reacting somewhat positively towards this. Maybe this is just bitcoin rallying a little bit. It's only up 1% today, but stretch is actually up 11.5% on the day since opening. 83.16. What's your read on that? Do you think that's actual confidence from investors, or do you think this is just. Stretch was oversold. Now people are taking opportunity to swing it the other way for a trade.
- [18:42]
C
I mean, there's a. There's a material news event, right, which is positive for stretch holders, which is negative for, to some extent, strategy holders. Right. So I think that the most important thing that we're seeing, like, honestly, 11% up on a. On a. A security, which is supposed to trade at 100, which now has two years of Runway, is. Is almost surprisingly little, given the amount of conviction that sailors trying to exhibit over his, like, you know, willingness to continue to pay that obligation and not just like, sort of let it. Let it dwell or let it disappear. He kind of floated in Prague we could always pause the dividend based off of this posturing. It doesn't look like he's actually, you know, going to pause the dividend. So to some extent, I think that the market is responding somewhat positively to what I would say for them specifically is overwhelmingly bullish news. If you're a microstrategy shareholder, I think that it's sort of the opposite. But you could also see this as being the end of significant dilution events. Right. Like, we know now we're below 1m nav. We've opened the door to selling bitcoin. The dilution will stop at the very least, but you may begin to unlever a little bit by unloading some of the Bitcoin into the market and then unfortunately dealing with some of the reverse reflexivity that is a result of selling bitcoin and seeing MicroStrategy shares go down more than Bitcoin. Right. On the way up, MicroStrategy was able to benefit from buy more Bitcoin, Bitcoin goes up, Microstrategy shares go up more credit profile of MicroStrategy increases, tap the whatever funding market you want, issue more shares, issue more stretch and then buy more Bitcoin and see Bitcoin pump again, increasing the shares of Microstrategy again more than just the underlying Bitcoin. But now it's the other way. You sell Bitcoin, you're going to crash. MicroStrategy shares more than the price of Bitcoin went down, which you know, has that sort of negative reflex loop. But you know, as when the market tides swing one direction, they can always begin to swing the other direction. So with a lot of pressure on MicroStrategy to the downside, you know, it could be like a top underwater at some point and we could start to see it, it pushed back up. But that's unclear, right? Depends. Depends when they start to dip into that Bitcoin reserve, right. Like they have given themselves, I think. What is it? Is it 17 and a half months of Runway?
- [21:15]
B
Now assume, assuming they sell the 1.25 Bitcoin, I assume, right?
- [21:20]
C
No, so they have already sold, I believe it's $1.15 billion worth of, worth of MicroStrategy equity which has given them
- [21:28]
A
a 2 point something billion dollar.
- [21:30]
C
Yeah, a $2.25 billion, $2.55 billion
- [21:35]
A
USD reserve.
- [21:36]
E
Yeah.
- [21:37]
C
So that gives them a continued period with which that they can begin to or continue to fund the dividend. However, when they decide to dip back into that, the money jar, sell the bitcoin, sell the family jewelry. Right. MicroStrategy is beginning to operate sort of like a household. Right. The first thing that you do is you spend your paycheck for MicroStrategy. Its paycheck was probably the stretch, right. You sell stretch to pay the dividend of stretch and also to buy more bitcoin. Well now, okay, we're running through the paycheck. Okay, well now let's get into the savings. So that's the treasury, right. And first MicroStrategy essentially paid off all of like a specific series of debt. I forget exactly how much it was, but that went through the vast majority of the treasury and it brought it down to about six months later, left Right. They couldn't issue any more equity, so they don't have too much salary left coming in. And their treasury was starting to disappear and was down to six months. But now they said, okay, we're going to. We're going to sell the last. We're going to scrape the last amount of our income that we possibly can. But they're about to get to the point where they're selling the family heirlooms, which is the bitcoin, because they don't have any more of the spigots to tap into in order to continue to bring cash into the company, in order to continue to pay obligations.
- [22:55]
A
And this goes both ways. So the thing is, Bitcoin hit 126k. Saylor didn't sell any. And a lot of that was just because he was absolutely gung ho about never selling any bitcoin. Now, that is. I know it was backed up against the wall, but now that his hand is forced and they are doing this, what does this look like in a bitcoin bull scenario? I mean, imagine bitcoin's going up. Now we have an interesting scenario where the seal has been broken. Saylor can now sell without the market questioning if this is the right decision, because he's already done it now at 60k bitcoin or.
- [23:37]
C
So
- [23:40]
A
imagine the scenario where bitcoin's in a bull market and now Saylor has a different strategy, which is selling digital credit.
- [23:48]
C
Yeah. I mean, it depends to what extent he continues to actually act on the idea of selling bitcoin. But it's important to note that we're right now in a bear market. So the downsides of this can come out before, before the, you know, the positive side. Right. If six months goes by, a year goes by, we're still sitting at trading at $60,000. He has a bunch of converts that he's gonna. He's gonna need to pay out. So that's gonna. That's gonna be a significant tap on, on his reserves as well. But then if we do see bitcoin begin to rebound to the upside, there is a bit of a compression on the potential high end because you do now begin to factor in this potential seller of bitcoin. So it can dampen, it can dampen upside over the long term. And over the short term, it doesn't actually do much to dampen downside. Right. But if we see bitcoin begin to jump back up and he actually does begin to sell some degree of bitcoin, that should dampen volatility to both the upside and the downside.
- [24:48]
B
Francis, Perhaps as a closing question, unless Charlie has anything else, going back to the household analogy, does Saylor have to choose between his two favorite children here? I mean, when you look at Stretch and the strategy common stock, if you look at the financial mechanisms, they seem to be at odds with each other. If you want to fund the dividends for Stretch, you either have to sell common stock or, or you have to sell Bitcoin, which is the whole reason why the common stock has any value. And then, well, are the two favorite
- [25:16]
C
children in this case stretch and MicroStrategy Equity, or the two favorite children MicroStrategy Equity and Bitcoin? Because I think those are the two that he really.
- [25:24]
B
That's interesting. Yeah. So maybe unpack that I was talking about, let's say just for the point of the view, the equity itself, are the children.
- [25:35]
C
Okay? Well, yeah, I mean, to some extent, if bitcoin dips anymore, then he absolutely does. Right. Like because there's, there's a third child in there as well which are essentially converts which if bitcoin is trading sub 60k, he's going to need to pay out a good percentage of and that's going to take a lot, a lot of cash, right? He has to pay that out. That's real debt. Right. Then he goes and he looks at his equity stack, right? And he has Stretch and he has pure equity in mstr. And if he's already paid out both, if he's already paid out the converts, it's going to be very, very challenging and he absolutely is going to have to choose between those two favorite children. However, if Bitcoin can stay above $60,000 or can stay above sort of the convert, like the convert positive side, then he probably doesn't, Right? Like he has a good amount. He's really rebuilt the treasury quickly with this issuance. The question is like now he's sort of at the end, he can't do much more with what he has. So he needs Bitcoin to continue to rise from here or start to rise from here. If it goes down any more from here, he's going to have to choose between the two favorite children over a two year period.
- [26:48]
A
My last question, we had Thomas Brazil on, on Friday, really good interview. And he said Sale has to rip the band aid off. It looks like he's about, he's ripping the bandit off here. Seems like the right decision. But zooming out, what does this do to the market's confidence in Saylor's conviction about Bitcoin Isn't that, hasn't that been the foundational brand of Michael Saylor and strategy or can this evolve?
- [27:20]
C
Well, I mean, I don't think that this, to me questions Saylor's commitment or conviction in the Bitcoin trademark long term. The way I see this is Saylor has essentially been taken kicking and screaming to the point where he has to sell his bitcoin and he's doing it at the last moment where one could possibly describe it as still being a responsible financial actor, conviction wise. I don't think Saylor's conviction has gone anywhere. The only thing that Saylor is trying to be mindful of at this point is being stopped out on his trade because ultimately this is a leveraged bet on Bitcoin and if you get stopped out, then you're no longer playing. And I think the number one rule in Bitcoin over the last, you know, 10, 15 years has been just be able to continue playing. So it's important that sailor is cognizant of the importance of being able to continue to play the game for another hand rather than being stopped out on a levered long. But the conviction is clearly there or I don't think he, he, he would have made this decision much earlier.
- [28:17]
A
Francis, thank you so much. Glad to have you on the show. You have been our sailor whisperer for some time.
- [28:24]
C
Thank you.
- [28:24]
A
Cheers guys, Appreciate it.
- [28:26]
B
Thanks, Francis.
- [28:27]
A
See you next week. Love me some Francis Corvino. Shout out to Lygos. We'll do an ad read for them in a moment, but before that, we've got Ethan Vera of Luxor in the wings. And before we bring Ethan up, a word from our sponsor, Luxer.
- [28:49]
B
This episode is brought to you by Luxors Commander. Bitcoin miner management software for enterprise operations. Luxors Commander gives you real time fleet monitoring bulk remote commands across your fleet. And Intelligent Miner, that's an automated profitability engine that runs every five minutes and adjusts your power settings to live hash rate markets and energy markets. Ercot backtests show 10% improved profitability with intelligent mining over binary mining. Commander Pro is $100 per megawatt or a 25 basis point pool fee adder and you can try it free for 60 days. If you'd like to get started, go to Luxor Tech forward slash commander to learn more.
- [29:31]
A
All right, we have Ethan in the wings. Let's bring him up. Ethan.
- [29:38]
B
Good.
- [29:38]
A
Gentlemen, hello.
- [29:41]
E
Always hard following up, Francis.
- [29:43]
B
Well, we had the sailor whisper. Now we're going to have the hash rate and GPU whisperer. On specifically to talk about AI financing, which I don't think we can get enough of this topic because half of the news that we cover at this point is financing, more or less. It's financing and deal flow and how the two interact. So to set the stage, Ethan, could you walk us through the financial stacks we're starting to see from NEO clouds and other AI operators. What mix of financing are we seeing between equity, corporate, paper and project level or GPU financing and what are the terms and interest rates that we can usually expect from those different types of levers?
- [30:30]
E
Yeah, and to your first point there, financing is a crucial piece to this. I kind of think of it in four levers, right? You have your, your power and your site, you have your financing, you have the hardware component and then the offtake, like how you actually monetize that hardware. And so financing really enables all of those steps and really the hardware purchase. This space has evolved quite a bit recently. GPU backed financing, where you're collateralizing the compute hardware, really emerged as an asset class in around 2023. Early companies like Macquarie were doing it. You started to see some of the early iterations of like Blackstone and Blue Owl with some of their core weave deals coming out. And it's evolved a lot since then. I would say for the majority of our partners that are looking to deploy new NEO clouds and stand them up, most of them are looking at this on a project basis. So they have a parent company, they're looking at a new deployment that they're going to turn into a NEO cloud and they're looking to finance into that entity structure and core. We've really kind of paved the way for that model. And so you set up a subsidiary and you're raising debt and equity financing directly into that vehicle. Typically what we see is debt providers are providing around 60 to 80% loan to value, meaning if there's a purchase order of say 100 million for both compute Nvidia chips as well as a modular data center, the debt financier is putting in 60 to 80 million of that as kind of collateralized ring fenced debt. From there you have to bridge the rest of the gap, Right? Where does the rest of the 20 to 40 million come from? A lot of that will need to come from equity capital. And debt providers want to see some skin in the game. So that equity capital might come from the parent company making investment down or it might come from directly in equity partner investing. And you can also structure it with a layered prep sleeve so you can have preferred equity and then common equity over top. And then the last component of financing there is really the down payment from the compute contract. Typically what we see for NEO cloud is they're selling forward multiple years of compute, they're getting a down payment, something around 10 to 20% of that contract upfront and they're using that as a way to fund their hardware purchase. So with all of that stack, that's really how Neo clouds are getting off the ground running.
- [32:49]
B
So Ethan, on that note, it seems to me like in theory there's a chicken and egg problem here where you need signed offtake in some cases before you can get financing. But a lot of buyers won't take that until they see chips up and running. How do new entrants actually break? How does a new entrance like break that loop? Or does that not even really a problem in practice because there's so much demand right now for computer.
- [33:19]
E
It's in theory and in practice and it is kind of this challenge where you have three balls in the air. You have your hardware purchase, you have your purchase order just waiting there. You have your financing which usually like a credit fund or a debt provider will go through all the diligence first and then they'll wait for the offtake to be signed. And then lastly is you have your offtake and the offtake wants to see that you have hardware procured. Obviously that requires financing. And so how do you as an up and coming Neo cloud, make sure all those balls land at the same time? There's a few different approaches we see. One approach is to first sign a non binding loi with the offtake. So a NEO cloud will say, okay, I'm going to sell forward three years of my compute. The buyer compute says, okay, I'm interested at these terms. We're going to do non binding for now and we're going to have milestones based on completion that will make it binding and then start to kind of bring in some different down payments as part of that contract. The other way we're seeing is just really if the offtake is very interested in this specific compute cluster and they have a lot of faith in the operator, they might just sign a directly binding MSA right away. And so they might just say, okay, you know what, we are limited on compute, as you mentioned, we're strapped for supply, let's just lock this in. We have faith in these guys and we're going to do it. So there's different approaches there, but it's certainly one of the major challenges and that Chicken and egg problems is a great way to describe it.
- [34:45]
A
Ethan. Let's talk GPU depreciation and lending against that. Because GPUs are iterating very quickly. The compute market is pretty. There's a lot of different types of compute, very different than Bitcoin mining where it's all one hash function. So does this create a mismatch between NeoCloud's lenders since GPUs can depreciate maybe quicker than the loan term because these loan terms are longer than it seems that some of these GPUs can remain relevant.
- [35:23]
E
Yeah, I think we've probably talked about this a few times on the POD around ASIC financing. The classic example we gave was like November 2021, like ASICS and S19. JPro at the time was $100 a TerraHash. So it's $10,000amachine. Lenders were coming out with somewhere between a 1 1/2 to 3 year loan and they were lending at an 80% loan to value. What happened was that over the course of the next 12 months the Asics drew down in price 80% and S19J Pro was trading around $20 a terahash. And so a lot of the borrowers just said hey, this is your collateral package. Take your machines back, I'm going to walk away. And so it's very important that lenders are kind of thinking very strategically about where residual values will be during the contract. Not just at the lifetime, but making sure that they are being thoughtful in that piece. Now there are some key differences here between the AI market and the Bitcoin market. The first thing is the AI market. Most of these companies have long term offtakes, which is like a hedge, kind of what we talk about on the mining side with the derivative instrument. But they basically have a customer that's guaranteed to make these payments over three years. Of course there's some counterparty risk if that counterparty can remain solvent, so there's some cash flow de risking. That's a different story here. And then on the GPU front, lenders need to think through these markets. I mean the technical lifespan of a GPU running In a tier 3 data center is going to be like 10 years plus. So it's not like a question of hardware failure, assuming it's, you know, Nvidia reference architecture, tier 3 standards, etc. It's more of a question of economic lifetime. Is this GPU going to be worth anything by the end of the contract or during it? And then it comes down to looking at different factors coming up in the market, you could start to model different analysis of like, hey, what's the chances Nvidia cuts their margin and significantly reduces prices? We saw IBM do that in the late 70s and it significantly impacted all of the creditors in the space because they cut prices by almost half overnight. Then you could think of the AI bubble. Is it a bubble? Is it going to run out of funding and is there going to be a lot less demand? Meanwhile, all of the supply chains are obviously at full throttle and you're going to see this kind of switch of supply and demand. So there's all these analysis that creditors need to do to think through that residual value and make sure they're deploying capital in a safe way.
- [37:50]
B
Yeah, too many corollaries with the ASIC market. I think maybe the only one being that, you know, obviously the revenues are just so gargantuan and there's more, I don't know. We've seen GPU prices in rental terms per hour change a lot.
- [38:07]
A
They've been up and down all over the place. H1 hundreds had this spike and they're back down now.
- [38:12]
E
So.
- [38:12]
B
Yeah, and I'm just curious, Ethan, how much do you see that actually playing into terms or you know, credit worthiness for some of these companies or the appetite to finance these builds? Because Matt was, Matt Williams of at Luxor was pointing out something I think really notable on our last segment with him, specifically that he thinks some of these new compute futures are kind of solving for the wrong problem. They're just showing rental hours for the GPUs, but, but they're not showing the actual potential revenue of monetizing that gpu. I mean, you can use that as a proxy, but saying a G, you know, an H100 can be rent out for however many odd dollars an hour is not the same as saying this company brought in X amount of revenue over this time period by running those things. How much do the do the GPU rental hour figures actually matter right now or is it kind of just noise?
- [39:07]
E
It matters a lot what the specific NEO cloud can get for their rental because lenders want to see what their debt service ratio are and how can their cash flows actually cover their liabilities. And so the actual contract itself matters. The indexes don't matter at all really. I think they're good for understanding directionally where the market's heading. But as Matt probably talked through, there's a lot of discrepancy between what these contracts are coming up at. For example, a Top tier Neo cloud like a Nebius or a Core Weave or a Lambda can rent their GPUs per hour at a much higher rate than a new Neo cloud with no proven track record. There's difference geographically, there's difference from a technical perspective. How much storage and memory does it come with? And so those aggregator indexes are useful just directionally, but on a deal by deal basis, the lender is looking at, okay, what does this compute contract look like? Can they get a really high rental rate? Can they get a strong down payment to ensure that this counterparty is locked in? And also look at the counterparty on the extreme end. We hear these deals of fluid stack backed by Google, that's really good paper. People are willing to underwrite that all day at very good terms. Then on the other hand, you have a startup that say raised 5 million and then they've now entered into a $25 million contract over the course of three years. They're not going to be able to fill their payments unless they raise capital during that time. Lenders are going to look at that a lot less favorably and give a big haircut to that. So there's a lot of things lenders look for in these deals and GPU rental rates, the actual one is quite important,
- [40:48]
A
I guess, maybe. Last question, Ethan. I'm just curious about the AI Capex cycle. Put on your Chris, you know, pull out the crystal ball, tell me what you see, read the tea leaves. How are you feeling about where we are in the AI Capex cycle?
- [41:08]
E
So I think I've been maybe in a certain bubble came back from Nvidia GTC Taipei, where the incentive for obviously all of those players, including the Nvidia team, is to not describe it as a bubble. I think the positioning a lot is, hey, there's still a huge surplus of demand and supply is just so far behind. Right. And that's not typically an imbalance you'd see at the top of a bubble. You'd start to see that, you know, tip the other way. And I think we're far from there. So what we're seeing really is like folks coming to our desk and looking for very long dated compute contracts, which I think is a good sign. So, meaning there's compute offtake buyers that want to access computer, that want to partner with our NEO cloud network to start deploying clusters of V3 hundreds like six months and beyond. And so it feels like there's no signs of slowing down. And really in some ways it continues to accelerate.
- [42:05]
B
Is there anything I guess A kind of follow up on that. Ethan, is there anything that you could see that it would slow it down? Because when I look at the lending side of things with some of these A listers like coreweaver, Nevius, their credit spreads have just absolutely shrunk. Shrunk. Recently they've been able to get much more favorable financing in the ballpark of 6%. We've seen that from Iron too. Which being honest if you asked me that two or three years ago that iron would be getting debt at 6% I would have like laughed, you know, thinking about how, just how far we've come with that. Is there anything that you could see that would throw a wrench into it, whether it be depreciation for machines, potential tenant renegotiations, breakthroughs in how we do inference, etc.
- [42:51]
E
Yeah, I mean I think there's a few components to that question. Like first on the lending side it's gone so competitive like from the early days that we were talking about earlier with Macquarie and like Blue Owl and Blackstone, like to, to today where so many credit funds want to get into the space and any credit fund that didn't have any exposure to AI, like you know those analysts are being called by their bosses and saying hey, why don't we have any of like AI in our book? And like how do we get access to some of this? And what that's led to is just a lot of capital chasing the same deals and those, those margins like really compressing for the lenders as well as rates coming down for the borrowers. So that's just kind of been a supply balance from the lending side. But in terms of like what, what could go wrong here? I mean I think largely like it's those same risk which is like how does Nvidia price their equipment? That's going to be a huge force impact and then what does AI spend look like? Are we going to near the top of capital coming in for AI companies to do inferencing, to do training and all the rest of the side of compute. And that could certainly slow down at points and it most certainly will at certain points and that pendulum will swing, it's just a question of when. As of today it doesn't feel like it's swinging but that could be, you know, that could change pretty quickly.
- [44:11]
B
All right, Ethan, third time's a charm. This is actually the last question I want to ask you, but you mentioned you got back recently from or you're going what the conference you went to in Taipei. It was experienced GTC Nvidia. Nvidia, yeah. So you kind of already mentioned one of those takeaways is Nvidia saying, you know, we're still bullish, which of course they're going to say that. But were there any other takeaways that you had from that conference that really kind of struck you, you think worth sharing?
- [44:37]
E
Endless. Endless. I mean, obviously the big push right now is towards Asics, so Nvidia's acquisition of Groq highlights that. But it's clear that the future of AI factories and token factories is going to be a mixture of deployments of GPUs, ASICs, and obviously much more CPU heavy, which kind of highlights like the new Vera Rubin from Nvidia with a big focus on the cpu. But really for Asics, the first use case was for inferencing. For certain inferencing tasks, ASICS can do it much more efficiently than GPUs. But then now we're starting to see that even within certain tasks like training, between regression and diffusion, you're going to start using Asics for certain tasks within training. So AI factories are going to be much more purpose built. And so when you think of a bitcoin miner turned neocloud, they're going to be building, going for a much more unified solution. They're not going to be buying GPUs from one person, networking from another person, ASICS from another person. It's going to be one integrated solution.
- [45:37]
B
Yeah. We covered the jalapeno News from OpenAI last week and it's interesting that that's one of your biggest takeaways from Nvidia's conference, because it seems to me like this is something that they're going to have to respond to in some way, but maybe it's too early. It seems to me to be so incipient this, this change to ASICS for AI that is kind of like anyone's ball game in some ways. And I don't know, maybe Nvidia won't be the kingmaker in five or ten years.
- [46:06]
A
And with that, Ethan, thank you so much for hopping on the podcast. Thank you for bringing your insights. Appreciate the whole Luxor team. We'll have you back on soon.
- [46:18]
E
Awesome. Always fun to chat.
- [46:20]
B
Thanks, Ethan.
- [46:23]
A
All right, we are gonna talk bit deer signing a new lease. We've also got Hive CEO Ivan Kulik Kilik on. But before that, a word from our sponsor, Lygos.
- [46:46]
B
Michael Strategy, Michael Saylor. Michael Saylor's strategy is actually going to sell Bitcoin. He's the reason that you are poor right now. And he's also the reason that hedge funds and lending desks are getting liquidated. Maybe not the actual reason but everyone loves to have a scapegoat and it doesn't matter what that scapegoat is because when Bitcoin crashes and someone is over levered they've got to sell down their position. And if that someone is not you because you're not holding the keys to your Bitcoin through your preferred crypto lender, what are you doing? Don't be the next FTS or Celsius victim. Do yourself a favor and check out Lygos Finance if you are interested in Bitcoin backed lending. They are our preferred Bitcoin backed lender here at Block Space because they use Bitcoin native smart contracts to make sure that you are always in control of of your Bitcoin. With Lygos there's no bridging, there's no rehypothecation and there's no wrapping. There's just pure self custody. Bitcoin backed lending with rates as low as 8.5 to 10% APR. Go to Lygos Finance to learn more.
- [47:56]
A
All right, we have yet another
- [48:01]
C
former
- [48:01]
A
Bitcoin miner, now AI data center company moving to to or announcing a deal in Norway. This is the great European ification of data centers. This one is the biggest deal we've seen, I believe.
- [48:19]
B
Yeah, I mean assuming it goes through obviously. And this is also the biggest move to date for Bit Deer in terms of its AI business line as we. I'll get to that in a second. I'll just read the headline here really quickly but Bit Deer Technology Group signs co location lease agreement for Tidal Norway AI data center site. I'm reading here directly from the press release filed with the sec. There are more questions than answers currently about this lease but the tldr it's for their Tidal Norway facility which they own and operate and they are planning to revamp for up to 225megawatts of gross. It loads. It's important. That's not specifically critical it load but that's the whole enchilada per what Charlie was saying at the beginning of this show that there's a lot that goes into this that's not even just powering the GPUs. There are no details on this right now. Bittir is targeting 180 megawatts for the site by the end of 2026. Harris Bassett, the chief strategy officer of BitDeer said quote signing this agreement marks an exceptional step in bitdear's execution of its global AI infrastructure strategy. We look forward to sharing comprehensive details of this transaction, including terms and strategic context once the lease has become effective. Now we also have some notes here from Matthew Siegel at Vaneck. Before I go to a few more second takes seeing VanEck guesses the first phase will be 50 megawatts and the entire 225 megawatts could be worth $15 a share at $6 million per megawatt CAPEX. That's according to Vaneck Research, or specifically Matt Siegel himself. Like you said, Charlie, we've been covering Europe a lot recently. In fact, our next segment with Aydin Kilik will deal with Hives Build out in Boden, Sweden, a facility that they are in the process of acquiring, and their first AI tenant that they've signed an LOI for have not disclosed yet. All of that subject to closing. But we're starting to see some of these bitcoin miners turn AI companies move towards the European sector or around the world. We might as well as we're rebranding to an AI data center podcast, we might as well rebrand to a Eurozone podcast in some cases as well, because that's really been where the deals have surfaced over the last few weeks. And there are a number of reasons for that I'm sure that Aydin can speak to. We've kind of pointed out that ultimately these companies are going to find power wherever they can get it, so it doesn't really matter where it is. And Norway and Sweden, specifically the northern European countries, because they are flush with hydropower, seem to me to be emerging as the, you know, tier one sites for these data centers in Europe. I assume that doing business there might be a little bit easier than some of the places on the continent. It's also a reason why they were the only places where you can mine bitcoin in Europe until AI came and stole the show because they have cheap enough power and they have a lot of it. So just another recap. This was Bitdeer's biggest move into AI yet. They are trying to get a data center for AI off the ground in Massillon, Ohio. Blockspace listeners and readers will remember that there's currently a lawsuit with a shared tenant on that site against their revamp because they're saying that it is disrupting shared utilities on the site. So this one tenant, it's a steel manufacturing company, is saying that Bit Deer's construction to turn that into an AI data center is actually disrupting Their ability to share utilities on the property. And as I understand right now, that is in the air as to how that will pan out. But if Bit Deer signs this, this will be their biggest AI deal to date. They have, I believe, some small capacity in Southeast Asia running clusters right now, but it's very small. It's just only a few megawatts at this time. So this would be the largest footprint for them yet.
- [52:23]
A
Yeah. Bit Deer, one of the largest miners, has a big power pipeline, if I'm not mistaken. Like 2 to 3 gigawatts.
- [52:32]
B
Yeah, it was like 3 gigawatts. And most of that, or a lot of that is planned capacity, where they don't actually have the PPAs in place yet, but they're strategizing for how to get it. But, yes, they have massive capacity projected into the future, and they're already running quite a lot over a gig of infrastructure.
- [52:48]
A
Yeah. So huge miner, operated at scale, tenured as well. Want to do the AI pivot, but they haven't been able to demonstrate that they are. They've got stuff locked in. So this is pretty big, pretty significant. And I'm looking at the markets right now. Most of your neoclouds and powershells are down, whereas Bitdeer, if I'm not mistaken, I think is up today, actually.
- [53:14]
C
Let me.
- [53:15]
B
Yes, it was the only one in our market update that was actually.
- [53:18]
A
Actually, it's down today. Now it's down, but it was. It was up at some point.
- [53:24]
E
Yeah.
- [53:26]
A
So market seems, you know, may at least briefly, did respond positively. Again, we. As you said, we. This is not the announcement of a finalized deal. It looks like Matthew Siegel of Vaneck says somewhere in a tweet that we could see it. We, you know, within a month, you know, probably expect confirmation, and then we'll see. This is still very early in the pivot,
- [53:59]
B
and I'd be really curious as to why we're seeing specifically in these European zones. There's. They. These companies are not really. The deals are either taking longer or they're just reticent to actually announce some of the specifics. I mean, I assume in both these cases, that's the case because Hive is still working on finalizing the tenant for the Sweden site because it's subject to that site actually closing and them acquiring it. They have approval from the Bowdoin Municipal Council, but same thing here for Bit Deer. There are very scant details, and, you know, obviously you're not going to announce any details before you have them finalized. Sometimes I wonder if there's any benefit to actually announcing the announcement of the deal. You know, it's like we're announcing the details. Sorry, what was that, Charlie?
- [54:50]
A
It's like they don't want people to forget about them. They're like, yeah, we are doing this. We've got so, you know, keep everybody, keep us top of mind.
- [54:58]
B
But we've also noticed that we talked about this with Asher Ganute a few weeks back or a month back now at this point about, you know, they signed a deal recently with a hyperscaler, but they did not announce who the hyperscaler was. And they actually kept the, I believe, most of the specifics for how much the deal was worth under wraps. And he was basically saying that we've gotten to a point in the game where these companies don't want to give any edge for how they are signing deals and the terms they're getting. And you kind of create this. You create this environment to where if you announce all of this publicly, then investors start asking questions. Well, company A got this deal with this, this neocloud or this powered shell. Why can't you do the same? And so they're starting to keep some of those specifics a little closer to their chests. Yeah, the.
- [55:54]
A
It is interesting though, that bit Deer just kind of comes out, describes the deal terms a little more transparently than the other ones. So different ways of doing bit business.
- [56:06]
B
Yeah, it was a pretty scant. It was a pretty scant press release. Kind of like you said, it's like, hey, guys, we're still here. Don't forget about us. And I do wonder about that with some of the larger public miners that have not made as much headway as their peers. The ones that come to mind other than Biti are really Mara and Riot. Riot has their AMD deal, and I believe AMD has had a few extensions on that they've exercised, but so far very little from MERA other than announcing that they're using Starwood for consulting for how to build their data centers, or they're using them as the partner for the actual construction of some of these AI data centers. They have the Xion acquisition as well in France, but TBD about whether or not that will lead to anything meaningful in that region. There's been a lot of chest beating, but there hasn't been a lot of deal signing. And it does make me wonder
- [57:07]
A
which
- [57:08]
B
thesis will actually bear more fruit. The ones, the early movers here that got in on the early part of the AI Capex cycle wave, or the ones who have sat Patiently and are working on signing deals and maybe seeing what has and hasn't worked from some of their peers and they're going to get in on a later round in the whole AI Capex cycle boom. Obviously, I think a lot of that will depend on what, where the timing is, because this whole bubble could pop in the next few months and then we're basically dealing with years of rebuilding to actually get to a place where these companies can either sign deals if they survive, but I don't know, if you listen to the Mike Alfreds of the world, we're going to rip this thing into 2028, or if you go
- [57:52]
A
to an Nvidia conference in Taiwan, then you come back bullish, just like, just like.
- [57:59]
B
I mean, it does sometimes feel just like a John McAfee. This is the new paradigm. You don't understand the mathematics of the blockchain. Right. I mean, that is kind of. That has been in some cases the only defensible argument I can make for some of the valuations in the sense of like, people just think this is going to be so astronomically game changing for efficiencies in the market, specifically with the company's ability to. A company's ability to do more with less. The idea that AI is going to be such a transformational technology that these valuations actually could make sense on a long enough time frame depending on how accretive this technology is. But it seems like we're so top heavy and frothy right now, Charlie and I
- [58:48]
A
says, you will see. I mean, Bitcoin's at 60k at some point. I do, I do. I'm like, okay, the risk assets, which are crypto assets, historically, they're all down, so how frothy can it possibly be? It's just frothy connected to a.
- [59:04]
B
You know, that's the whole market right now, though. That's what's driving all the, like, that's where all the credit's going to. And I think that, you know, really, I do think one of the only obvious things is like, once this current credit cycle stops and that money stops flowing into AI, that's when you could potentially see cracks with regards to what we're seeing with this current Capex boom, because that's obviously what's driving all of it. But I'll shut up because I believe we have Iden Killick waiting in the wings.
- [59:35]
A
Yeah, we've got Hype CEO Iden Killick in the wings and we are going to bring him on up here, here and talk hive. Let's Bring them on up. Hey, guys, welcome. Welcome back to the show. Nice hat, by the way.
- [59:51]
D
Oh, yeah, the. This is a limited edition gray. You've had to be with high for at least five years to earn the. I'm just joking. You know what, it is in a way, because we've been at the conferences all over the world for the. I mean, really since 2017, but, you know, I think in full force in effect since 2021, where, you know, Frank's done a tremendous job branding Hive with sort of the, the, the gold B logo, which actually comes from the Vatican. And so any rate, and Gucci sort of emulated it, and so it's very iconic and we've done different colorways, et cetera. And this is sort of like the, the tech material hat and we have the, the dad hats. We have one in crimson red.
- [60:36]
A
So, like, before I let Colin ask a real question, you got to tell me about the Vatican thing here. I'm. I'm enraptured with where does this tie into the Vatican?
- [60:47]
D
Yeah, yeah, I think so. Like, often people think it's like the Gucci bee, but, you know, Gucci obviously being an Italian brand, and Frank could explain it better, but like in the iconography, somewhere in St. Peter's Basilica, they. They do kind of have a bee like this. I think that's where. That's where it comes from. So that's kind of off the top of my head, the extent of it. But yeah, you know, always if you're ever at a conference and you see any one of our executive team, sometimes you'll see someone wearing these hats and it's just. It's just a fan or a shareholder who got. Got some merch from the booth. But, you know, we always welcome speaking to anybody. So you're. You're always, you know, you know, invited to drop by and say hi.
- [61:33]
B
So Aydin, that B is buzzing all the way to the Great White north in Bowdoin, Sweden. And I. So I really want to touch on the whole enchilada here with the LOI and the acquisition. But so a great place to start. Specifically, why did HIVE decide to make the acquisition of this data center in Bowdoin, Sweden, and kind of take root in the Swedish market more permanently?
- [61:57]
D
Well, the thing is, Sweden has been a stronghold of high for really like coming on seven, eight years, no, eight plus years now, because Bowdoin was running since 2018. That was the original super site. We had 130, 000 GPUs and it was doing over a million dollars a day. That one site during the Ethereum mining era and it was a long term lease with the municipality of Bowdoin. And so it's hard to buy property from the Swedish, from any Swedish government owned entity. But we've invested deeply in the community, we've talked a lot about it, we've talked about investing and sponsoring the Bowdoin hockey team and it's called the Hive Arena. But the other thing too, we've paid a lot of tax to, to the country of Sweden for, for operation. So I think, I think that that's recognized. I think they realize we're good stewards of assets. We're an excellent data center operator. You know, when I went down for an inaugural visit in 2021, when I joined Hive, four levels of executive management, our account rep, his boss, his boss and his boss came to meet me at Vattenfall. Vattenfall means waterfall in Swedish. So it's the national utility company, obviously hydroelectric power. And they told me, at least at that time and I'm pretty sure we're still up there. But at that time we were the largest demand response participant in the country of Sweden. So I think we've done a lot and really if you're going to start investing, you know, nine figure sums, I think our early estimate for construction at the site, something or other to the tune of 175 million USD to convert the current Bowdoin facility of 32 megawatts to a Tier 3 liquid cooled power shell.
- [63:50]
C
You're going to want to own that
- [63:51]
D
land before you go making, you know, upgrades and improvements that are pretty deeply entrenched in the infrastructure and you can't just pick it up and move on. So it was important to us to own that real estate and so we got approval to buy it and we press release that.
- [64:07]
B
And, and why specifically Sweden? Like what, what makes you want to take root there? Like why is this a market where you're, where you want to build?
- [64:14]
D
We already, we already have taken root there. I mean this, this property's been a stronghold since 2018. So we've been operating here for eight, going on nine years now. So, you know, it's. People can build bitcoin mining infra for half a million megawatt. Rough numbers, right? But HPC tier 3 can be 8 to 10 million a megawatt per critical IT load. And then on top of that your GPUs are 30 million a megawatt, where ASICS roughly could be a million a megawatt. So we're talking orders of magnitude more investment. So we felt comfortable operating in this jurisdiction originally because it had phenomenal power pricing. We were, we still do. We Ford hedge typically we target 2 and a half cent USD power 2 and a quarter sometimes, which is phenomenally competitive. And we still have the ability to sell back to the greatest spot pricing like in bare market economics. Just, you know, we've talked about this in the past, but I think it's always good to remind people why I truly think Hive is pound for pound the best operator in the game. So we'll hedge power at 2 and a half cents USD if spot rate goes up and if hash price drops and we're like hold on, we can actually sell back to the grid and make more money selling back to the grid than mining. Right now we will do that in the crypto winter of 2022 after the three arrows calamity and all that stuff, hash price dropped at the time, at the time, which this sounds like, obviously not such a bad hash price. But again ASICS were far less efficient back then. Hash price dropped to $50 a pet ash a day and a lot of our peers were bleeding. We mined profitably during that quarter. It was calendar Q4 of 2022 and we made 3.2 million in demand response selling back to the grid, which again, you know, this is, this is serving the market and, and selling power at the grid, not being opportunistic but really working within the guidelines and parameters that the utility company has and says hey, this is what we would want to sell power back to the grid for. In fact we prefer if you sell it to us, we could service the grid. We're like brilliant, we'll tune down, sell power to the grid was more profitable, do that than actually mine and, and it, and it served our shareholders well. So it's, it's, I think earlier I said the, I would like to think they view us as good stewards of, you know, the community of, of you know, the asset base, which is the data center in this case. And of course for our shareholders, I think that we're good stewards of capital. So it's being very thoughtful and intentional. Even during the 2022, 2023 era when we didn't see a good ROI on ASICS, a lot of our peers were just scaling at any cost to hit multi billion dollar market caps to really get on that capital markets flywheel. But when it looked, when you looked at, we had two different banks, it was Cantor Northman, they did an analysis of a dollar per x/ hash, how much money did we raise, how much money did our peers raise and how many exahash did they actually bring online? And that includes infra and paying for Asics, which is a really good report card because it's otherwise hard to track. How are people roiing, people running their ATMs etc? We were something on the order of 20 million per exit hash, which our peers were anywhere from 40 and a lot of them were 60 to 80 million per ex hash. Why? Because you overpay for ASICS, right? And nobody's checking and it's dilutive. You're never going to ROI if you're overpaying for asics, but when you just want to scale at all costs mentality, and we've never done that. So I think that now we've navigated, we've mined profitably through every single quarter over the last six years and now we've been building this GPU cloud business. We have 5,500 GPUs running, we have 35 million ARR today. We're now contracted with over 100 million of ARR in our HPC business unit. With the announcement of the deal with Cohere, which is super exciting that will come online, be installed and deployed Q4 calendar Q4 this year.
- [68:25]
B
Speaking of timelines, Aydin going back to Bowdoin, specifically a few milestones to clear closing of the acquisition of the land and closing the deal for the first tenant. Can you give us a rough idea of how long this revamp will take to start servicing AI loads? Is there a target date for first integerization that you are looking at now? Roughly what are your thoughts there for the timeline on this Bowdoin site?
- [68:50]
D
Well, the other announcement is that we also put out last week on Thursday that we signed an LOI with an investment grade sovereign Swedish technology company. And based on indicative market rates for bowdoin, it's about 150 bucks a kilowatt a month, which is actually higher than you're seeing a lot of our peers stateside that are in secondary markets throughout the US doing long term deals at $100 a kilowatt hour, $130 a kilowatt hour. So 150 bucks a kilowatt hour I think is a phenomenal unit economics. And if you look at, if you do the math, it's like a four year ROI. If like our initial CAPEX estimates are 175 mil and if you do the math, 32 megawatts of utility load on a nominal 1.3 PUE gives you 25 megawatts of critical IT load on at 150 bucks a kilowatt a month. It's easy to go one hundred and fifty bucks a kilowatt times 25,000 kilowatts. I. E 25 megawatts times 12 months is 45 million a year cash flow that a long term contract could potentially and indicatively throw off based on demand for that region. So that's like a four year ROI on the retrofit. I mean typically data center HPC retrofits or new bills are like 78 year ROI. So the economics are phenomenal and in terms of timeline etc, so we'll update the market accordingly and we'll put out a press release and happy to hop on with you guys but you know, we really wanted to let the street know, hey guess what? We own the site. It's going to be an HPC site and then boom, guess what? We've already got a investment grade top quality client that's interested and in fact it was quite prescriptive. They want to do approximately 10,000 Grace Blackwell three hundreds. So be hybrid, liquid cooled, air cooled. So we're very excited about that and I think that the street, you know it, it was. We had also done a convertible financing at a premium. We did a cap call.
- [70:55]
B
Yes. And y' all upsized that I believe as well. And a quick point of clarification, I knew about the LOI with the milestone I was talking about specific is getting the details out for what exactly the terms will be etc. After the site closes. So I was just more curious about when we might expect to see or a timeline for revamping the site. But the convertible note news also really big in terms of $100 million convertible note, 15 million upsizing after y' all announced it. So clearly a lot of demand.
- [71:28]
A
Yeah, I am interested in time like as you guys have been. You say you've been operating out of this site for a long time, you've been operating in Sweden for a long time. I feel like the actual tactical way of retrofitting a site which used to be mining cryptocurrency into now the new age GB300 racks. I think a lot of folks, they get mystified by the numbers and the size of the power deals and the rates and financing. What does it actually look like to retrofit these racks tactically? I mean what are we doing? Do we have is it conventional server architecture or is it just a complete overhaul of the site? Give me a little window into this process.
- [72:14]
D
That's an excellent question. And by the way, earlier when I was trying to explain the economics, I was trying to demystify that. I was trying to say how do you get from an indicative rate of 150 bucks a kilowatt a month to an annual figure? Because you, you've got people coming out saying, hey, you know, we just signed a, you know, a five billion dollar contract for some site that we're going to do colo. So like, how do you get there? So let me first again, I'll recap those economics. You always take the dollar per kilowatt, right? Which is like the industry. So if you're going to go sign colo, it really is it like in bitcoin mining, we talk about dollar per kilowatt hour. This is actually dollar per kilowatt per month. This is very similar. Tomato, tomato. And so that's why it's good if you're sort of an avid mining enthusiast. And again, really for hive, this has always been our DNA because when we were going from bitcoin mining to ethereum and then even kicking off GPU cloud, how do you compare the value proposition of each business unit? Well, the common denominator is $per kilowatt hour, right? And then that's how you really can evaluate what's the profitability of my GPUs versus my ASICs, etc. So it's very similar here. You look at the dollar per kilowatt per month. So you take that what is your critical IT load and you multiply that. So 150 bucks a kilowatt times say 25,000. I'm giving you indicative, right? And we'll update the street, of course, with you know, leases when every, you know, when definitive contracts are announced, et cetera and so forth. By the way, then I'll get into the capex and scheduling question. But for everybody listening, you go dollar per kilowatt per month times your critical IT load. So 150 times 25,000 and you times that times 12 to get to your annual, that'd be in this case, using those numbers, about 45 million a year. And if you were to sign, oh, I don't know, a 10 year contract, just for example, it could be longer, could be shorter, It'd be a $450 million total contract value. So now by comparison, if you had a site that would say 300 megawatts, say some of our peers in the US have like a 300 megawatt site and the utility load is about is 32 megawatts of Bowden. These larger sites we're seeing, you know, the ciphers and hot do are about 10x that size. So that's how you get to like a four and a half billion dollar contract. You're, you're taking your ARR per year and multiplying it by the term of the contract, be it 10, 12 or 15 years. So that's, that's on the revenue side, on the cost side and on the scheduling side really what you're doing is you have to order long lead items which are habitually the chillers, the UPSs, which are basically fancy batteries that make sure you've got steady power if anything happens until your generators flash up. So diesel generators. So these are big industrial grade diesel generators and you know they'll take a few moments to flash up to provide that secondary power. And until they do and that secondary power stabilizes your UPS's. Again, fancy batteries are your, they're your bridge. So those are habitually the three things that you need. And we could be talking lead times upwards of a year. Depends. You know, a few months ago chillers were actually having the longest lead times. What you want to do is order those, put a deposit down, then you have what's a purchase order. And you know, crypto mining world, it was very different. You often in bull markets you had to prepay big deposits up front and you know, things wouldn't order for, for a long time. But it's a bit better because you know, for GPU, I mean GPUs are heating up now you've got to put deposits and, and you've got to pay everything, you know, before the ahead of the GPU shipping. But you know, once upon a time with GPUs we were able to get net 30 payment terms. They would, you know, super micro would ship GPUs, we pay them three days later, which is fantastic. But it's very similar demand economics. But these long lead items, you got to put a deposit and you got a PO that gets accepted by the vendor. Now you've got an estimated delivery date. And so that's really what any credible counterparty in this industry is going to want to see because that will dictate your ready for service date for the data center. So until you have those long lead items you're really your dating item. It's very hard to fundamentally I'll use the word convinced or broadcast a go live date or completion date, AKA the ready for service date for your data center. So once you got the long lead. You got the ready for service date.
- [76:35]
A
You go.
- [76:35]
D
Okay guys, this is going to come online mid Q4 this year, whatever it may be. So I would say stay tuned for as we provide those subsequent updates.
- [76:44]
C
Really.
- [76:45]
D
In the last quarter, we've ramped up our global between Canada and Sweden, 440 megawatts of HPC Tier 3 conversion capacity. And that includes the Gigafactory in the Greater Toronto area, 320 megawatts, 70 megawatts at New Brunswick, which are a flagship bitcoin mine in Canada. And of course now we've got Big Bowden, 32 megawatts, Little Bowden, 7 megawatts, collectively about 40 megawatts. And so it's really exciting. So we've, we've really, you know, the narrative, if you watch my interviews, has been we've been land banking by substations. That was how I was alluding to the fact that, you know, we're focusing on building that land and power pipeline. It's not just land anywhere in the middle of nowhere. You, you've got to buy land where there's power close to it. And so either you're building your own substation like in Paraguay. We are, we're, we're land banking. Our Iwazu site, where we've got 200 megawatts running is in the backyard of the regional 1.2 gigawatt Itai Poo substation. And, and that pulls right off the hydro dam at 500 kilovolts. On our site we have a 200 megawatt substation which we now expand into 300 megawatts. By the way, not to be confused, in all of Paraguay we currently have 300 megawatts running. The other 100 megawatts is in a region called Valenzuela. But in Iwazu, we're expanding our footprint. So that was this quarter. Hey, land banking by substations. Boom. All the news comes out now we've got a footprint. And by the way, if you include that extra 100 megawatts that we have coming online in Paraguay, substation is landing next week. All the civil work has been done Again. We build heavy infra. It puts our global pipeline of HPC for over 500 megawatts. And I will note that we also put out a press release last week that we had a proof of concept with a brilliant research team out of Columbia University in New York where for the first time they were able to do a pre training run from New York to Asuncion. Over 5,500 miles away. And the results were remarkable. Number one, it validated really our proof concept in Paraguay. Number two, the results were so novel that they actually submitted the findings to the largest neural network conference in the world. And what they were able to do was using an optimization of a program pre training algorithm called Muon Realized performance from our eight hundred and forties which are five years old by the way we used to mine Ethereum with them. They're still going.
- [79:13]
A
Yeah, I remember those.
- [79:14]
D
Oh yeah, they're still there baby. We get every ounce we go, we go the distance, we go the full 12 rounds like, like rocky when it comes to our compute and, and, and if they're end of life we'll see is there is there some novel way to optimize. And they were as performant as H1 hundreds in this use case which was. Was tremendous. So huge, hugely impactful. I think it's kind of like you know when we built our own Buzz miner. We're the only company to successfully build a working ASIC with Intel. A few others had you know, tried but you know it's really hard to get it working. We built over 10,000 of them. There's a lot of stuff you learn and there's like at the firmware level, at the integration level, you know, thermal management, et cetera. So we've got a lot of deep tech under the hood at Hive and I think that it's good to talk about that. And now you know with Craig Tavares joining our team and he's got 20 years of experience building data centers at Cogeco and Apple. You know we, we've just as Jensen like to say scale up and scale out and so now you know we've really got a global footprint of pipeline for the next year. I think our white and I think our 2 year forward target is 700mil ARR for HPC AI business unit 200 of that we hope to realize what we will realize end of this year as we ramp that GPU cloud again. We're 105 million contracted right now. We've got a couple big another big deal in the so cohere was the one big deal we announced the other big deals in the wings hope to announce that over the next month and then with that deal will be this is all our investor presentation just kind of paraphrase it save people some, some some time. It's that second deal would get us to about 170mil ARR. And again the data center capacity for this growth near term is all with Bell AI fabric. So you have an opportunity where we could ramp our GPU footprint really quickly. Get 11,000 GPU use up from 5,500 today, 200mil ARR. Spun up end of this year. And then next year in early 28, we're just retrofitting our own sites to tier three, and that'll add another 500 mil ARR.
- [81:27]
B
So, yeah, a lot of opportunities for expansion. We'll be keeping a keen eye on it, especially on the announcement that should be forthcoming. Aydin, thank you so much for coming on, breaking this down. Best of luck with all the revamps, and best of luck to you all for the rest of 2026.
- [81:44]
D
Thank you. And by the way, I looked it up. It's called the Barbara St. Peter's baldacheen, the Barberini Bees Herald.
- [81:53]
A
Okay, there we go. Yeah, that was. Look, you can tell me also about your data centers. I want to know about the bee.
- [81:59]
E
Just.
- [82:00]
A
Yeah. So thank you so much, Aydin, for your time. Thanks for coming on.
- [82:04]
B
Thank you, Aydin.
- [82:04]
D
Thanks, boys. Thanks for having me. The pleasure.
- [82:10]
B
All right, I think that's a wrap, Charlie.
- [82:13]
A
That's a wrap. If you are sticking around, make sure to tune in Every weekday Eastern, 1pm, 1pm Eastern, every single day. Featuring quick hits on AI data centers. Bitcoin, Bitcoin markets. Bitcoin tech markets, emerging tech. This stro is brought to you by CleanSpark. Nasdaq listed ticker CLSK. Thank you, CleanSpark. We'll see you all tomorrow.