How Low Can We Go? — Data Center Monthly

[00:00:00] Speaker 1: I'm ready. I'm good. We are live for Data Center Monthly. It's been a little bit of more than a month than we've done this. And of course, Nate and I are joined with a special guest, Tanner. Tanner, how are you? Doing awesome, man. Thanks for having me. Yeah, thanks for popping up. [00:00:15] Speaker 2: Thanks for coming to our therapy session. [00:00:19] Speaker 1: Yeah, we're all going to zero together, I guess. So yeah, tonight we're going to talk about the broader macro, why we would think what's happening, how long it's going to last, what we expect, near term, far term. Then we'll talk about some individual news specific to Nebius, CoreWeave, and Iron, as well as maybe some other names too. Tanner can't be here the whole time, so we're thankful for the time that we have with him. So we're going to start off with, of course, the price action real quick and then the broader macro as well. Let me show this right quick. We always look at this because we're here to make money and boy, there's not been a lot of money to be had. And so we can see it's not just CoreWeave that's fallen. CoreWeave is down 36.5%. Iron's down 47% over the last month, 33% and 32% down for APLD, which had really good earnings last week and also Nebius. So a lot of pain. You know, we'd have to go back further than three months ago, six months ago, to see one that's not actually bleeding terribly, which is Nebius, and it's basically flat. [00:01:19] Speaker 2: Just a cash position. [00:01:21] Speaker 1: Boy, yeah, like pretty much. Year to date, you know, Nebius is still up and the others are getting absolutely crushed. But, all right. So let's start with the broader macro stuff. We can share specific news. I can bring that up as well if we want. But what do we think is going on? Why is, obviously, it's not just CoreWeave. If CoreWeave was down 40%, everything else wasn't, we'd be like, yeah, CoreWeave bonds. That's it. So, I guess, Tanner, since you're our guest, why don't you start? What do you think is really the culprit and where do you think we go from here? Like, when does it rebound or does it just go straight to zero with everything? [00:01:58] Tanner: Yeah. Well, I mean, it's such a broad question. So where should I start? I think that with the concerns around CoreWeave, they are very true in the way that their success is not guaranteed. And, in fact, if anyone's been following any of my commentary, I've actually been buying a lot of Nebius and not adding to CoreWeave, even though we're down like, you know, half the price of where I've boughten before. But the concern is that not now, but at the time of refinance, that they're going to have to pay way more for their debts and that they're locked into contract prices that are not nearly where the spot rates are today. So I think that the companies that have a lot of access to capital or GPUs that are online and not sold out for five years are going to have amazing pricing flexibility. And I don't think that CoreWeave lands at the top of that list. You know, you could put SpaceX on that list. You know, maybe Meta might step into the ring one day and they'd be pretty high up on that list as well. But Nebius obviously up there, these sort of names. So I'm just a little bit worried about the names that are heavy into the debt CoreWeave. Oracle is another one. However, you know, it's now under $400 billion of market cap and they've got huge business lines that are outside of the cloud space. But Nebius is the one that I'm adding to right now. I don't even remember your question, to be honest. [00:03:43] Speaker 1: So what's the cause of this pullback? And how far, how much more do you think that we have? Is it, you know, are we looking at down another 50% or? [00:03:54] Tanner: Well, I think 10-year yields going up, like those, the investors are going to want a healthy spread. And I think that spread widens if you believe that you're not going to get paid back. So while demand is raising right now, if the 10-year continues to climb, it's not even going to matter. Because, like I said, they're locked into long-term contracts and you're not going to be able to adjust your price fast enough on an old three-year-old Microsoft deal that still has two more years to go if their cost of capital continues to climb. So, you know, every single company that has, that's making money in a form of a spread on interest rates is going to get killed if, you know, risk-free rates continue to climb. [00:04:44] Speaker 1: Yeah. No, that's fair. Nate, what about you? Is there anything that you'd push back on or anything else you see like, hey, this is really a contributing factor? [00:04:53] Speaker 2: No, I think he's entirely right. I think what we saw a year ago was a deal meant the stock pumped. Everyone cared a year ago in, you know, October, September, November of last year, what mattered was the validity of the business. The business model making sense, actually getting backlogged. And then in March, we saw, or I guess in May, we kind of saw a pivot of, okay, now you have the deals, but is this going to make you money? Are you going to be able to pay back the debt you're going to capture from these deals? And so Roy, I have, I shared my screen real quick. And I think this, this image here is just literally a direct correlation between why we're seeing this all. It's just risk taken off the table. It's, it's these companies being re-rated because of the amount of debt. Now, obviously I think something like a Nebius is being the baby thrown out with the bathwater, but at the same time, they're going to need to reset just like they did today. And we'll talk about that. But Nebius is down the least. If you look at all of the Neoclots, since they're all time high, Nebius is down, I mean, it's down 45%. We're still hurting. But compared to all the other Neoclots that are down 60, 70, 75%, it's still been re-rated the least in a sense. And I think that's because of the debt. It's because of how they financed the growth and because of how they've attacked the financing aspect of the business. And so I think it's just quite literally just the market taking risk off the table or what they both perceive to be risk from all these backlogs and from all these quote-unquote prepayments or not prepayments but contracts signed by the Anthropics and the Open AI's of the world that is hundreds of billions of dollars at this point. And so I think the market is, I honestly just think we're at a point where the market is just going to wait and see, like, hey, you say you have this contract. Let's see it actually be actualized into revenue. Let's see it slowly become an operational profit over time. And that's again why I kind of picked Nebby's here because I think they'll be there first. [00:06:45] Speaker 1: So do you think that that means that we're going to have this price overhang for like multiple quarters or? [00:06:52] Speaker 2: Yeah, it's tough. This is definitely to me closer to, I mean, honestly, very similar to October of last year where that lasted six months. Now, then we hit in October of 31st, I think literally was the all-time high last year. I think it was, wasn't it Micron earnings or someone, some company's earnings was the all-time high. Maybe it was NVIDIA, I forget. October 31st last year I think it was the all-time high. The difference this time is the all-time high was June. So I do think we're getting closer to the bottom. We have so many macro things like oil, Iran, midterms that are ending soon. I do think that's a big part of it. But it's hard to tell, man. I don't know what, you know, I guess my question for you guys is what changes the sentiment? What comes out where like, okay, now they can pay back the way we think. Now they're, you know, I don't know what we've seen over the past three months has been up and to the right positively. And so like, I just don't know what changes to make the market go, oh, wait a second, we're wrong. [00:07:51] Speaker 1: You know? So I think a large part of it is related to politics and midterms. So like it or not, Donald Trump has very clearly given the stock market performance as the litmus test of his administration. And so I think that there is definitely some effort to try to undermine that as we head into elections and even more so specifically for data centers. Like he's come out and basically said if somebody's against data centers, they're basically a communist. So not in those words, but I mean like Joe McCarthy from the 1950s would be proud. But no, like I, so I think some of that is going to go away. not all of it obviously after midterms. Now on what you said, Tanner, I agree with you. Like the challenge is, CoreWeave is not insolvent. They're not going to be unable to pay these back. Most, a lot of the loans are not interest only. Like they're actually paying down the principal so they don't necessarily need to finance. Now with the senior convertible notes, if those don't convert, they are looking to need to refinance there. And so there's still financing options. But the problem is the most clear option. So they have a bunch of money they could tap into without raising for their debt. They're fine for at least a year with CapEx and everything without having to raise again if they don't need to. But let's say if prices for all these guys stay down for another six to nine months, they're going to want to build out capacity. But if you raise through senior convertible notes, where your share price is is meaningful because that's where you actually have the cap calls. If you're going to sell that, that's where they actually convert. And so if your stock is wildly overvalued, like honestly, Nebius got ahead of itself. It did. Like it's a great company. It's amazing. It deserves a premium. Absolutely. But that's a great thing for Nebius because they're able to then use that for those senior convertible notes. CoreWeave does not have that option. Like they'll still, they still have things that they can raise. But like if they raised senior convertible notes right now, I mean, they might be like 4% interest rate and convert it like 75. That's terrible. But like SpaceX is one of those we'll talk about here tonight too. Like that's one of the benefits of that opportunity in play is it trades at a pretty lofty valuation. And so, you know, Elon's talking about building out almost 20 gigawatts of capacity like it's not going to happen like over the next Elon numbers. But you need to raise a ridiculous amount. Where's that going to come from? Well, the good thing is SpaceX is pretty expensive. Maybe not overvalued. I don't know. But like looking at where they are like right now, that's a great thing for them. And so, so I agree with you, like Tanner, like there's, that's really the risk in my mind with CoreWeave like right here is if things stay depressed, then they're going to have to slow down their build out which isn't great. Now, as far as like the unsold capacity and everything, they tend to sell out six to 12 months in advance. So they do have most of 2027 sold out because they're near the end of 2026. So there's more coming online that they'll sell out. But yeah, they don't benefit like Nebius does or SpaceX does as well. [00:10:57] Speaker ?: So, [00:10:57] Speaker 1: but I really think that for this high beta stuff, CoreWeave, IREN, Nebius, APLD, all these guys are caught up in the same thing. A lot of it's just macro and I think some of that fades. I wouldn't, like historically we've seen, usually it begins a little bit before midterms when the election is clear as far as the outcome. I don't think that the House or the Senate is as clear enough as it should be for that certainty. But we see historically a 14.1% gain in the S&P 500 in the six months following the elections on midterm. Now that's a small sample size of just since 1970, but there is a strong upward bias. And I think that we'll probably see that still, but I'm not sure. What do you think, Tanner? What is it going to take to see this actually turn around and instead of every single time we do something on data centers, it's like, oh, it's down more. [00:11:50] Speaker ?: Well, [00:11:51] Tanner: I think that the choppiness is going to constantly happen. I think that we're going to take this sort of, you know, markets are going to 2x and then drop down by 50% and then 2x and then drop down. Or maybe not that math. I think that would equal out. But you get what I'm saying, you know, two steps forward, one step back. But essentially, I think where the market's getting this more wrong than not is the flexibility that these companies can have due to the insatiable demand. And everyone's saying, okay, well, if CoreWeave stays on this route, this is going to be a problem and they're going to have very, very thin margins. But who's to say that the future deals that they end up signing, like Nebius, for example, are not prepaid by 50%, 60%. And then even Nebius goes, okay, well, if you want this data center before them, we need 70% prepayment, we need 80% prepayment. And will every company be able to do that? No. But in what they consider an infinite demand environment, there are customers who can prepay and the flexibility of this industry will essentially lead to, who knows, NVIDIA, maybe, you know, maybe they're not backstopping, but they're investing. Micron's got a lot of capital. There's a lot of companies with a lot of capital that potentially might want to make sure that this ecosystem is staying very strong. So I would just say that these are not static markets and then they're very flexible. [00:13:22] Speaker 1: Yeah. Yeah, it's, we usually don't see it go up quite as fast as it came down, but like usually when we see whatever that is, and it could just be like no news, nothing has changed, and the market makers are like, hey, we're pouring in heavy for whatever reason, it is violent to the upside as well. So do you, any more on that, Nate, or do you want to, should we talk some macro specific stuff? [00:13:48] Speaker 2: No, I think Tanner's right. Just, I think real quick that it changes quickly. I mean, we saw in July that was, I mean, that was probably a one-off event, the July event with the liquidation of Leopold. That's a little bit different, but even in March, I mean, it moved quick from March to June. That was three months. Most, a lot of these stocks doubled, tripled, quadrupled in three months and now they're down 50%. So kind of like Tanner said, we could see a reversal after midterms and a massive boost up and then we're going to see the same, you know, 50, 60, 70% haircut and sentiment go back down the drain. So it's just, honestly, it just become, it's just a market, right? Like it's just a rollercoaster of the markets. [00:14:27] Speaker 1: Yeah, it's a frustrating rollercoaster of a market for sure. And, but like the one thing I think that is nice is we're far enough along in the eye trade outside of a few people that just are sticking their head in the sand and ignoring reality. I think most people don't doubt the demand and I don't think that they are doubting that it's fairly durable. Now, there are some that are like, yeah, two or three years out maybe and fair enough. Like we can't really predict two or three years out as fast as things evolve in the eye place. But at least for the next couple of years, like demand is just, like everybody is saying it, that's in this space. Every single person, every single player, there's not one, even like the weakest links that are like, oh yeah, demand is starting to soften. It's ridiculous right now. And then, [00:15:09] Speaker 2: can you share my screen one more time? I think the other thing too that I'm looking at for just overall, you know, markets is also what heads funds are doing. So we can see here that, you know, a lot of these things are looking like previous periods. And so on the left is the percentage that heads funds are selling versus buying. And so we can see it's at the low that we saw in March and actually approaching even lower. Same with when it comes to the gross net leverage from hedge funds where see it's, you know, at a pretty low point. And so I don't think this goes on for much longer, but at the same time, I also don't know shit. So like this could go on for another three to six months, but I just, the earnings are too strong, man. And I think that's what helped the market a lot even in the face of the Iran war. I don't know what this market would have been like without good earnings. I mean, it could have been buggly, but it has helped the market a lot in the face of the Iran war, oil going up to what, 110 a barrel at peak or something like that, which is just absurd. So yeah, I think we really need that on Iran war to end and midterms to be over and we could be back off into the races. [00:16:21] Tanner: Reminds me of like the Kanye speech whenever they're like, oh, what would this look like if earnings weren't good? I guess we'll never know. That is great. [00:16:34] Speaker 1: So this is, speaking of macro, like one other thing, this is a not, the SOX, the semis index and everything. This is, Matthew Siegel from VanEck noted this, semis are now cheaper than the market unless this time is different. So if there's no earnings cuts, I mean like we're looking at 17 times in a space that's growing because I think it's expected to grow 40% next year in terms of earnings. That's a lot better than the broader market. So like we can look historically at how this tends to rebound when there's that much of a disconnect. So it doesn't mean it's going to happen tomorrow. I think it's overdone. Pain could continue for a little while as well and some pain before the midterms is not too unexpected. One of those things that seemed to be kind of that catalyst was this. This seems like ancient history. October 8th, there was opening that came out and their annualized revenue neared 50 billion, below earlier 70 billion. And there was a lot of problems with that statement because it wasn't like internal figures or anything like that. It was just, I think it was Axios that was estimating. And it seemed like here, October 8th, like if you look at the pricing on a lot of these data centers, that's when we started seeing a whole lot more weakness as well. So I know I said like earlier that the demand is pretty much solved, but there are still some people that say otherwise. And one more piece of news, and I want to get you guys' take on this. We also saw this last weekend, Microsoft pulling out of the UK N-Scale data center. And the big part of that was this is located around the London area and they were told that they would not be able to connect to their power grid until the 2030s. And so people are like, aha, Microsoft canceled a contract. Therefore, between that and of course the Anthropic or rather OpenAI stuff, the demand is really not that durable. What's wrong with that picture? What's wrong with that argument? [00:18:28] Speaker 2: Yeah, I think, and I'll answer first. Funny enough, when you actually say October 8th, I'm pretty sure that was the near-term top, wasn't it? I think that's the exact date of when all these, everything has been sold off. And again, I think people confuse that it's just been AI. It's been Rock Lab. It's been ASTS. It's been Robinhood. It's just high beta. So for people that think it's just AI names, no, it's just high beta names. It's just the entire market taking risk off the table. Appreciate the five bucks, by the way. But yeah, I think, what was I saying? I forgot. Okay, yeah, October 8th, the opening high revenue. Yeah, so the problem with that and what frustrates me is opening on Anthropik need to freaking go public. I am sick of seeing these, he said, she said, rumor of what we think is going to happen, what Bloomberg reports versus what opening I says, no, that's not true versus wait, wait a second, we're going to kill the world. Just please go public so we can see numbers, so we can track you against other companies in the sector or in the market. It is just so frustrating like you said, when that estimate was not even for OpenAI themselves, it was just a massive, oh, we thought it was 70 billion, but actually it was 50, uh-oh, that's not what OpenAI guided for because they have, there is no official guide because they're not public. And so I think that's the first thing that just annoys the heck out of me is they can get the constant delay of IPO and these VCs just wanting a higher, higher valuation. The second thing on the N-Scale deal is like genuinely just made me more bullish the proper NeoClouds like Nebius and Corby. Like when we hear people saying, you know, I think Nebius said, you know, five or six people fighting for GPUs that just now has one more customer that needs GPUs and guess who that goes to? It goes to the companies that actually have time to power like a Nebius, Corweave, iREN even. And so to me, it's just more bullish for my company or the companies that the NeoCloud companies that are actually doing it properly. But I can understand why sentiment-wise it's like, oh no, demand's falling off a cliff. These companies can't build data centers because they just bucket them together. So that's why we, that's why we saw that sell-off as well. [00:20:34] Tanner: Yeah. Tanner? Yeah. I mean, saying that the demand is falling off a cliff, these are just like almost factually wrong. But I'm surprised that there's not more commentary on the data center monthly around what might be the biggest NeoCloud. SpaceX. [00:20:54] Speaker 1: Oh yeah, of course. We're going to talk about SpaceX. [00:20:56] Tanner: We have to talk about SpaceX. That is such a crazy freaking report. I don't know if you guys have been focusing on that, but what was the question? Sorry. Is the demand [00:21:10] Speaker 1: durable? [00:21:11] Tanner: Yes. [00:21:12] Speaker 1: Yeah. Next question. Let me, let me play. I think this is too good of a clip to not play. This is, you know, GTC Berlin. There was a ton of great stuff that came out of this. Nebius especially I thought was just absolutely shown during this, but let me actually remove it and make sure that I've added the sound. Apologize. Yes. Tanner's acting like Zuck. All right. Let me know if you can't hear this. [00:21:42] Speaker 4: Jensen, we're obviously in this environment where the cost of capital has increased and as a result investors are increasingly scrutinizing the business models, the balance sheets of the Neo Clouds and clearly that is creating some headwinds whether it's Firmus or CoreWeave or NextDC. You've backed a lot of these companies financially, rhetorically, financially. So my question is how far is NVIDIA prepared to go to back Neo Clouds? How much further could you go in extending your balance sheet to support those companies? Would you ever, if it came to it, bail out one of these companies or is it time for Neo Clouds to stand on their own two feet? One, [00:22:20] Speaker 5: we invest in them. That's different than backing them. We invest in them because it's a good investment. I can prove it. The reason for that is because all of the Neo Clouds, their demands are incredible. Their demand of their compute is incredible. And the contracts, they have contracts from large companies, small companies. Their demand, they're completely sold out. Now, that translates, however, to needing to secure more land power shell, needing to secure more compute, which is, therefore, needing more capital, more working capital. their company's total contracted value is incredible. All of theirs because the demand is great. And so they need working capital. Now, you just mentioned that because the demand for capital is so high that interest rates are starting to go up. But remember what you just said. It's because demand for capital is so high. Now, why is the demand for capital so high? Because the demand for compute is so high because demand for AI is so high. That's what's driving it. Isn't that right? Okay. But remember, at the core of this, the core of this is demand. This is a different, it's not a, it's not a structural issue aside from the demand is incredibly high. We always love when demand's high. everything else will get solved when demand is high. You can't solve the absence of demand. But when demand is high, everything else will get taken care of. And so I'm quite confident that they will find solutions. And their solution starts with the fact that they have excellent customers, demand is really high, and they just need working capital. And working capital will come. It gets solved. [00:24:20] Speaker 4: Jensen, we're obviously in this environment where the cost of capital has increased and as a result investors... [00:24:25] Tanner: Pause it. [00:24:29] Speaker 1: You're muted, right? You're muted. It's way over there on the other screen. It's hard. It's like, where is it? And it's behind something. Yeah. Thoughts, Tanner? [00:24:39] Tanner: He's right. I mean, well, he's right and he's also being generous about one comment saying he's not backing them. We all know he's backing one of them. He's invested in the rest. But I think that he wants people like saying backing and you're supporting and they won't live without you. These are the concerns that investors don't want to have to deal with. Saying that these are strong businesses. If we weren't here, they'd still be standing. Are companies that you can get behind and feel comfortable backing those bonds, backing share prices, backing secondary rounds. that's what you want to talk about. So as a partner, he's being very supportive there. In terms of, like, I truly do think that what he said is true and it seems like the market is discounting it so much. Insatiable demand will be met with creative solutions to problems. Micron coming in. Investors bringing in more prep. I talked about this already, but it's so true. If you're saying, essentially, we're charging $50 million per megawatt because Anthropic can make $100 million per megawatt because the legal company that they're selling their tokens to is making $500 per megawatt, there's huge margins for everyone here. And whenever there's huge margins, Anthropic can charge more and then they can, you know, potentially support the companies that are building those neoclouds closest to their customers the most. Whether that is in Australia, whether that's in Europe, UAE, wherever Nebius or IREN or CoreWeave is, they can support them with capital or BlackRock or, you know, Jane Street even did a compute deal with CoreWeave and said, by the way, we'll also invest a billion. Yeah. You know, as part of it, they realized the capital problem but they, as a capital provider, has a creative strategy to get compute over other customers. They can offer capital and they can also buy a lot of compute at good pricing. I think that we'll see more of that. [00:26:55] Speaker 2: Absolutely. [00:26:56] Tanner: Nate? [00:26:57] Speaker 2: Yeah, I can say I 100% agree. If he comes out and says he's backing them and I'll be the big brother and save you, it doesn't look great, right? Like, they need to, they need to, you know, show that they can stay on their own two legs. And also on the demand front, like, again, I go back to my anecdotal experience working at a tech company. Like, the customers I'm talking to, there is no signs of slowing down, only ramping up. And this is from every sector across banking, across commercial, across consumer retail. Like, every company is just saying, hey, we want to do more. Can you help us? That's all I get. You know, that's what I would deal with every single day of my job. And so that's just my anecdotal bellwether of like, look, that's my, honestly, my first sign of, if they go, ah, we understand, we're going to, we're going to take a pause. That's where I'm like, okay, that's going to have the trickle down effects, but I'm only seeing the opposite of, we want to go, like, honestly, the big thing too is we want to go open source. We want to keep building. We want to keep understanding this, you know, agent movement. And so, yeah, the demand is insatiable is, you know, a rough descriptor of what's happening right now, in my opinion. [00:28:01] Speaker 1: So Tanner, is SpaceX just going to satisfy that demand fully? Because the pace that they built already, what Elon has said as well as possible, is that just all the demand, SpaceX is going to be the big winner, nobody else is winning. What are your thoughts with SpaceX? And then Nate, we'll go to you too. [00:28:18] Speaker ?: Cool. [00:28:19] Tanner: I'm going to be a little bearish here and say I, it's left unanswered right now. I think that the amount of compute that SpaceX can stand up is surprising. I think the margins that they've been able to put up are unbelievable. And this is in part due because they're not, and whenever I say they're not just a neocloud, I'm not even talking about the space program, I'm talking cursor. I'm talking the fact that they have their own models, grok bots at the frontier on the agent side, that that is allowing their operating margin to climb higher. And you're going to see companies like Google or Meta or some of those other businesses mimic that where CoreWeave and Nebius are just going to be, you know, probably step away from that, have nothing to do with the models and offer great software and compute on the cloud side. So is there a possibility where we see a glut of compute and that glut is caused from SpaceX and in fact they're fine with that glut because they have another product to sell? That's a possibility. Do I think it's going to happen? No. And the reason why they're building out 20 gigawatts or 13 gigawatts or whatever the number that you guys think is realistic, I think 20 is a lot. [00:29:39] Speaker 1: Yeah. I think probably 10 is a lot but I mean they could realistically bring like 8. [00:29:46] Tanner: But you know what? Like Meta said they're also going to get to 14 by next year. A lot's coming online in the next quarter. Who knows? And by the way it's not going to be yeah, let's not think 20. I think the actual number that they said was 13. which would put them on par with Meta as well. I don't know if it's necessarily going to put Nebius and CoreWeave out of business. I just think in the data center monthly conversation SpaceX has to be on that list because they hold up a premium but at the end of the day if rental prices start to go down and GPU prices start to go down they might have one of the great offers in terms of GrokBot and that's leading them to operating margin positive and none of the other neoclods have been able to hit that. I think they hit that way sooner than any of us would have assumed and mind you that was with depreciation and then on top of that that's with a 1.2 billion dollar one-time stock-based compensation with their acquisition of Cursor. so like at a quarter where you would have assumed negative numbers they still pulled out positive I think that was like that was the turning point quarter for me for SpaceX where I'm like wow you know what they do deserve a premium is it 2 trillion I don't know I didn't buy any but [00:31:17] Speaker 1: yeah it's a little rich but it's not as rich as I would have expected before the earnings yeah [00:31:22] Tanner: for sure and the second they showed off that earnings you realize okay they should be the most expensive but there's a spectrum you know [00:31:32] Speaker 2: yeah [00:31:33] Tanner: but anyway [00:31:34] Speaker 2: I think I think you're right to say that it's impossible to say whether it's extremely bullish or in terms of will it cause the glut like obviously if it becomes 13 15 18 gigawatts like I think it's almost impossible to map what that actually means for demand do I think demand then plummets to a point where GPU prices come down and there's not good economics not next year but who knows two three four years from now as especially SpaceX continues to build and continue to up their gigawatts as of compute I would agree that it's I'm not gonna sit here and say oh there's no chance or oh there's 100% chance like honestly I don't know now what I the question I'm just curious what you guys think and it could just be a minute point but you know SpaceX is a competitor in terms of the model space obviously with Grok and Grok AI or yeah technically Grok AI do you think these open source companies like a reflection would still just not give a crap who gives them compute as long as they just get compute and still work with SpaceX and you know yeah I would assume so right like a meta and everything else or do you think they prefer to work with an EBS who gives them open source like I would just curious [00:32:45] Speaker 1: yeah until the demand environment really changes and it's I think we're a long ways off of it a lot of it is coming online but it looks like demand is not going to be satisfied anytime soon at all and even like the way that Elon's talking although I'm a little skeptical on this he said hey I think SpaceX is going to use most of our own capacity like with what they've done and that's kind of the funny thing about SpaceX is when they were going ahead and building their own capacity it wasn't because they were like oh yeah we're going to become a hyperscaler they just were like we can't get freaking compute from anybody we got to just do it ourselves and that's actually it's kind of funny because that's one of the things that happened with Tesla when Elon was there early on they're like hey we can't get whatever it is that we need we just got to do it ourselves and you know they did want to be vertically integrated but they had to be even a lot more than they expected because of that but regardless ecosystem is growing you know absolutely rapidly but yeah like they've showed that they're very good like I like this comment another day another sector really SpaceX is certainly a direct major competitor with unique advantages last week telecom this week the neoclouds and financing is another one like if we look at their financing and the terms that they'll likely be able to get if they want to go through traditional debt it's still going to be really expensive but it's going to be closer to what hyperscalers are paying not what the neoclouds are paying so a lot of advantages there [00:34:08] Tanner: there's real advantages if you listen to SMCI or some of these names of the cost of time to compute like to get power online I think in SMCI they said that eight racks costs if it takes you 30 days extra to get those racks online would cost you approximately three million dollars of missed revenue and they're talking about companies that obviously have more than eight racks they're setting up hundreds they're setting up 1500 racks at a time you're talking about a lot of missed revenue so the fact that SpaceX is by far the fastest company to put to stand up a gigawatt at Colossus 2 is a unique advantage okay and there's extreme advantages if we go to talk about Nebius or something like this you know being native to the EU being all of like it's not something that the EU is super open to do or really any country is open to do is just open their doors and say hey come on in Google and Amazon and let's bring in all these American companies to you know build data centers directly for that one customer and we're not going to see the major upside they want to give that to local businesses and then contract out that compute and so I think a lot of the neoclouds are going to be around and thrive from localization as well whether that's in Australia UAE Europe because you're just going to pass a lot of that red tape much much quicker being native businesses there but yeah [00:35:52] Speaker 1: we've seen every single one of the big three neoclouds move in that direction this last quarter I mean they've been doing it anyway but over this last month so Nate [00:36:03] Speaker 2: the latency is a good point I mean even think about like an HFT high frequency trading firm they're going to be closer to New York than the one that's in LA or the west coast right definitely so yeah I think the latency for especially for Nebius that is now probably the most international kind of with all the other hyperscalers having that localization is definitely a big piece obviously of the story you know I think we also saw and we'll probably talk about in a second like they're also expanding to possibly Saudi Arabia or in the UAE you know pretty I think they're in like 17 or no no 17 13 different countries or something 12 different countries across the world and so that's going to continue only to expand which is you know the only similar type of companies are obviously the hyperscalers that have that latency from an international perspective [00:36:56] Speaker 1: yeah Cloudflare too but it's a different product but [00:36:59] Speaker 2: yeah fair yeah [00:37:00] Speaker 1: but yeah let's go right into Nebiusen since Tanner you don't have too much longer let's look at Nebius specific news Nate if you want to take the lead and kind of tag team this with Tanner as you get that and pull it up too I bought some Nebius sadly it wasn't today it was yesterday but I really like it'll be down another [00:37:20] Speaker 2: 5% tomorrow so no big of course yeah [00:37:23] Speaker 1: we can always it's the never ending dip on all these so that's the great thing is if you have money so all right here's your screen [00:37:31] Speaker 2: yeah so I think well usually what I do is I start you know a sequentral order from starting in September but let's do let's do the newest to the oldest because it's more fun now we have Tanner here for the next 10 so the big piece of news was literally this morning Nebbi has secured 3.5 billion of financing from Citibank let me actually find my other post that actually has it in more detail but essentially it's a let me actually find it real quick should have had it pulled up but I thought I didn't let's see here [00:38:00] Speaker 1: while you're pulling it up like the nice thing is it shows that people are still willing to fund these build outs [00:38:06] Speaker 2: yeah and I think the unique thing about this deal here too is it's only I've never seen this before by the way 364 days of maturity which I've never seen in terms of a debt deal means it has to be paid back within a year so by October of 2027 this should be fully paid off now in terms of the deal itself pretty good terms in my opinion you know you had a 6.5% on the low end and 7.45% of the cap and if you actually think about just on an interest percentage or interest amount basis it's about 235 million for the full $3.5 billion loan and up to 260 million so you're looking at you know an extra 65 60 million per quarter that Nebius now has to pay on their interest on the balance sheet but again it's a very quick one year loan it's very aggressive for what they're doing it seems like they're trying to obviously use this to build quickly and it's great because I think it's kind of what we're going to go into I think in the next couple years 7% isn't cheap but it's also nothing compared to the 9, 10, 11% things we've seen before and I think this will again [00:39:22] Speaker 1: keep good is this an interest only loan or are they paying down the principal as well I'm assuming interest only [00:39:28] Speaker 2: I think interest only I don't know I'm pretty sure interest only [00:39:31] Speaker 1: yeah wouldn't make as much sense if they were paying it [00:39:44] Speaker 2: talked about a little bit today but [00:39:46] Tanner: yeah I mean not really like the payback periods of the deals that these companies are able to sign like Nebius especially SpaceX I think they said their payback period now is like six months in length it's like even if it was obviously they wouldn't be able to buy the chips with this capital set it up and then also monetize it for a full year in that amount of time but it just goes to show you these types of deals can be so short because you can really do quite the payback in such a short period of time due to the pricing strength due to the demand that compute has right now so honestly Nebius has much smarter people than myself they know what they're doing here and by the way it's so surprising that it dropped so much today how long have they been saying that they were going to do this like it's just been constant we have $40 billion in backlog we're going to do a lot more than this in terms of the $750 million and then the stock was [00:41:09] Speaker 1: dropping dropping after hours all the data soon as they're falling more they're all going to zero together but not really [00:41:14] Speaker 2: yeah and then actually maybe I can share this as well I'm not going to do it I want to share a video remove it no yeah go ahead go ahead remove this one then I'll share this here this tab as well cool thank you let me know when ready I I just want to showcase the video of him being asked about financing and essentially said see if this will work with audio can you hear this [00:41:47] Speaker 6: a much more disciplined approach yes [00:41:49] Speaker 2: sir [00:41:50] Speaker 6: what's behind that is that your history in running Yandex is that sort of the philosophy you have just [00:41:55] Speaker 7: we're building a real company not very loud is measured by earnings yeah I [00:42:05] Speaker 6: behind that is that your history in running Yandex is that sort of a philosophy you have just [00:42:09] Speaker 7: we're building a real company real company is measured by earnings how much value you generate and you just need to be healthy businesses business we generate cash [00:42:20] Speaker 2: yeah it's just the most like Russian engineer response I've ever seen in my life we're just we're just a company we're generate cash I do love this approach though and this is why I invest in that because he just gives answers at the most black and white answer ever we're just going to be responsible and provide returns for shareholders and it goes back to why they've been so smart on the financing front and why they've taken it the way they have and that's why they're probably again given that premium now on another big piece of news that happened two days ago on when actually yesterday God time flies there's so much going on Nebi assigned a preliminary framework with Humane which is one of Saudi Arabia's largest sovereign wealth funds and DataVolt to be the anchor customer at Neon's OXYGON AI campus the details are up to 50 megawatts half of the first 100 megawatts now under construction phase one is planned at 360 megawatts total capacity comes online in Q4 of next year this would be Humane's first customer Nebius being Humane's first customer and this is actually signed at DTC Berlin so this news came out of DTC Berlin they probably kind of proposes there again it's very preliminary it's sort of like an LOI letter of intent it's not fully signed yet but it's just another kind of breadcrumb of Nebius continuing to expand and trying to secure power now this is a from what I can tell a co-location so again you know Saudi Reba comes in there has the capital works with DataVolt to build a data center and Nebius comes in puts in their GPUs and kind of just tries to get compute online right because of the demand but again it's great to see kind of goes back to the latency of just being in different regions I think you'd be the first of the neoclouds to be in the UAE maybe I don't know Roy I [00:44:17] Speaker 1: don't know of any others I don't think so [00:44:20] Speaker 2: I [00:44:21] Speaker 1: think actually I think they're one of the very small neoclouds is I don't remember which but [00:44:27] Speaker 2: and the great thing too about the UAE is it's but a lot less likely with delays from just getting on power online standpoint and again this would be in Q4 of next year so pretty early signings of it but again just a great deal that hopefully they sign and it'd be really cool to get the door with a sovereign wealth fund that's a big big play we're talking hundreds of billions of dollars so it'd be a great kind of customer to have in there thoughts on this dinner Roy [00:45:07] Speaker 1: I mean like I think the continued expansion outside the United States is to the benefit of most of the neoclouds and that localization that locality just being where they need to be like there's obviously some geopolitics related to the Middle East but I think that that helps diversify their business so that if there is something that happens in one country or region or whatever you you're not just only there [00:45:31] Speaker 2: yep now next piece of news before you get on to that I want to [00:45:42] Speaker 1: respect Tanner's time so Tanner if you need if you have a little bit longer great but if not are there any I've got like two minutes okay so give us any final thoughts [00:45:54] Tanner: no I final thoughts in broad strokes there's not the first sort of large amount of uncertainty that's been generated in this market at risk of repeating myself I just think that focus on the demand the demand will solve everything else you know new models will end up coming those models will end up generating new productive ways of using you know artificial intelligence then we'll need a lot more tokens and right now I think that's the agent boom you're seeing Claude Duet and Dots and Muse and all these and by the way those agents use way more tokens than we do way more that's real by the way you should look up the difference in the way that agents use tokens versus the way that humans do and then the fact that there's going to be like essentially 1000x the amount of internet traffic via agents versus humans this is all headed in one direction certain investments will outpace others and they're not all going to be equal winners but but I think in this type of you know just sheer gold rush for compute it would be hard to not win on most of these names but still be cautious doesn't mean that it's core weave can't find itself in trouble and that's why I'm just pouring new money into names like Nebius because they don't have those same debt concerns right now and you know they can start to ask their new customers about higher prepayment and higher dollars per megawatt that's also going up SpaceX has talked about it Nebius will follow suit yeah I apologize if I didn't answer everything in the proper way but I remain long I remain excited it's just this is a constant high beta market where whenever things fall these fall hard and whenever it goes up they skyrocket let's not forget these are the same companies that have gone from 20 to 80 from 80 to 300 and now we're just having to face some of these moments where it that nate put to [00:48:41] Speaker ?: and [00:48:41] Tanner: to out at to [00:48:43] Speaker ?: to to [00:48:43] Tanner: to [00:48:44] Speaker 1: out to [00:48:45] Speaker ?: to [00:48:45] Speaker 1: to [00:48:46] Speaker 2: no uh i there's no i some swedish stole my handle that's hilarious i will unbend that well thanks [00:48:54] Tanner: for helping on man we appreciate it guys thank you so much always appreciate you guys coming on my show i apologize for the short term i just have so much work with freaking bloom tonight um love you guys hope to see you guys again soon maybe we can catch another show um yeah i was gonna ask if we could do this on a friday or something i'm like i'm just so slammed today but thank you thanks for [00:49:15] Speaker 1: having me yeah thanks for coming on peace man later so real quick this is a good comment how come there's not much discussion about 80 of proposed data center builds next year won't even be brought online so go back to the the in scale uh comments that we had uh were hey they had to cancel that microsoft canceled the contract because of the power issues that they couldn't bring the capacity online until the 2030s which doesn't even mean 2030 it could be 2035. so that's that helps uh basically support the resiliency of the demand i don't think it needs any help i think demand is just absolutely crazy anyway but like that's that's not going to be that that's one of the reasons why nate and i and tanner and others primarily focus on the larger neoclap place because they've shown a strong track shown a strong track record of bringing online capacity rapidly being able to fund it build it get it done without any like unanticipated delays with a few exceptions there's always going to be when you build dozens and dozens of sites occasional times whether something it's delayed but um yeah this is more so i think the little guys the ones that don't have that resilient supply chain [00:50:26] Speaker 2: that don't have the talented staff and so forth this is the all birds and the the commercial or the consumer retail companies pivoting to data centers um yeah i think you even you know eric i think it's a great point though to be fair you know a lot of these smaller data center players like kind of you know core we've said oh my god what roy said i don't know why um i'm not changing [00:50:49] Speaker 1: my name to core even until we see 500 on it so oh my god i think it was called you dr then if we [00:50:55] Speaker 2: get by yeah like like roy said i mean it's the bigger players have kind of shown that they're not a part of the 80 percent i mean that'd be city centers nothing's been delayed yet we saw some rumors of it possibly earlier this year so far everything's on track and so obviously i'm aware of what's going on with the data centers and in the timelines and so i'm always being aware of that um but it is a great point like who knows what's going to happen it could get even more political over the next six to 12 months right like well this is not saying that's not going to happen right it's definitely [00:51:23] Speaker 1: possible and that's something i'm also going to monitor yeah the the uh with the midterms like i think the biggest political impact is going to be the gubernatorial elections and so that's uh most of the most realistic pushback against data centers has been either at the state level or even like closer locality and uh we if we see a strong blue wave that's also going to have impact and ramifications on the gubernatorial elections and so there will be a little bit overhang there too but yeah i think a lot of this is just mostly related to elections and we've seen at least in past election cycles not that the the major talking points immediately drop off but they they certainly lose a lot of impetus and steam the people are tired of talking about them they want to move on to whatever the next thing is and yeah that's a to our benefit because things are so beaten up but i believe you have quite a bit more uh nebius i do i do so let's look at that uh we'll do iron next and we'll wrap [00:52:19] Speaker 2: things up with core leaf perfect yeah so we i'm just going to keep going from the newest to the oldest just because i i want to keep people don't want people to go sleep um but the most recent one too was nebius signing or expanding with td sinx for those i don't know td sinx is essentially just a middle player that helps with go to market sales between enterprises palantir use them in their early days because they don't have a sales team i think essentially what they do is to connect enterprises and companies that say hey we need x service right and so in this case it'd be hey we need compute we don't have we don't know where to get compute and td6 is oh we have this vendor called nebius and that's literally what this this company does they're essentially like sales consultants that are the middleman between the vendors um and this is what helped palantir as well and so this is a great partnership they originally partnered with nebius for the americas last october so this signifies their expansion across europe which makes sense with the coinciding of you know gtc berlin i think that's exactly why this press release i think was like the night of gtc or something the night before but yeah this is great too to see them continue to expand their channels they obviously have the balanceer channel they obviously have nvidia as a big customer channel but this is another big one td cynics is a really really big uh middle vendor um for stuff like this and so that's super cool to see as well you know um next piece is another loi they signed so this was an loi to be the first customer at volts 600 megawatt gigafactor so again another loi to be the first customer on a at a site so essentially they're looking for anything they can get it's what is what this signifies to me i think this is the third letter of intent in like the past two weeks we'll talk about another one too where they're like hey we know you're building something we'll be the first one there like we'll be there don't worry we'll pay we'll be there we want we want compute yes they'll be the anchor tenant which gives them probably better terms by the way as well to be the anchor tenant so they're looking for anything they can get um and this specifically was 100 megawatts or phase one is 200 megawatt buildings which is about seven billion euros two and a half billion for construction four and a half billion for compute hardware power from the north sea offshore wind via eneco um apparently nebbius can just walk if the product gets delayed which is pretty interesting um and also the permits target by the end of 2027 so again another q4 probably of 2027 for power so we're looking again guys like a year out for you know power for these uh letter of intents uh but again just another way for them another channel [00:54:51] Speaker 1: for them to secure power so pretty cool to see oh yeah yeah power is like that that's really i mean it's continued to be one of those big uh focal points uh it's what it's why one of two reasons why i'm skeptical on spacex's ambitions as far as yeah that they'll be able to build that out is you've got to power it uh now they are going to do it very well but uh you just can't magically conjure up like gas [00:55:17] Speaker 2: roy what's your i have a question for you on this because i was talking with a few people of like maybe they can't get the technical watch if they acquire like because right now we have a cypher at three billion we have these other companies where they can just buy them for like nothing like do you think we're going to see some acquisitions over the next six months from spacex with these new [00:55:35] Speaker 1: clouds yeah i mean i think gartner was one of those that was talking about this too and a lot of what they had said i think is incorrect but we will see a lot of consolidation in the space and a lot of the smaller plays and opportunities will get consolidated into the hyperscalers um we can throw a spacex i think into the hyperscaler conversation now too and even like some of the neoclouds will be of of mind and appetite to gobble them up too because they're all beaten up terrible valuations it makes a ton of sense um i i would say the only reason not to do that is if there's some sort of contractual issue there um or you know that that company is just a mess but a lot of these i think are just right for acquisition uh especially at current valuations i don't think i think core we've like obviously is too large iron nebius i think that they they could hypothetically be bought out but i think it's going to start with the uh the smaller ones like you mentioned yeah like cypher keel wolf [00:56:29] Speaker 2: i mean there's so many names um yeah i don't know i was just thinking about that it's like right now especially right now they're down 70 percent like this is the time where you could you could buy cypher to premium and still only cost you six six billion which for spacex is you know pennies on the dollar that's nothing um for talking about 10 gigawatts you know it's like 20 of their shares or something yeah yeah it's like that's nothing uh a couple more things on my side we'll kind of all wrap up soon um next piece was nebius actually achieved nvidia exemplar cloud on the b300s um for training so if we take a look on nebius or nvidia's website now nebius has three which again i'm not going to sit here and say these are the biggest things ever but they're relatively important um nebius now has three of the gpus that their exemplar cloud provider for um the only one that has more than them is oracle with four so again kind of just putting them in the next i would say getting closer to a hyperscaler it's still got some time to go in my opinion um but just another little variable that they need in my opinion to become a reputable hyperscaler you know in that conversation right so um and then the last piece i'll cover well actually two more um they actually acquired a company called infrise so in a crazy i didn't even know this is a thing before they acquired this company but essentially it cuts the time to launch and scale large ai models again i guys i am not a software engineer or a devops engineer so i'm not going to sit here and say like i know everything what's going on all i know broadly is that it just cuts down the time to launch ai models on gpus something called cold starting which again that's a thing in cars and things i've done before i it makes so much sense that it applies to gpus but i never even would have thought about gpus in that capacity um and just a quick visualization of like what this means is essentially you know a deep seek v4 pro might take 23 minutes to cold start but with inferize whatever proprietary software they have this is now they can get it done in 20 seconds so it speeds it up 70x which is important when you're talking about basically spot pricing and having a gpu utilization at 100 this is something important and actually can add a lot of revenue over time um as a business so pretty interesting acquisition again i'm not going to sit here and go over the whole technical aspect of it because i don't to be honest fully understand it to the you know ability to explain it but it sounds cool and it sounds like it'll maybe save them some money and save them some time so it seems like a again a smart acquisition to both [00:58:59] Speaker 1: sorts of cloud platform yeah core we've launched something similar to that as well and i like like you i i was you know i'm in in this space like i'm not a an engineer or anything like that so i lean on those that are uh but yeah it's like okay cool cost savings good and then last pc news here um was nebby [00:59:19] Speaker 2: assigned another um deal with aib data centers for 50 megawatts of it load backed by 65 megawatts of power aib data centers is a small data center developer they're actually a public company which i didn't realize when they partnered with them at the time too um essentially they build and operate facilities made specifically for ai high performance computing um you know the deal is 50 megawatts of uh critical it capacity in the southeastern u.s um backed by 15-year utility agreement for 65 megawatts already in place so again just another data center trying to partner with someone partner with anyone really it seems like that has the qualifications and it has the power to come online um and pretty cool deal i mean this is great for nebius as they continue to try and expand um well yeah and any thoughts on this roy i mean it's kind of standard practice for them now to keep partnering with these companies [01:00:12] Speaker 1: yeah i mean it's not a nothing burger but it's kind of like when palantir just started signing contract you're like okay good another one like they're executing i don't know what more can be said uh i like two pieces of nebius specific stuff that you didn't mention and there's a lot like i know you're not coming everything i i liked uh arcadia also weighing and saying yeah we're still seeing these really elevated prices as well the you know the short-term duration a lot of it is spot price and then the bid markets as well and then related to that they launched the dashboard that kind of shows and illustrates that and and actually i don't know if you have that yeah go ahead and pull that up i i mean this is one of a few things that led me to actually buy some nebius and add to nebius like i i like this i mean i i think arcadia out of like the three is the best ceo uh like period i don't think it's close uh honestly and tranter's not really like the technical lead for corey so it doesn't matter as much but yeah regardless like he absolutely killed it at gtc they've been building the right [01:01:11] Speaker 2: with the right strategy um yeah makes sense yeah i actually had a bunch of posts on like the pricing day increases they have but i i was like all right that's just that's kind of both one thing um because they increase prices on spot on pay as you go on preemptive vm vm so virtual machines you also had this come out so they had a bunch of essentially over the past month price increases in pricing um transparency or as a company and something that i think could be interesting is as this is online for months and months like you could probably continue to find a way to use this as your analysis right especially as oh yeah a banker um and so it's really cool to see i'm assuming that core we will follow this and other companies should follow this as a way to like i hope so okay spot pricing [01:01:52] Speaker 1: it makes i think the well yeah i think all of them are heading towards that direction but obviously the obvious has the most capacity out of the three um right like iron's bringing some capacity too but uh on this uh going back to inference i don't know the answer to this and if you don't you know we can punt it but chai asked uh do you think inference acquisition is showing its benefits with the change in tos of removal uh removing the price ceiling now that they can bring gpus [01:02:17] Speaker 2: on or take them out of this i think even just from an m a perspective i don't think they're gonna be brought in house that quickly i could be wrong i mean they announced the acquisition literally three weeks ago so if they could move from acquiring this company to then having them built in house to then be able to remove that saline in the matter of three weeks like that's insane execution like insane execution and so from just from an m a perspective i don't know if they could do it that quickly um now what he's talking about everyone is this right here so yeah on october 8th we introduced dynamic spot pricing for preemptive vms under which the spot pricing was capped at one cent below the on-demand service rate now starting november 2nd they're removing the cap they're just saying all right like demand is so insane right now we're going to remove the ceiling for gpu's pricing and so nebula legend made a great post on this um and so why this matters is nebbis offers two ways to run gpu on demand so customers pay a fixed hourly rate for gpu access and spot customers bid for temporarily idle gpu capacity at dynamically determined prices um the catch is that nebbis can reclaim the gpus at any time so essentially historically that spot prices were capped at just below on-demand rates and so now they're just literally getting rid of the ceiling there's now no cap on on-demand rates um and so we kind of gave an example of why anyone would pay more for you know interruptible gpus kind of gave an example saying a customer's reserve a inner gpus but is currently only using 500 the remaining 300 are temporarily idle nebbis can make those 300 available to spot customers and if the original customer needs its cpus back nebbis can preempt the spot workloads and he he goes really into detail about it but essentially this is bullish again on just the demand there's they're probably just literally seeing so much demand they're like all right let's just get rid of this ceiling like why do we even have it in the first place um yes but again i think to answer your question on infrise i don't know if the acquisition was a part of that again it's only been three weeks since they announced it so it'd be pretty tough to bring in-house like quickly but maybe and if it is that's i mean that's great execution [01:04:29] Speaker 1: oh yeah yeah they've been executing very well uh of course caught up like everything else and iran has been caught up in it it's actually been the worst performing of these um i did end up buying a tiny bit just one option in iron and uh there's a lot of reasons why this right here is not the reason i i do think it has the biggest leadership of everybody but yeah so uh iran is a new official sponsor of a football club a very well known like this is a just like the golden state warriors makes sense if you want to sponsor a sports team why not one of the better known with a lot of traction but i don't know about that strategy it's one of the things that has kept me away from being a realistic investor but i mean there is a lot of good news and things going for it like they they have power and like that in scale when uh in scale was unable to deliver the power microsoft backed out of the contract i'm like i mean that's the bull case of iron is not the leadership it's not their technical ability it's the fact that they actually have power and they have capacity they're bringing online and and part of that was that we saw this month and there's by the way i'm i don't own a lot of iron i do track it so i'm sure there's news actually i know of some news that i'm not going to be covering same with core weave you know i talk about for you a lot i'm just going to focus on what's the most relevant here but anthropic itself was rumored so going back to a previous earnings call they said that they signed with a frontier lab so that's anthropic open ai and it looks like it's probably anthropic specifically it's just 50 megawatt but again um they have a lot more capacity coming online too helios is already starting up and running and delivering to microsoft as well so their revenue profile looks like it's going to change quite a bit and right here bringing the sites 245 megawatt of usable capacity online one month early would add 1.3 billion plus in revenue um so really good piece of news we saw the stock actually really performing very well right after that and then it got clobbered along with everything else um any any thoughts on that and then we'll get into like some uh two pieces [01:06:41] Speaker 2: from the ceo yeah again it's sort of back to the this one at least this one has a little bit better breadcrumbs or the real madrid deal is the ties to their spain site like at least i can kind of you can maybe conceptualize that a little bit of why they did it the golden state warriors one literally makes no sense to me but this at least has some sort of ties so like i understand the influence of being in that space also being in spain um i know jim made a post kind of saying like hey nebbia spends the same amount per year on um paying the fee for the data center right i believe regulation [01:07:16] Speaker 1: fees yeah yes the regulation yeah yeah that was his point is in a i miss that piece thank you uh nate and also neil uh the real madrid sponsorship plus 300 megawatt site in spain uh this week so it was [01:07:29] Speaker 2: yeah seems like it was tied together so yeah i think so again i think it's it's bullish and um great for them but yeah um just on that on the partnership yeah i think seven seven million years a year if they can if that's what helped them get the 300 megawatts in spain then so be it i you know no problem there um but i'd like to start seeing them close deals and producing revenue and then maybe the deal that these partnerships are sponsorships will start making more sense as they actually have [01:07:58] Speaker 1: cash flow coming from deals yeah uh absolutely agreed and yeah i um i really like to like i've been very critical of uh their leadership and i will continue to be like if i don't like something but you know of the three arcady absolutely killed it gtc multiple interviews just nailed it i mean he usually does um roberts was better than entranter and tranter basically the only thing he did was circulate an internal memo out uh to employees and was like hey we're good guys we're good um would like to see more they may be like they may be a little bit more conservative and like hey this is a quiet period before earnings but that's not stopping iron that's not stopping nebius uh from weighing in and so this i thought was a really good read a lot of questions this week on funding so here's how a gpu typically gets flooded iron first customer signs a multi-year take or pay contract and agrees to a prepayment on our recent three-year contracts prepayments have been 45 to 55 percent of the associated gpu capex i i believe do you know remember what nebius is is i think it's 50 to 60 i think is what they said i thought it was yeah a little higher courses i think somewhere between the two second the financing is committed secured by the contracted gpus on the microsoft contract we secure an a minus rated facility around at around six percent which together with prepayments covers 96 of the gpu capex more recently for a non-investment grade counterparty at mckenzie we secured financing at nine percent fixed different environment now they would not get the same rates but which together with the prepayments covers more than the full cost of the gpus that's the gpu side the data centers are a separate long-term asset base and we own and you could not have said this five years ago now we can because that depreciation cycle the three years and then they're useless that's been disproven it's it's certainly not in this environment i doubt that changes anytime soon uh but uh yeah he goes on and he talks about uh you know the depreciation um multi-billion assets and then he said here there's more customer demand that we can build for as we noted at our recent earnings 2026 is largely sold out recent three-year contracts are at more than double the prior year's pricing and demand for anything we can bring online in 2027 covers it several times over with prepayments a focus in every negotiation i like that piece right there this is one of the ways that these things are going to get built out it's not just iron it's not just nebius it's not just core weave it's all the little guys as well if you have capacity there's such a ridiculous demand you're seeing um you'll probably see higher than this for prepayments and yes capacity is going to be sold out well in advance as well meanwhile announced capacity keeps slipping across the industry this was before the spacex uh earnings which is the point the signing deals isn't the bottleneck in this market bringing gp online is and that's our focus and related to this and then i want your your thoughts as well this was a is this yeah this is like i've not seen this uh somebody mentioned this it's just a two minute clip uh from their ceo let me remove and reshare actually because we won't have sound otherwise um tab there okay perfect [01:11:07] Speaker 8: and also caught up with daniel roberts the co-ceo of nasdaq listed ai infrastructure company iron and asked him how much customers are willing to pay for soaring compute prices absolutely and it's an integral [01:11:22] Speaker 9: part of every negotiation we have with customers and this is what is really interesting about the current environment we understand the sentiment in public markets we understand the increased focus from investors but when we're talking from customers nothing has changed the demand is still there i mean jensen just mentioned 34 times increase in token demand this year alone so it is quite disconnected the two ends of the market but we're very respectful of that and we need to fund this in a sustainable [01:11:51] Speaker 10: way and what about the the rental or the sort of the revenue or the the rental prices for what you're able to get at the moment are those also continuing to rise as we've heard from for example over at spacex yesterday the earnings came out cfo was talking about some of the the rising rental prices are you seeing that similar um dynamic play out they're not going down but are they going up [01:12:13] Speaker 9: look it's hard to say because it's anecdotal data points right you're having conversations with 10 15 20 customers pricing refreshes once a month they're not going down the evidence suggests they are going up yeah but equally there's a direct interface between the price you receive per gpu hour and the level of prepayment by the amount the customer is willing to fund of your own build out [01:12:35] Speaker 10: you previously said uh 2026 demand is sold out you're talking about three-year contract pricing i think more than double uh last year are you expecting those contract prices to rise but the [01:12:47] Speaker 9: demand is still there so 2026 is done and on the next few months we're going to be quadrupling our operating revenue so it's about a billion of operating revenue today over the next few months that will go to four billion next year we are having conversations with customers covering multiples of all the capacity we can possibly bring online and in the last month or so 2028 is now coming into focus but the demand is there like the key question for us is how much of the funding of this build out can we get these customers to pay for us so whatever capacity you have online in 27 will be sold out effectively and 28 similar dynamic it is it really is like any gpu you can bring online you have got multiple people or wanting it out in berlin argent awesome nate thoughts yeah i think it's interesting [01:13:37] Speaker 2: as he says the last part i took away i think everything else is pretty generic we heard from the call of him just saying like the quadruple and four billion of operational revenue which i think is how they track air are um but what i heard at the end was basically the conversation now is how much can you pass and so i think we could start seeing 70 i don't 80 percent repayments would be pretty insane but i think we start seeing like a 70 percent repayment which then it becomes you know nebius for example i said they're going to do a gigawatt a year they only need to fund 20 billion of capex a year then rather than having to fund 30 to 40 because of that 70 percent repayment um so yeah it could be it could be pretty nuts but then it becomes a race to the bottom of oh who can who can we get to that prepays that let's just prepay the lease with like a cypher or a wolf because they need customers right um but then it's also well who gives us the best uptime and downtime on jpus so yeah i'm curious to see over the next few quarters of like what that prepayment gets to because i think it will just keep [01:14:34] Speaker 1: going up as demand keeps going up right yeah yeah it's it like the demand is absolutely relentless we're getting that read from everybody uh there and that that's one of the things is like with the prepayments usually it means like if there's zero prepayments if we lived in that world then you would have a higher pricing in your contract like that that's a trade office prepayments help a secure capacity that you are desperately needing and b usually get slightly better favor uh terms if you're the end customer um so the the trade-off is like right now the prices are so ridiculously elevated that things are so tight that it could be that nebius or core weave or iron is like okay we we are understanding that there's concerns with the capex so we're gonna receive five dollars less per um um watt uh that then we would uh otherwise but you're going to give us eighty percent prepayment or something like ridiculous like that yeah and that could easily happen like there i wouldn't surprise me if and it'll be kind of curious if we we hear just that on earnings calls or if that's included in the pressers with what they're happening and this one this this is not entirely true black bomb but there is a truth to this but um he says the other issue is that they lock in a price contract by the time build gets going their cost skyrockets the margin isn't what they plan for the the way that the supply chain works especially for these larger neoclouds is they have um these prices largely locked in place there is some room uh for movement depending on what product that they're specifically getting but a lot of them is like they're contracting those deals out well in advance it's um which is great but also if the prices plummet really rapidly some of those like core specifically with memory have pricing floor agreements as well so there's really not as big of a disconnect between um okay we contracted for this price but way way way more expensive like we saw for instance uh nvidia uh mentioned that they're going to raise prices on their units uh starting in january i mean that's to be expected but everybody else is raising their prices too so it's not um and not really uh it's not the the point where you're like hey we're selling or we're doing these gpus we're deploying them and the payoff period is going to go from like a year and a half to eight years because of something like that it might be you add a month or two yeah um so yeah iron uh themselves like i really liked uh the way that they handled those conversations uh they will bring on some capacity online like the leadership yeah people are are talking about of course corey's insider cells but they're talking about the kind of egregious i think it is egregious package that iron has there as well but um the demand is relentless um it it's very much needed what they have and so yeah i i thought that iran performed better than expected not in terms of the share price but as far as a company over the last quarter uh so like yeah i'll take a little flyer on it for earnings um anything else on iron otherwise i will cover core weave no nothing nothing else on my side okay let me go ahead and share my screen back again and so there's just a few technical type of things and we want to talk about some broader items as well um corey launches the core of zero trust ai framework keeping customers in control of access to the models data and rental environments this is one of those uh themes that we've seen although it's gained a little bit less prominence as of late uh just being able to control it and have uh your uh data sovereignty uh so rather than like worrying about like either you're running on these models that all of a sudden you are not actually in control of your own data that that's core weave or that's whatever llm that you're using this is a pretty good solution for that there's gonna be a lot of these there's of course already some of these like on the palantir and nvidia side as well um but yeah i really like that so together these six controls minimize corey's role in the customer trust chain and that's good a trustless system basically like hey we are you know you can run uh these critical secure data points and these workloads on corey and i think you know corey has like corey federal as well like when i read this they didn't mention i believe federal at all but uh you know that seems like just a natural client for them because a they're american company b they have atomic capacity c i mean they're really really fast because they operate kubernetes directly there uh on on the bare metal and then d you have products and offerings like this with the zero trust solutions so uh i think they're doing the right thing they understand they're in client and that's something that would be very attractive for them um beyond this uh corey've announced just like maybe it's just like iron as well some expansions this one is in india so adani connects and so they are launching in uh mumbai 240 megawatt data center as well and there's three different 80 megawatt buildings specifically um it sounds like that this is just the beginning of their expansion in india there's gonna be a lot of people in india that like there's there's so many places overseas like whether it's uae india wherever like ai is going to be global it's come in the united states but it is going global and all these neoclads that basically this they are have mastered these supply chain issues yes there are financing questions on them um this build out is going to continue relentlessly so any thoughts on either of those two um bits of news and i've got a few more as well no i [01:20:05] Speaker 2: think it's it's great to see it expand to india um i know nebius also has pretty strong confirmation of a site in hyderabad um india as well so yeah i know i know i know from speaking to a few people who actually live in southeast asia that indonesia is probably the best thing for your buck in terms of power and pricing so i do think core even nebius will probably announce something in indonesia over the next year or two as well um as they move more of their department teams out there and kind of [01:20:30] Speaker 1: grow their brand awareness and everything over there yeah it wouldn't be uh it's going to be everywhere and yeah there's going to be specific logistics there but like one of the things i liked about this is they're partnering with locals like the udana group is like massive like they they're able to handle those type of things uh so i don't really have any concern with that build out specifically um a little more news here so uh they had launched uh ford uh corey forge launched so they had this massive conference at the end of september uh i think they had let's see 1500 um 4500 customers um which is pretty amazing and uh this specifically uh just added another layer what they what they do uh improving agents uh and of course agentic workflow so here's basically the bullet points from this uh so they have corey area now available and it's a coding agent uh weight and bias models that was again from the acquisition um you know over a year ago uh cord notebooks agent lens and so forth and so all these are specific towards inference and agents and that's definitely the direction that they need to go as well um a lot of people like hey they're just bare metal that's all they do and that's just out and out a lie like it just maybe that was true before the weight and bias acquisition but it's no longer true at all anymore so um and let's see here what else did i have oh yeah so this one's really relevant so obviously corey bonds are moving massively and it's it sparked a lot of attention as well this specifically is uh one of those senior convertibles which is why you're not seeing as much of a change in terms of rates i guess i should have pulled up the yield to worse instead uh but what i thought was really interesting we had blackstone who reported earlier today let's see i think it's this one and uh blackstone is one of those that is a major backer of corey's debt and financing but also across the entire space and so the ceo he talked about the demand too but like especially at the end of this statement like think about corey like that's the biggest question is like is corey going to go belly up because of these bonds and stuff that's not what corey pays but it is indicating what the price of debt is for that specific structure but he says demand for computers accelerating we see no signs of that slowing down as ai use cases ramp for both businesses and consumers and the world increasingly engages with the vast potential of this technology at the same time there's a critical scarcity supply that is only intensified seldom in my career have i seen such a supply constrained market of this scale like these are not dummies blackstone like they see a huge swath of the financing needs of the world current public attitudes towards data centers will likely only amplify this scarcity case in point like the smaller ones they might uh they bear the brunt of that public outlet backlash corey fast 51 sites that are active nebius had they're going to hit 20 next year uh over time as the public comes to understand the advancements that have occurred in data center operations and destruction including with respect to water usage and funding for electricity these attitudes should change we'll see we also anticipate new laws and regulation codifying these advancements will be implemented of course there are risk factors to consider there will be situations where investors don't have sufficient capital and projects can't procure power in scale for instance equipment or the necessary approvals at blackstone we approach investments in this area with the same rigor and focus on risk adjusted returns that have defined our investment process for four decades in ai infrastructure for example our data center portfolio is built on durable long-term contracts with highly creditworthy counterparties so at least for this isn't like a specific decor we but i you know i wanted to bring it up for sure in core weave it doesn't sound at least to me from that like that these financing guys are are i i well i think there's two things one they're saying current deals that we have they're just fine and second it doesn't sound like that they're willing to just back away and say well we're gonna fund less of these or maybe not about any of these at all what do you think yeah no i think so too and it's [01:24:46] Speaker 2: i don't know i'm just perplexed man i've just i would agree that i see these companies like blackstone even like today city bank with nebius like it feels like what they're doing what they're saying are two different things um or at least what the market perceives of what they're thinking versus what they're doing or two different things i mean we're seeing these bonds and cds's go up um we're seeing the market essentially sell off you know high beta names and really a lot of ai because of this concern of paying back debt but these bankers are so willing to fork over cash um and truckloads of it and so it's i don't know that's kind of confusing i don't really understand what the market sees versus what's reality and again it could just be a very detached sentiment from market right now which it probably is like you mentioned earlier more tied to midterms and in oil in iran um but man it's it's yeah i think it's just another indicator in breadcrumb of like look blackstone's giving money city bank just gave money today to nebius for great terms it's going to continue to there's going to continue to be financing needs um and i think even spacex saying they're going to do even if they do eight gigawatts like the amount of hundreds of billions of capital that's going to take is going to have to come from somewhere um and i think elon to be fair to elon i think he would guide that knowing he has these conversations already started and knowing these things are coming so what they're doing versus what the market perceives are two different things that's happening right now and it makes no sense to [01:26:11] Speaker 1: me honestly yeah it's it's uh like you see the demand is relentless you see it's not going to be satisfied like even if elon's pie in the sky projection is is made like it just people don't understand like how the demand is structured it's it's not just if it's human adoption like based on like muse and personal assistants and stuff and and uh talking to chat gtp that'd be one thing or even like the enterprise workplace environment it's like well that's cool it's good demand it's going to grow rapidly but the agentic like what's happening with that in agents spawn agents and sub-agents and this tanner mentioned it the thousand x increase of non-human traffic that's going to be on the internet in the next five years that's a read from matthew prince from cloudflare like that's that's coming that is what is spawning the most demand and the more capacity that there's actually available it's in the the more efficient the cost per token like it spawns additional use cases and we've seen that play out this year that anytime we get close to people actually even beginning to think that there's not going to be enough um capacity then they're wrong so real quick two other pieces of news nate i see that you need to go do you have uh five more minutes uh i got two or three and i can hop and i got three okay um so we have the uh cluster max of course core we've joined uh by nebius as well iran doesn't really offer this so don't read too much on iran not being there um corey have also announced uh lmd 1.0 an open source kubernetes framework uh and so i really like that like they're very very strong with kubernetes they have the the uh project lead from google that basically set the the standard of manage uh kubernetes specifically and lastly corey of ceo defends company's financial strength this is the one thing i didn't like is entranter sent out an internal memo saying we're good they don't have what we built it's like do better man like honestly yeah it's you if everybody else is in the space is going on and doing interviews and weighing in and talking like that didn't change the price trajectory of any of those but like [01:28:31] Speaker 2: do better like honestly yeah especially when the i mean they were i wouldn't say they're the reason for it but at the same time like they were the big scare that happened and that's what shot a lot of these names down and was the talk of the town for a few days like as a ceo you come on you bolster confidence and you say no we're fine i mean i think you and i could go on and literally just talk about what we talked about for last hour and a half and do better than just like yeah we're good don't worry like that just that's just nothing for investors and you know there's no confidence at all and that [01:29:00] Speaker 1: was for more for like the internal employees and everything but like still like come on um nate i'm going to give you the uh opportunity to sign off uh us so this is gonna be a little bit shorter by the way we're going to probably do ours shortly after the earnings uh next month and so it'll be a little bit earlier in the month um and we'll look for a special guest uh next month as well but nate yeah [01:29:22] Speaker 2: final thoughts awesome yeah i think um everyone brace yourselves we could be there could be some more drawdown but also hey we could be close to a bottom just stay liquid uh i see a lot of people on x gambling please just stay liquid do not get into leverage or margins obviously not financial advice do what you need to do but at the same time be careful and be be a smart investor um yeah i appreciate all of you watching appreciate everyone coming every month we'll definitely be doing post earnings next month probably a little probably second week november maybe earlier the third week of november before thanksgiving but we'll definitely be doing one post earnings and that'll probably be a long one but yeah appreciate everyone watching appreciate everyone following along and please go subscribe to [01:30:04] Speaker 1: roy put them on my channel and we'll talk to you all next time and follow nate take care guys hang in there