Anthropic Takes xAI Colossus Compute, Shivon Zillis in OpenAI Trial, Kalshi’s $22B Valuation

0:13Welcome, everyone, to The Information's TITV. My name is Akash Pasricha. It is Thursday, May 7th. Today on the show, we are unpacking the latest news about XAI's deal with Anthropic. We've then got the latest from the Elon Musk OpenAI trial from our reporter who has been at the courthouse all week. And we'll end with a discussion about Calci's big funding announcement today and some exclusive reporting that my colleagues published about Polymarket. It's going to be a fun show, so let's get right on into it. Anthropic signed a deal with XAI, or as the company likes to be better known now, right now, SpaceX AI.

0:50The agreement is that Anthropic will utilize all the capacity at XAI's first Colossus data center for its own operations. Joining me now to break down the news is Raoul Martinique, the CEO of data center management company DataBank. Raoul, welcome back to the show. It's great to have you here. Hey, thanks, Akash. Thanks for having me. So what was your reaction to this news between Anthropic and SpaceX AI, it is now called? Yeah. Well, look, anything, obviously, in Alon's world is interesting. So, look, XAI, obviously, Frontier Model Company, you know, it was started a couple of years ago. They built out this colossal data center footprint in Memphis at a really rapid pace.

1:35But by all accounts, you know, and it's been reported, you know, the XAI, the Grok chat box and LLN isn't doing too well. You know, usage is significantly down. I think all 10 of the founders left the company, you know, Musk admitted that, you know, it kind of needs a reboot. So when I kind of see that and then kind of see what's happening with Anthropic where, I mean, the company's on fire, right? I mean, their cloud code is getting a lot of traction in the enterprise space. It's a really useful tool. I think the CEO said that, you know, usage has got up 80X in one year. So I think it's probably an opportunistic way for XAI to kind of offload the cost of this infrastructure to Anthropic, who desperately needs it to support their customers.

2:27And probably shows that, again, Anthropic is compute constrained and needs capacity. And XAI has excess compute given the not so successful progress of its LLN. Right. So the big question that I'm trying to figure out here is we also saw the deal between XAI and then Cursor, the other fast-growing coding company. And look, my read on that deal was Cursor needs its own compute capacity to service its own customers. I mean, I kind of thought that the Cursor compute needs would come from Colossus, the XAI data centers. Now we're hearing that the first Colossus facility is entirely going to be used by Anthropic.

3:13And so I don't know where that leaves Cursor. Have you heard anything? Yeah, I mean, look, Cursor obviously was, you know, the poster child for kind of AI adoption last year. Again, Anthropic seems to be taking over that kind of coding capability. You know, my understanding is Cursor's out there trying to raise money right now. They're struggling to do that. one of the reasons is you know they had reportedly on negative gross margins which i think is kind of you know that's another story that i think is now really starting to come to the surface right which is these ai companies you know they have these kind of flat rated tools that per seat tools that people can use but when they use them and because they're agentic they generate so many tokens and when you think about generating a lot of tokens what you should think is hey that's a lot of compute cycles.

4:02That's a lot of GPUs. That's a lot of power. That's a lot of cost. So I'm curious to understand what is the profitability model here, right? Because if cursor is growing like a weed, but it's got negative gross margins, you can't make that up in volume. Right. And the question that I have too is where are they going to get their compute from if it's not going to be from XAI? And I wonder how you might think about that question. Yeah, well, look, taking a step back, just broadly speaking, and as you pointed out, we're a data center developer, so we're bringing kind of data center capacity to market that supports a lot of these deployments from hyperscalers to large enterprise.

4:43Q1, we're just finishing up our deck to report to our board next week. It was a really successful Q1. If you look at hyperscale earnings, they were all kind of strongly up to the right, you know, accelerated growth there. They've all kind of increased our CapEx amounts. So right now, the sector is, you know, accelerating, you know, even after kind of some of the huge announcements last year's. Again, you got some cross currents, obviously, OpenAI is pulling back a little bit. But broadly speaking, I would say, you know, there's just a lot of demand for kind of compute capacity. And the hyperscalers are sucking up a lot of that.

5:23The frontier model companies like Anthropic are sucking up a lot of that. But remember, when demand increases kind of on a quarterly basis, you can't really match supply to it, right? That's because data center and then GPU and installing those in the data center, those things are measured in quarters, years, multiple years. So that's a much more inelastic supply chain than people - What about the - let me jump in here. What about the language around orbital data centers? This was an intriguing part of the fresh release here. I think it was about as vague as it could get. I think all it said was that Anthropic has, I think, expressed interest in, you know, the Orbital Data Center project broadly.

6:06There were no commitments. What is your read on that? Yeah, look, I mean, this is, I think that whole Orbital Space Center stuff is a bit of kind of the things that are going around SpaceX. I mean, let's pause for a second and realize that SpaceX, which is a rocket company, now basically has an LOM business, a frontier model company within it. And it has Twitter, you know, X, right? I mean, those aren't really kind of, you know, synergistic assets. You know, obviously, they're spinning it as a synergy story. I mean, the whole data centers in space story, that's another one where when you actually unpack it and think about the technical challenges that need to be surpassed for that to be a viable solution compared to terrestrial data centers, by all accounts, in my opinion as well, we're way off into the future, at least 10 years or more until that becomes viable, if at all.

7:03right because you're going to need to have kind of quantum reductions in the cost to move stuff into space you're going to have to have significant advances in optics to be able to move the data at the speeds that these gpu clusters need in space and you're going to have to solve the thermal problem which in space heat doesn't radiate you need to kind of support you know move it over a large surface so there's just a lot of a lot of things that need to be uh you know solve before that becomes real. So I think it's more aspirational at this point. Right. Now, we were talking before the show, before we start taping, we've got CoreWeave earnings tonight.

7:42I wonder just as you think about CoreWeave, what are the types of questions you have going into those earnings and what should we be watching for? Yeah, look, what I think people are going to be watching for is obviously their growth profile. you know, how is that growth accelerating? Is it kind of plateauing? You know, and then also their capital intensity, right? I mean, the CoreWeed business model, it's a solid business model. I mean, cloud is the same thing, right? You deploy infrastructure, you lease it out over time. Obviously, we now have this new just specialized infrastructure around GPUs.

8:18So there's, you know, there continue to be questions about, you know, the long-term, you know, life of that infrastructure. Again, the recent results are pointing to a sustainable business model, but I think all those things will continue to be looked at, but I expect them to have a really strong quarter. And just to bring a full circle here, Raul, I mean, after this XAI Anthropic deal was signed, I mean, there were a lot of people out there asking the question, including you know, there was a TechCrunch article headlined, is XAI a neocloud now? I mean, is that the business that they're in? Do you see that being, you know, I don't think this is what Elon is hoping for, but I mean, functionally speaking, that is kind of a business model that they're now in, right?

9:05Yeah, no, that's a true statement, which again, I think shows that this is probably more about just, they had this huge cluster, they spent tens and tens of billions of dollars on it you know they weren't getting usage out of it through their own products or monetization how do we monetize it ahead of an ipo where people are going to be looking at the bottom line so i think it's more probably an opportunistic um scenario than some long-term kind of grand strategy right well the rule i want to thank you for coming on that is raul martini the ceo of Databank here on TI TV. The information's Rocket Drew is covering the Elon Musk OpenAI trial from the courtroom all week long.

9:47Here is his latest update for us. Siobhan Zilis took the stand in Elon Musk's lawsuit against OpenAI. Zilis is a fascinating character in the history of OpenAI because she's played many different roles. She's been an advisor, she's been a board member, she's also been an employee of multiple of Elon Musk's companies, and she's the mother of four of Musk's children. So there was an open question of how her testimony is going to fit into the lawsuit as it's unfolding. And in fact, both sides kind of tried to claim her for their own and use her testimony as a lens to explain their broader case. On Musk's side, they sought to use her testimony to show that even as a board member, she was uncomfortable that the board didn't have any information about the launch of ChatGPT, and that she was uncomfortable with Musk's personal stake in Helion Energy, which is a startup that OpenAI has done business deals with.

10:45On OpenAI's side, they sought to show that when there were negotiations going on over what the future of OpenAI could look like and what kind of structure it could adopt, that a for-profit subsidiary was definitely on the table, but so was full-on conversion to a for-profit structure and also absorbing OpenAI within Tesla. Now, we knew that that was something Musk had proposed, but during Zillis' testimony, it became clear that that proposal was pretty fleshed out. And in fact, they were working on announcements and events that would go along with Tesla starting its own AI subsidiary. And that, in fact, Tesla was looking at hiring from OpenAI.

11:25It did eventually hire Andre Karpathy, who was an early member of OpenAI, a founding member, but was also looking at hiring Sam Altman and thought that Sam Altman, among OpenAI's co-founders, would maybe be one of the people who was most likely to leave to join Tesla. So that path for OpenAI and the idea of pursuing AI within Tesla was more developed than we previously knew. So Zillis was a fascinating character on the stand. She referenced her neurons. She bantered with the lawyers about their shared Canadian accents. And overall, I think, emerged unscathed from some of the worst accusations against her, which were that she was funneling information to Musk behind the scenes.

12:12Ultimately, she was able to explain that her role in OpenAI's history involved a lot of back and forth between different parties. And in her telling, she tried her best to make sure that she stayed neutral and was aware of any potential conflicts that came from her relationship with Musk. So that's how her testimony shook out. Musk's side still has a few key witnesses to call before they rest their case. They're going to play some deposition videos from former board members of OpenAI who were involved when Sam Altman was fired, and we're still expected to hear them question Ilya Sutskover, who is a co-founder, and eventually to question Sam Altman themselves.

12:51So that's what's coming up next in the Musk v. Altman trial. CalShit raised$1 billion at a$22 billion valuation today. It is the latest development in a fierce rivalry between CalShit and Polymarket. To that end, my colleagues have some inside details on the current state of Polymarket's business. I want to bring on Michael Rodden and Yueqi Yang, who published that story this week. Welcome to the both of you. It's great to have you here. Yueqi, I want to start with you. What did you make of the announcement this morning of Kaoshi's fundraise? So two things stood out for me from Kaoshi's announcement this morning.

13:29First, he said it's obvious that Kaoshi is trying to get institutional backing for the company. One of the investors listed in their current round is Morgan Stanley. And we know that Kaoshi's biggest rival, Polymarket, has the backing of ICE, which is the parent company of the New York Stock exchange. The other thing that stood out for me is that the valuation of CalShea at$22 billion is much higher than the valuation of Polymarket. We reported last month that Polymarket has been trying to raise funding at a$15 billion valuation. So that's smaller than the$20 billion valuation that CalShea has already received.

14:10And we'll get into more details about what explains the discrepancy between the two companies. Right. Well, so Michael, let's pivot then to talk about the story that you and you actually published this week. You did some reporting on Polymarket's US business. Tell us a little bit about what you found. Yeah. So it's very topical in terms of like Kaoshi pulling ahead in fundraising. Polymarket was sort of the market leader for a long time. And it's well known to be like sort of like the dominant sort of player in these prediction markets where you can put bets on the weather, the war in Iran, who knows.

14:49But more recently, Cauchy has really stripped ahead in terms of trading volume users. And it's had this big marketing campaign alongside of that. Part of the reason for its success is it has this regulated US entity that pretty much all of the trading flows through for Cauchy. um whereas polymarket it was banned from the us in about 2022 um because it was operating an unlicensed uh exchange so it actually moved its uh sort of legal headquarters to panama and has operated this offshore platform ever since and it's it's done using crypto like uh customers use crypto they use virtual private networks to to access the site um but more recently it's trying to move back to the us and become sort of like uh like kalshi like a regulated uh entity that's that's um and part of and you and you had some good reporting about about their ceo with the about the us entity ceo right yeah and and so like we were curious as to like why is polymarket uh losing the race against kalshi um and what we were curious about was the person leading the us licensed entity for Polymarket, who's this guy called Justin Hertzberg, lives in Florida.

16:11But a lot of what he seems to be doing doesn't seem to relate to Polymarket at all. He has a number of other sort of businesses that he spends most of his time on. And this role you'd think would be important because it's the exchange where all the trades are flowing through. He's the CEO or the nominal CEO of it, but he just doesn't seem to be there. And so we were trying to pick apart that in our story. And meanwhile, Kalshi is just racing ahead in terms of customers, trading volume, etc. So, Michael, did you get in touch with Hertzberg? Did he tell you anything about these other businesses or what his involvement is in the current polymarket business?

16:53Yeah, it was really strange because Justin Hertzberg is identified as the CEO in a couple of letters to the regulator, which is the CFTC. but he doesn't share the role of Polymarket US CEO on his LinkedIn. He doesn't mention it on any podcasts that he appears on, which are numerous. He doesn't mention it in his online biographies when he writes for, say, Forbes. He's the head of this company that white labels trading platforms for an area of the financial markets on the fringes called Prop Trading. And this is like where a lot of day traders spend their time. It's like gamified trading. It sort of resembles a casino in many ways, which is, I mean, I guess, not unlike the prediction markets themselves.

17:37But he's set up hundreds of these prop trading firms, which is a very interesting space because it's very sort of like on the fringes of the financial markets. It's very hard to like see what rules take place there. And a lot of people end up losing money by trying to gain access to these firms where they're trading sort of their own money and investors' money. But yeah, so he's traveling around the world, representing those companies that he operates. He's signing deals, like acquiring new prop trading platform firms. Meanwhile, Polymarket is just sort of not really at peak operational in the US.

18:20It's got a customer wait list of 1.3 million users who are waiting to trade on that US legal platform. but they've been having a lot of engineering problems in the back end they've had some some issues with their payments infrastructure that they're trying to rectify so their engineers are public polymarkets engineers are talking publicly about how there's been a lot of engineering problems and they're trying to fix them so it just seems like this situation where not everything has gone to plan for polymarkets return to the u.s and meanwhile this this person who's meant to be the CEO of their licensed entity is sort of off doing his own thing, which I think is very strange.

19:02Yuechi, I kind of want to get to the big question here around prediction markets. And we'll talk about how all different stakeholders are thinking about this question. I mean, is there a reckoning coming for prediction markets? You know, we can, Michael, I'll come to you in a second about, you know, what government officials are thinking about this question. But Yueqi, I mean, how is Silicon Valley grappling with this question? How are investors thinking about it? Traders? What are people telling you? That's a good question. I think on one hand, a lot of people, including investors, that back these prediction markets thinks that we're still at early stage of prediction markets.

19:41So far, most of the volume concentrates on sports bets. And so it is not yet entirely reaching a mainstream audience in terms of providing and allowing people who are not interested in sports to bet on this variety of other such as weather, such as economic data, etc. And that's something that digital markets are trying to build out so that they can diversify their revenue. But at the same time, I also talk to people who think that we're probably going to near peak, the peak of prediction markets soon, in a year or so, because the regulatory environment will change once we have a change in the administration.

20:28Currently the Trump administration is very friendly with prediction markets, but that may not stay the same once we have a different government. And there's also just this almost like indefinite headache with state regulators where a number of these stakes are suing prediction markets to prevent them from offering sports bets, which the states think are just equivalent to unregistered sports gambling offerings. So there are these regulatory headwinds that are coming for prediction markets. And meanwhile, these markets are just getting traction with users. Michael, you actually touched on it, but tell us a little bit more about how the Trump administration is thinking about the prediction markets landscape.

21:16Yeah, it's interesting because these platforms really came to prominence during the 2024 US election, which Trump won. And the Trump family really loved the prediction markets because, unlike a lot of the polling, they were very favorable in showing that Trump had pretty much like an unassailable lead. particularly after the debate with Joe Biden, where everyone pretty much, you know, couldn't We all remember it. We all remember it. And so, you know, Donald Trump Jr. became very sort of like intertwined with both Cauchy and Polymarket, a venture fund where he is a partner invested in one, I think, I'm not going to say one of the firms, but he's an advisor to both.

22:09And so they've been very supportive of this industry. And under the Trump administration, the CFTC, which is a regulator for contracts markets, has sort of rolled out the red carpet for these firms. They are representing the markets in the battles, in the legal battles with the states at the moment. The chairman of the CFTC has been on the public record very clearly defending these platforms and their ability to do anything, even when it seems very contradictory to the statutes on the books. But more recently, that's shifted a little bit. And, you know, there may be a variety of reasons for this.

22:51But recently, there was an army officer who won$400 ,000 betting on the ouster of the Venezuelan president just moments before US forces sort of captured him. And he was indicted the other day. Polymarket did notify law enforcement about the identity of that army officer who bet on the platform. But Trump was speaking to reporters and he sort of lamented how the whole world was becoming a casino and that he was never really in favor of it. That might just be sort of his, you know, Trump says a lot of different things at a lot of different times. So he might just be sort of responding to public pressure there.

23:40And the CFTC hasn't changed tack in its battle with the states where it's acting as sort of like a partner to the platforms in a lot of those cases. But the states hate it. You know, they fear that they'll lose a lot of revenue. from sports gambling and the regulatory environment sort of becomes very unclear when the CFTC starts regulating sports bets, which is usually the domain of the states. So it's working its way through the courts now, but for now, it's very contested. Uechi, can I ask you one last question? You know, prediction markets, I'm thinking about all of the startups that popped up around the crypto booms.

24:26And these are not just trading companies. There's all the different infrastructure companies. There's so many different layers to crypto that sort of gave rise to our whole ecosystem of startups. And we were talking just earlier this week about Andreessen's new fund and certainly Han Venture's new fund. Do people in your orbit, I mean, do they expect a new crop of startups to pop up around prediction markets, not necessarily the markets themselves, but I imagine trading technologies or, I don't know, maybe data analysis, you know, reading the markets, data analysis. Like, is this, is, I haven't even looked at it, but maybe if you look at the latest Y Combinator batch, I wonder if you'd see a bunch of founders saying, I'm going to take advantage of this moment.

25:11Yeah, there are already a crop of startups that are popping up in the prediction market ecosystem. I would say though none of them is generating as much revenue as the main platforms, which are Kaoshi and PondMarket. There are some startups that are providing analytics tools. I think Michael wrote about startups that are allowing people to copy the traits of what they call insiders, people who have this track record of making highly unusual, well-timed, profitable trade so that you can benefit from their strategy. I also know people who are building what they call prime brokerages for market traders.

25:53So there are startups that are forming and investors are looking at them. They're also looking at new prediction market platforms themselves. Some of these are trying to specialize in a certain type of bets. but I think most people agree that this is a market where it's currently dominated by the two biggest platforms and it's very hard to break their dominant as a new platform. Right, and Michael, are these startups any less murky? You know, we talked about this earlier in the show, but are they sitting on any better ground in terms of regulations? you know like if i'm gonna be fair like this is a very sort of like emerging industry you know this is a developing story this is a developing story like the rules have not been set um you know it's like so like even calcium polymarket you know these are run by really young people um you know it's it's strange like writing about this in industry because you know a lot of the times that people who are younger than me which is you know a reversal from like what i've normally done um but yeah you actually this is this is michael saying hey yeah i'm a little i'm a little old here for this um no but it's it's it's it's the wild west you know so it's like it's very unclear what the rules are you know a lot of these people building a lot of these apps that sort of leverage um these trading networks uh run by anonymous founders you know it's it's on telegram you know they're they're using vpns to access the sites you know it's it's flooded by like market makers and, you know, people who are looking to like sort of front run trades.

27:35And, you know, it's very different to traditional financial markets or what you'd expect. You know, if we live in a very strange time, even in normal financial markets and in regulation there. But yeah, the prediction market space is still, you know, you have to have a strong stomach to wade through there. Right, right. Well, I want to thank you both for coming on and helping us to make sense of all of it. It was some great reporting. That is Yuichi Yang, our crypto reporter, and Michael Rodden, our finance reporter, here at The Information. That does it for today's show. A reminder, we are on this stream Monday through Friday at 10 a.m.

28:13Pacific, 1 p.m. Eastern. If you can't make it then, episodes are available on theinformation.com, on our YouTube channel, or wherever you get your podcasts. Make sure to follow us on social media, on X, Instagram, and TikTok. I'm already excited for our next show, tomorrow. Have a great rest of your Thursday. Bye-bye for now.