Elon Musk’s “Ego” SpaceX-xAI Bet, Ranking AI Models, Groq’s $7.6 Billion Payout

0:13Welcome, everyone, to the information's TITB. My name is Akash Pasricha. It is Friday, February 13th. It has been a volatile week in the markets with AI fears fueling sell-offs across the market, hitting software, wealth management, real estate, and logistics stocks. Pinterest shares are also tumbling after the company reported its slowest quarterly revenue growth rate in two years. Today on the show, we are kicking off with our editor's cut. We are breaking down the bold bets that Silicon Valley is making right now and unpacking how close technology like orbital data centers and humanoid robots are to becoming a reality.

0:52I'll then speak to the CEO of Arena, formerly known as LM Arena, about the current state of AI benchmarking and whether today's model rankings actually reflect real world performance. Finally, we will wrap the show with our AI and finance reporter who has two major stories to unpack, a$7.6 billion payout to Grok shareholders, and this weekend's big read on Blue Owl Capital, one of the key financiers behind the massive data center buildouts powering the AI boom. It's going to be a fun show, so let's get right on into it. There has been a lot of conversation lately about the far-off bets that will take Silicon Valley into the next era of technology, Orbital data centers and humanoid robots, for example, have seen some good progress.

1:37But based on where we are today, it will still be a while before these tools can become a commercial reality. This week on The Editor's Cut, I want to bring on Martin Peers and Nick Wingfield to help us put some of those bets into context. Martin and Nick, welcome back to the show. It's great to have you here. Hey, Akash. Thanks for having us. Nick, I want to start with you. So you have written about some of these far-off bets for our Weekend magazine. tell us a little bit about what your observations have been about how the dialogue has changed around them recently. Well, if we're talking about orbital data centers, the real reason that this has been thrust into the kind of public discourse is the SpaceX IPO.

2:19I mean, it's pretty straightforward. This is an IPO that is expected to happen in the second half of this year, the company has said, Elon Musk who said, and it's become a major plank of the SpaceX story, and one that I think we're going to increasingly hear from Elon Musk. Now, there are a variety of reasons for it, one of which, if you listen to Elon Musk, he'll talk about 24-7 free power from the Sun. There are other benefits. There's another thing that I think also we're starting to hear more about, which is really the difficulty of building these kind of multi-gigawatt data centers on Earth.

3:04And so in effect, NIMBYism, you know, people who don't want data centers in their backyard on planet Earth, may be driving some of the dialogue that we're hearing about putting data centers in space. But Nick, have there been any major breakthroughs at all lately in the last couple months? We know the narrative has changed and SpaceX has brought it to the forefront, but has anything changed around our capability to actually commercialize or build or launch this technology? The biggest change is with the transportation to orbit. So SpaceX has now successfully tested their large Starship rocket multiple times.

3:49And that rocket is critical for lowering the cost of getting kilograms to space. And you really need that kind of capacity in order to deliver the number of satellites that are going to be required for these orbital data centers to work. So that's critical. There have been some other tests that I don't think have really demonstrated that you can do these data centers at scale in orbit. And we've also seen, you know, some research from Google, which is interested in this type of technology as well. But really nothing that has proved beyond a shadow of a doubt that orbital data centers can function in orbit.

4:34So, Martin, if Elon wasn't behind these particular bets, you think they would be in the headlines at all? Sure. I mean, I think Nick actually wrote a story about this a couple of years ago, right, Nick? Well before Elon sort of embraced this. I do think, though, we should step back and look at what's really going on here. really Elon has used to has become used to being the center of the tech world right he gets an enormous amount of attention much of which he drives you know by putting out these ridiculous announcements and a couple of years ago his position began to be threatened because OpenAI was suddenly getting all the attention so Elon who had been involved with OpenAI from the beginning but then had had a falling out.

5:25Elon starts his own company, but unlike OpenAI, which has raised a huge amount of money from various people, Elon hasn't got the money to fund OpenAI as much as he needs. So what does he do? He creates this ridiculous thesis about sending data centers into space justifies merging XAI into SpaceX on that thesis, which really doesn't make any sense at all. And now we're all sort of talking about it. It's crazy. This is really all about Elon and his ego and his desire to be in the middle of the whole debate. I think in the end, Elon will not be the main mover in AI and the people who are investing in the SpaceX XAI my IPO are going to be disappointed.

6:16So, Nick, one of the questions I had for you was, look, when the internet was being developed in the 90s, was this how people talked? Back in the dark ages when Nick was a young lad. Okay, well, maybe, Martin, if you want to admit to being old. No, no, Nick's old than I am, so he's... I was going to say, I was like, Martin, I was going to Nick to be courteous to you, but you could jump in. So, look, my question really is, isn't this the way you have to talk about these moonshot bets to get any kind of attention, fundraising, traction about it at all? I mean, what is the alternative, I guess? Well, we used to talk about things like this back in the 90s as vaporware.

7:01And, you know, you would announce, then the moonshots weren't, you know, orbital data centers. It was the next version of Microsoft Office or a database or something like that. But very often, companies would talk about these sort of far-out products. But those really were, I would say, a lot more achievable. And the logical leap to get to those next versions of the product that didn't really exist weren't as great as the ones that we're seeing right now. Which is not to say orbital data centers will never work or that robo taxis from Tesla, at least, will never work or optimists. I mean, they probably will.

7:47It's kind of a question of the time horizon. And so, like, do you need to talk about this? I think you do need to talk about this now if you want to tell a story to investors. And that's largely what this is about. I don't think and maybe it's also partly to spook the market, you know, in competition, because, you know, you really can have a broader impact if you start to talk about these things well in advance of when they're ready. So, Martin, what about from the investor's perspective? I'm trying to think here, you know, what could have changed in the last 25 years? I mean, people talk about where investors, what asset classes they look for, returns that, you know, exceed what the market is giving them.

8:37And we've seen the venture capital industry kind of, you know, grow over time. And I'm just trying to think openly about this. I mean, could part of this be that, I mean, you know, you have to take on more risk today. You have to sort of make bigger bets. I think what has changed, I mean, obviously many things have changed, but I think a big thing is that this focus on individuals, this sort of blind belief in individuals, I mean, we see it in the political arena and we see it in the business arena. And Elon has created this sort of image of himself because he's had success with SpaceX and Tesla as this just, you know, visionary that people will just, they have put blind belief in what he says.

9:22But if you actually watch what he says, he's constantly wrong. And I just think people need to be a little bit more skeptical of some of the more outlandish things that he says. and particularly if you watch him on X, he says the most outrageous, extraordinary, bizarre things. I think his success has really gone to his head and he just now, he's like, you know, believes anything. You know, he's kind of like believing his own BS kind of thing. Nick, what do you think could change this in the end? How could this narrative shift going forward? Is it a business going wrong, a technology not panning out, investors getting sick of it?

10:07I don't... You know, it's hard to imagine Musk being more wrong about orbital data centers than he has been about fully autonomous Teslas. I mean, he's been promising that now for years and constantly missing deadlines. And I mean, depending on how you view the potential repercussions from that. I mean, Tesla's stock certainly has suffered, but for a long time, it hadn't. And, you know, I think this is just built into the bloodstream of the tech industry. I mean, we haven't even talked about it, but there are some non-Elon players out there who are also making some really bold promises about what AI can do, about what agents can do, that agents are going to, you know, kill software as we know it.

11:00And none of us really know if that's going to happen. And will they pay the consequences of those types of predictions? You know, it's hard to say. It certainly seems like there's not been a huge penalty for Musk, but he may be a a unique case? I mean, I would say that on that point, the fears about AI hurting, you know, Salesforce or servers now, you can understand where that's actually coming from. And that might eventually end up happening. It's just that there might be a different, it's just a question of the actual timeframe. But Elon, I mean, he is the one who's been saying these most ridiculous things about AI that, you know, people won't live in poverty anymore.

11:49Healthcare will become magical. I mean, the guy is like smoking. I mean, he's obviously, you know, he's on something because you can't believe anything he says, really. I think it's just, you know, fiction. Right. Okay. Well, Martin and Nick, I want to thank you for coming on. That is Martin Fiers, our co-executive editor, and Nick Wingfield, our features editor here at The Information. Benchmarking has been very important to the story around AI models in recent years. Of course, there has also been a lot of conversation around benchmarks, not actually reflecting how successful certain models will be in the real world.

12:27I want to bring on someone whose business is at the center of all of these discussions. Anastasios Angelopoulos is the co-founder and CEO of Arena, previously known as LM Arena. Anastasios, welcome to the show. It's great to have you here. Thanks for having me on. Tell us what is Arena, formerly LM Arena. Arena is a company dedicated towards measuring intelligence. So what we do is we run a benchmarking platform. It's a live evaluation, unlike any other evaluations, where real users, tens of millions of them, are coming to our site, arena.ai, using AI for their daily work tasks, whether they're doing software engineering or legal tasks, medical tasks, writing.

13:11and so on. And in so doing, they're giving us feedback that we use to construct a public leaderboard that's shaping the future of the AI industry. So I think a lot of us will know Arena and LM Arena from that leaderboard that we've all been refreshing to see every time a new model comes out, who is on top in terms of text and video and and audio and stuff like that. How do you actually make money? The way we make money is by helping labs and enterprises evaluate models pre-release or AI systems pre-release. So what happens is that during the course of developing a model, before, let's say, labs want to release it to the public, they might want to try it out on some users, you know, to get the distribution of real-world use and see how well they're doing and try to understand its strengths and weaknesses know how to improve whether that be hey this model is not so good at math and then they maybe go and collect more math data to try to improve it or whether that's coding data i'm trying to understand where maybe it's making making errors maybe it's making code that doesn't compile maybe not solving users tasks we help them understand this on the distribution of you know our real users through these cloaked models with these code names.

14:32And that's a service that we offer to providers to help. So these are like the major the major AI labs basically are your customers. Major AI labs and also enterprises. Yes. Okay. You should know that the public leaderboard, the one that you see on arena.ai slash leaderboard never touches money. Model providers cannot pay to be on the leaderboard. They can't pay to be off the leaderboard. They can't pay to change their score. and all public models are evaluated according to the same open source methodology that you can see on our GitHub. You can go to our arena rank repo and it shows you exactly how we take the public's votes and we turn them into scores.

15:12This seizure only applies to pre-release or, you know, cloaked models where providers are trying to sort of predict how well they would do if they were to release their model. Got it. So this is, I mean, this is kind of like, you know, know the church and state sort of system you have to have here which is that your your b2b business which is helping these ai labs assess their models pre-release I mean it sounds like that has to and is completely separate from once they are released then you have another team or you know another another operation basically which is allowing people to rank them I mean one one cannot really talk to the other well you know I think the the neutrality is driven by um people's faith in our scientific rigor.

15:57We're all former academics. Myself, my co-founders, Wei Lin and Yan come from Berkeley, where we were running this project as PhD students. Well, Yan was our professor. But Wei Lin and I were PhD students at the time. And just thinking about, hey, how do we do this? It was at the time of ChatGPT. And nobody really knew how to compare it to language models. You would put them on static benchmarks, like an MMLU or something. And it'd be like, oh, well, model a is better at taking the tests than model b but then you put them in users hands like you try to chat with them and it all you know basically fails it's like you it's like you can't talk to this model and it's clearly so dumb so what we did in order to fix that and it wasn't just us it was a large group of students at berkeley is create this arena type system that kind of became you the gold standard way that such chat systems are evaluated and has sort of propelled to this point, we remain very scientifically oriented.

16:57We continue to publish papers on our methodology, again, like with the very open source system. And by the way, if you think about our incentives commercially, trust is at the center of it, because if any model provider starts feeling like, hey, these guys are not in the game to be neutral, and if it's not a neutral playing field, then the whole platform loses everything. So I would say that's the real, that's one of the incentives also keeping us neutral. What do you see as the future of your business then in an environment where these models become a little more commoditized? And by that, I mean, you know, there's this whole discussion around right now today, the labs are all fighting with each other to have the best model.

17:41And every time a release comes out, everyone rushes to the benchmarks to see how they're performing. But look, there has been a discussion around three, four, five years from now, you know, will the competitiveness start to, you know, converge a bit? And will you really differentiate yourself based on the application that is installed? And in a world where the models do come commoditized, what is the future of your business? Well, let's distinguish between two things. There's the models becoming commoditized, and then there's the space fragmenting. so models can become commoditized while the space also fragments and by that i mean the cost of sort of the margin of inference can be driven to zero but at the same time what can happen is that different providers and different you know models are good at different things and that's kind of the world that we see ourselves going towards cost of inference certainly going down um you know a of these companies are talking about driving the cost of intelligence to zero um making sure that you can have like intelligence in your pocket it's at essentially free cost too cheap to meter blah blah blah but also we're seeing that different providers starting to specialize in different things so as an example like the consumer war is kind of being fought right now between chat gpt and Gemini but the enterprise war is certainly being you know it's not necessarily being led by Anthropik in terms of dollars that Anthropik is is kind of positioning themselves there in the market and growing quite quickly and so if you believe in a world where many flowers will bloom and there's a huge amount of space for AI to you know capture dollars everywhere in the economy then the natural incentive for these companies will be to sort of tile up the space and fragment and in that kind of world our business is actually quite valuable because we can solve that fragmentation problem for users we can help but you don't you don't think all the models in the long run will sort of because you know the an anthropic coding model against uh opening eyes coding model etc i mean right now the the rivalry that we hear about is which company has the best model for the particular use case that you're talking about.

19:58I mean, my read on the discussions was that in the long run, the cost of inference will come down, as you're saying. Also, it won't really matter who matter who has a better model because the differences will become marginal in some ways. It sounds like you're saying that you don't think that the differences will become marginal. I don't think the differences will be marginal. It's difficult. You know, part of the issue is that it's difficult to personalize to people's taste. And so all these models are going to have different tones. You know, it'd be like saying humans will commoditize. It's like, you know, yes, the models will become more and more intelligent.

20:36Okay, they might be, but there's not a one-dimensional axis called intelligence that these models are going to commoditize under. You know, you go to different people and you like different people different amounts. It's like, they might all be smart, but the way they talk to you, the way they present information, the particular set of skills that they have, are they more geared towards this? or they more geared towards that you know let's say i'm writing a scientific paper um maybe i prefer the the sort of um way that the scientific paper is being written by one model versus another maybe one model has a more empirical taste maybe it uh i i find it to be more careful on experiments whereas the other one maybe is more theoretical and you know it's just like these issues of taste are going to matter so much to the future of ai um and it's going to need likely human in the loop workflows and by the way expanding the platform that we have towards agents yeah even beyond the model and then assessing the agents then right yeah absolutely and human in the loop and that's where our platform has a particular advantage because we have tens of millions of users here that want to use the most cutting edge agents to do their jobs so now all on the human in the loop piece and this is a slightly slightly different piece but after the models are released then you You know, we know that reinforcement learning and human feedback is becoming increasingly important to helping to tune these models.

21:55How does your business then fit into that story? And how do you sort of connect to that phenomenon? Well, I think the story that we are seeing slowly play out in, you know, all of the AI story is that every real environment is becoming an RL system. right whether that's Salesforce whether that's arena whether that's you know a little environment that an RL environment company makes and sells to a lab you know or whether that's the real world via you know a robot going around picking stuff up and seeing whether it can actually perform tasks the purpose of arena and the reason why it's such a valuable platform is because it represents real users it is always live by um winning on the arena a model is able to predict the future because it's able to predict which answer a human will like better it's able to give them what they want for a new user coming in with a new question and ultimately it's interacting with reality and And reality is impossible to game.

23:09That is the future that I see, you know, benchmarks moving away from these sort of static benchmarks where it's just a list of questions and a test. And then you see the test once and it's over. You've kind of memorized the answers. And towards a live platform where you're really measuring utility to real people, them in the loop doing their work with you. Let me ask you one last question before you go, which is we see the models getting better and better. and we see them on the leaderboard as well, companies moving up and down. That is a very different story from AI adoption in the enterprise, which is a much slower moving story.

23:49What do you think are the core reasons that AI adoption hasn't taken off the way people would have hoped in the enterprise? And yet on your leaderboard, we see models getting better. I mean, they're getting so much better, right? We saw with 5.3, 4.6, they can do things they can't already do. Why? Well, let's just, you know, make one thing clear. Enterprise adoption, although sort of lagging behind consumer adoption, is not exactly slow. There's a lot of enterprises that are - Maybe compared to what investors are hoping for. Exactly. So, you know, it's not like the ecosystem is like sluggish and, you know, whatever.

24:28this is the fastest moving economic trend that's that i've certainly seen in my lifetime and probably that anybody has seen and the models are really continuing to improve um the the gains are substantial we saw it you know last week um with opus 4.6 that that model has added so so substantial leap in coding performance and in general performance over the last generation of of models from anthropic and um so the the purpose of this leaderboard and so okay why why is enterprise adoption moving slowly you can actually look at surveys of this and the number one issues relate to reliability it's am i able to safely deploy this model am i able to you know predict how well it's going to do on my users am i able to upgrade it am i able to understand whether it's going to be compliant with all the rules that I have in place?

25:24And sort of, can I measure its performance and continually improve it on my users? These are a lot of questions that enterprises are asking, which is why enterprise evaluation, which is yet another area that I think Arena is positioned well to attack, is poised to be one of the most important problems of the enterprise adoption story. Right. Well, Anastasios, I want to thank you for coming on. That is Anastasios Angelopoulos, CEO and co-founder of Arena here on TI TV. Shareholders of chip company Grok, which late last year signed a$20 billion licensing deal with NVIDIA, have finally started seeing some of that money flow back to them.

26:11Our AI and finance reporter, Myles Krupa, had some great reporting on how much they are getting back. And he also wrote this weekend's big read about Blue Owl Capital, one of the biggest data center financiers around. I want to bring on Myles to talk about both of those pieces. Myles, welcome back to the show. It's great to have you here. Thanks, Akash. Let's start with Grok. So we know that the headline number, Christmas Eve 2025, was$20 billion. It was the big licensing deal that NVIDIA did. You had some reporting this week about how much of that money is actually coming back to investors and how that number breaks down.

26:48What did you find? Yeah, so what we know so far based on what Grok has told shareholders is that Grok can expect to get about$17 billion in cash from NVIDIA over time through this licensing agreement. Now, that's going to happen apparently in three stages. And what happened this month is that shareholders got their first payout from the first phase of the licensing payments. So shareholders were told that three quarters of their stock would be redeemed and that that would result in$7.6 billion in proceeds to them. and and i i just i mean there there's still questions here that we have but the the 20 billion uh you know reconcile that with the 17 billion dollars in cash so the three the three billion dollar hole here i get that's still a question for us to who where that money's going who who's getting it it's a bit of a question mark i mean we could speculate about some things that might sort of fill that hole i mean there could be um potentially payments to some of the executives going to NVIDIA.

28:03There could be other things, earnouts. But from what we know so far, it's pretty clear that NVIDIA is preparing to pay Grok$17 billion. Right. So now let's compare this to some of the other licensing deals that we've seen other big tech companies complete. And I want to look at this from the perspective of investors who put money in Grok, because you wrote this in our venture capital newsletter, Dealmaker. How does the return that they are posed to get from the Grok deal, how does that compare to how investors did in the inflection deal or the Character AI deal? Yeah, when you look at the sort of last round investors in these different companies, it's kind of in the middle, roughly in line with what others have gotten.

28:56You know, Character was a 2.5x payout to its last round investors. Inflection AI, which went to Microsoft and kicked off this whole thing, was a 1.1x return for its last round investors. And here, investors in Grok's$750 million round last year, they're expected to get, you know, a 2x on the price they paid for their shares. So, you know, it's pretty much in line and it's a decent outcome for sure because that last round was only three months before this licensing deal. So it's a pretty quick return of cash to those investors. What do we know about what is going to be left of Grok after this licensing deal is completed?

29:43I mean, you know, we've heard about these shell companies that sort of remain. I presume there's some IP there that NVIDIA is not getting. Do we know what's happening with that? Right. Well, so NVIDIA's license is non-exclusive. So Grok can go out and license that IP to other companies as well. And the other thing that Grok still has is this cloud platform they were building using their chips. So, you know, that had some customers, I think, tens of millions of dollars in revenue. um you know i in my mind probably the most likely buyer is some other chip company that will want to buy grok basically to sort of lock up its technology so that no other chip companies other than nvidia of course um can have access to it you know i guess the thinking goes if nvidia found this you know worth a 17 billion dollar licensing payment maybe there's something there and maybe the best move for a rival would be just to sort of buy it to lock it up.

30:50Right. Let's pivot to talking about your weekend big read that you wrote this week about Blue Owl Capital. Why is Blue Owl Capital so important to the AI story right now and who are they even? Yeah, you know, they kind of came out of nowhere last year or 2024, rather, and started doing some of the biggest data center deals in the US, especially everything from OpenAI's Stargate project in Abilene to Meta's massive data center in Louisiana. and they were willing to invest billions of dollars in these projects not as a lender but by owning the equity in these kinds of special purpose vehicles that are being set up for data centers and so you know they're shelling out a lot of money and they're willing to take the riskiest part of these deals which is the equity so they're doing something a little bit different than the rest of the market.

31:53And they're doing it, you know, with pretty substantial speed and size. And, you know, kind of the part that is interesting about the Blue Owl story to me is that they are taking on a lot of the risk that the AI hyperscalers, the labs, you know, companies like Meta, for example, are keeping off their balance sheet in some ways. And, you know, you see it reflected in the stock price of the AI companies. I mean, you know, but they're all doing quite well from the AI boom. And yet Blue Owl Capital, I mean, their shares are down something like 50 % over the past year. And so I wonder where that leaves Blue Owl Capital.

32:33I mean, you talked to the CEO, the co-founder. What were the questions you had for him about this? Yeah, you know, I really wanted to understand the financial calculus. and you know it turns out that the kinds of investments they're doing in data centers is following a playbook that blue had blue owl had already sort of established in other contexts before in its real estate business so they were already you know doing similar deals with amazon for instance for amazon distribution centers and the way they structure them, they actually get regular payments on their equity. And in these data center deals, it's very, very hard for a tenant like Microsoft or Oracle to walk away before their 15-year lease is up.

33:29And so there are all these sort of mitigating factors that I understood better after the interview that make these equity investments a little bit less risky than they might seem on the surface. And did that surprise you? What surprised you about the conversation that you had with him? Yeah, I mean, his bullishness, I think you can kind of get a sense of it from the earnings calls. But, you know, he really, I think, sincerely believes that there's a huge arbitrage here, as, you know, people in finance would say that basically, this is a once in a moment opportunity to basically lend money to these huge tech companies at above market returns.

34:13And I don't think I quite appreciated that until I talked to Mark. And also, yeah, just all the ways that they're kind of de-risking the investments. I don't think I quite appreciated until we sat down. You know, that said, they are taking a bet on, you know, that some riskier companies like Crusoe, Oracle, CoreWeave, you know, that they're going to be around in over a decade to pay their bills and finish these projects. And if that doesn't happen, you know, they'll be left sort of trying to find somebody to fill their place. So, you know, it's balancing those risks. And I think it was interesting to hear how Mark thought about that.