SpaceX S-1, Anthropic Revenue Booms, OpenAI Cracks Erdős Problem | Diet TBPN

0:02John Coogan:We have a massive show. There's so much tech news. People said the technology industry, they were out of news. They weren't. There's plenty of tech news. They lied. They lied to you. All over the Wall Street Journal. AI companies are duking it out for prime placement in the journal. SpaceX got top billing in the journal with the IPO filing. SpaceX sets its IPO in motion. The SEC filing starts move to raise tens of billions of dollars in record debut. We've talked about this a lot on the show. Clock is officially ticking on SpaceX's huge stock offering. What are you laughing about? Ryan says, I knew Jordy couldn't do this two days in a row.

0:42Did I not do it two days in a row?

0:43John Coogan:Nope. Haven't done two days in a row in months. We can roll the tape. I mean, after reading that piece about us, I feel like some brand differentiation is actually to our benefit since some people can't tell us apart. And so I've become a fan of the split. Of casual Friday and Thursday? Yeah, of just a casual look over there, more buttoned down, buttoned up, buttoned left and right over here. Somebody's got to be buttoned up. Someone's got to do it. SpaceX on Wednesday revealed new details about its financials and how Chief Executive Officer Elon Musk will try to grow a sprawling enterprise dedicated to advancing cutting-edge technologies in space and back on Earth.

1:27John Coogan:The company disclosed the information in an investor prospectus. Publication of the document sets SpaceX on course to potentially raise$80 billion or more for a stock sale as soon as next month. The rumored date is what, June, July? Was it June 12th? June 12th? That's just 20 days away, basically. They're going to beat out Saudi Aramco that raised$26 billion when it went public in 2019. Musk has touted out-of-this-world objectives for the company, from deploying a huge number of artificial intelligence satellites in the future to colonizing Mars. Texas-based SpaceX has distinct businesses ranging from rocket launch to satellite operations to a nascent AI unit that is racing to catch up with rivals.

2:10John Coogan:founded by Musk nearly a quarter century ago. Yeah, it has been a long time. SpaceX revolutionized the commercial space industry. The company has grown from a startup with a handful of employees that almost went out of business to one of the world's most valuable private companies with over 22 ,000 workers as of March 31st. It controls technologies that competitors and even nation states haven't been able to fully match. SpaceX reported its revenue last year at$18.67 billion. and Dan Primack had a post saying that the business was smaller than he expected. He's going on CNBC today to talk about the IPO.

2:45John Coogan:Has had a couple interesting takes. People are going back and forth. Overall, the reception has been the S1 is an extremely enjoyable read. Kevin Kwok says it's the most enjoyable S1 read in a long time. Reads so easy like sci-fi or fiction. It's kind of the perfect post. Or just regular fiction. Because if you're pro-tech, you like space, you're excited about space, that could be a positive. Yes. But if you're a bear, you can say it reads almost like science fiction. Of course, the best TAM slide ever? Probably the best TAM slide ever. Sawyer Merritt has the screenshot here. SpaceX and IPO filing.

3:25John Coogan:We believe we have identified the largest actionable total addressable market in human history. We estimate that our quantifiable TAM is$28.5 trillion, consisting of$370 billion in space from space-enabled solutions, $1.6 trillion in connectivity across$870 billion in Starlink broadband and$740 billion in Starlink mobile, as well as additional opportunities in enterprise and government. 26.5 trillion in AI across 2.4 trillion in AI infrastructure, 760 billion in consumer subscriptions, 600 billion in digital advertising. That's massive. Well, is that for X? I don't know. The idea, so everything else is more believable.

4:12Everything else is more believable to me than X getting meaningful digital advertising penetration.

4:17John Coogan:Yeah. I guess the time matters here because a lot of these markets aren't this big currently, I think. I don't know. But I guess over time, you know, if you think about the next 25 years, the next 100 years, I don't know if these are inflation adjusted, but there's lots of things that could happen. For illustrative purposes of sizing our addressable market, SpaceX excluded China and Russia from global estimates. I feel like you might want to put in China and Russia over the next couple decades. Who knows? Maybe we become best buds with both countries. Anything can happen. World peace might come, and that's going to expand TAM.

4:55John Coogan:That's an economic incentive for world peace. I like to see it. There were some beautiful photos that were shared in the start of the S1. Lots of pictures to start, and then it gets very text-dense. But the photos were, I liked them. I thought that they were unique. I hadn't seen them that often, and they felt like they were kept in the back pocket for a while. and they, I don't know, they just like remind you of SpaceX. It's like a beautiful thing. Dan Primack says, incredible that Goldman beat out Morgan Stanley for the SpaceX IPO, lead left. Given that Michael Grimes returned to Morgan Stanley in part for this deal, of course, Morgan Stanley is on the deal.

5:29John Coogan:But that is, it is a big win for Goldman that DJ SpaceX is at the helm. Goldman Sachs and Co. LLC lead left in the joint book writing managers. But everyone's getting a piece of the SpaceX IPO at this size. Will Bitsky says, shout out to the Goldman analyst that was richly sacrificed to win this lead left IPO. It must have been an incredible amount of work. It's not just the biggest IPO of all time. It's not just this incredibly complex structure with multiple businesses. It's also, you're reporting to Elon Musk. Elon Musk is your client, and he's going to ask for things probably more aggressively than anyone who's a CEO of a company that's going public.

6:09John Coogan:So lots of winners from the SpaceX IPO. Luke Nosek is a huge one. He was at Founders Fund, co-founded Founders Fund with Peter Thiel. His next role will be leading. This is from a long time ago. He left to start Giga Fund, which was billed at the time as a new investment firm that initially will be focused on raising capital for Elon Musk's SpaceX, a Founders Fund portfolio company where Nosek is a director. And so David Kwan says, today I learned Luke Nosek left FF to start a fund exclusively focused on investing in SpaceX. There are a few of those that we're hearing about these days. Of course, exclusive does not mean 100 % of the capital went into SpaceX.

6:44John Coogan:It just means that they were very, very focused on that. Gigafund has a lot of different companies in the portfolio cover. We've had a bunch of founders on the show who have raised money from Gigafund. But SpaceX is where Luke is a director, deeply involved, and has focused on participating in many, many rounds. And so conviction will do that to an MF, says Pocket Jacks Capital. Lots of big winners. Frank asked Codex for SpaceX fair value based on the S1. Should be an interesting buildup to the IPO. What was the result from Codex? Sid lands around 1.1 to 1.5 billion. It's not bad. That's not bad.

7:19John Coogan:Bull case gets to 1.7 to 1.9 if investors assume anthropic sticks. AI infrastructure margins are strong. Starship unlocks major new markets. And public market scarcity drives demand. But$2 trillion means the market is effectively assigning something like$800 billion to$1 trillion into the AI orbital compute story on top of an already rich Starlink valuation, possible as an IPO mania print, but that's not what I'd call for fair. So we'll see what happens. I mean, the big news was the partnership with Anthropic where Anthropic is spending over a billion dollars a month, I think. It's ramping up - $15 billion a year.

7:54John Coogan:And that's huge for SpaceX, given that they did 18 and a half or something last year. This is a huge jump up in, I mean, they have to be one of the biggest neoclouds like overnight with this. Yeah, I was trying to find some of our conversations from last year where we were XAI and Grok was growing, but maybe not close to the rate that would require that much infrastructure. Yeah, finding product market fit on the actual distribution side. Obviously, we love X, but it's not the biggest platform. Yeah, it's certainly a shoot for the stars. And if you miss, you have a pretty great NeoCloud business, right?

8:36Anthropic has to pay way above traditional NeoCloud pricing for this compute. And so ends up being a great outcome for SpaceX.

8:47John Coogan:Peter Haig says, just reading the SpaceX SEC document, one thing that sticks out is the capital spend on AI is 3x that on space. It's an AI company with some rockets. which is a wild, wild pivot. It's the 11th hour. You know, this has been a rocket company for 20 years or 15 years, then an internet company with Starlink, but that was still so tied and so clear and so quick to get to like a logical link. Like you needed the launch capacity to build Starlink. And so you had this new capability, satellite internet, it was amazing. And it went from idea to launching the satellites to consumers actually using it when they're traveling, camping, off the grid, real, and then showing up in planes and all sorts of different applications.

9:37John Coogan:It became very, very relevant, very real, very quickly. And the Colossus XAI, that felt like a different company because it was, but it has just become so, so, so big so quickly. Yeah, and looking back at the plays Elon and his investors have made around this over the last year, right? There was that felt like somewhat of a coordinated effort beginning of this year and late last year when suddenly everyone started talking about space data centers very suddenly. Remember Gavin Baker started coming out talking about it. That's around the time when they sort of floated the December of last year. Floating the idea of like what the potential valuation would be.

10:15Started building that AI narrative. Started, you know, made a play for Cursor. You know, partnered with Anthropic, even though, you know, only a few months ago they were much more combative. Yeah. His name calling. So, yeah, he, you know, I think this is why Elon has been able to accumulate so much capital. It's like he is pretty much the best in the world at like making, just making plays. Yeah, making plays. And doing whatever it takes.

10:40John Coogan:Yeah. So the most recent play, unrelated to the news that made the front page of the Wall Street Journal, anthropic revenue surges set to post first profit. Sales seen reaching$10.9 billion in the second quarter, up 130 % over previous quarter. truly in the title of the, in the actual URL of the Wall Street Journal article, they call it mind-blowing growth about to propel Anthropic to its first profit. Absolutely fantastic execution. So Tom Brown, co-founder of Anthropic says, we're expanding our partnership with SpaceX and we'll be scaling up GB 200 capacity on Colossus 2 throughout June. Appreciate Elon Musk and the team helping us find good homes for the clods.

11:23John Coogan:Is Claude plural? I thought it was all one Claude. And the purpose of Anthropic was to build Claude. And Claude will eventually build Claude. But I guess you have multiple instances of Claude running on different servers on different GB200s. Anthropic's Q2 revenue is set to increase by over 200 % and will post an operating profit. The AI will never be profitable group is in absolute shambles right now. There have been a lot of folks who have been just doubting time and time again, will this ever make money? Will this ever make sense? And Dylan Patel sort of laid out on the Dwarkesh Patel show this idea that at a certain point, the leading models might actually be able to raise money or raise prices because they're driving so much economic value.

12:10John Coogan:Semi-analysis also put out a table showing for particular workflows that would take them, you know, $1 ,000 of human time that they would have to hire more people for, they were able to use AI and actually get an equivalent result for a tenth of the cost or a one hundredth of the cost or even, you know, a 30 % savings sometimes. Lisan Al-Gaib says, can someone check on Gary, February 23rd, 2026? He said, turns out Gen AI was a scam. I had to check the date on this because this seems like something he would have written in like 2024 and I would have been like, yeah, okay. Yeah. Maybe the usages are a little limited.

12:50John Coogan:Maybe where there is some sort of wall here, the data wall, or, you know, maybe we won't be able to, you know, maybe we'll need new, new paradigm, but to write this in 2026, when we're in like the fastest period of acceleration in terms of actual value from these models is, uh, pretty, pretty remarkable. I'm interested to see where he goes from this. Is he going to double down? Is he going to stick with this? It has been a couple months since February. I think the entire crypto boom and NFTs in particular just broke a lot of people's brains. Yeah. And VR, Metaverse too. Metaverse. It was another thing that was over-promised.

13:27Under-delivered. Metaverse, yeah, potentially even more.

13:30John Coogan:Yeah. Yeah. There was a lot of discussion about this will destroy Hollywood. This will destroy movies. Because with the Metaverse, there was never a moment where you could use a product and have a mind-blowing experience. Well, without paying for it. Like you had to buy. No, no, I'm just saying, I'm just saying like, period. No, no, no. The Apple Vision Pro demo, like there was a day you called me and you were like, why is everyone losing their mind on the timeline over the Apple Vision Pro? Do I need to buy one of these? And I was like, eh, it's kind of like a previous, not like perfectly there.

13:58John Coogan:I like it, but it's still the issue. I don't even think I, I don't think you're remembering correctly because there was no moment where I wanted to buy one. No, no, no. You didn't want to, but you recognize that it was the current thing when the Apple Vision Pro launched. For like that week, when everyone got them delivered and they tried them, there were a lot of people that were saying, They had Vision Pro psychosis. Vision Pro psychosis. A lot of people had NFT psychosis, all sorts of psychosis. We'll see how the AI psychosis develops. It goes both ways. Yeah, but anyways, comparing it, anyone can have a pretty wild experience with AI on a ton of different services.

14:36You could never do that with the metaverse.

14:39John Coogan:So Lisan Al-Ghaib is contrasting Gary Marcus's sub-stack post with what's happened in the AI industry. Anthropic valuation up 173 % since the start of the year. Posting profits in Q2, according to the Wall Street Journal. OpenAI valuation up 67 % since the start of the year. And OpenAI general purpose model solves longstanding and well-known Airdos problem. I believe it's Airdos. Airdos. Airdos problem without a scaffold. And so there was a lot of questions about how hard is it to solve these problems. But fortunately, we have Tyler Cosgrove, who's going to take us through what actually happened with this solution to this math problem that people are very excited about.

15:25John Coogan:Noam Brown said, today we are sharing that a general purpose internal OpenAI model achieved a breakthrough on one of the best known combinatorial geometry problems. Less than one year ago, Frontier AI models were at IMO gold level performance. I expect this pace of progress to continue. And Siddhar Ramesh, I don't know if he was joking about this bet, but he says, I have lost my$30 ,000 bet that AI would never solve the planar unit. Yes, I believe that was a joke. That's a narrative. That's a joke. Yeah, I think if you go into the comments. He jokes around a lot, but no. But a lot of people were surprised and a lot of people were excited about this.

16:01John Coogan:So take us through what actually happened. Okay, yeah. So I can basically go through like a simple explanation of what the problem actually is. Okay. So just for some context, Paul Erdos, kind of this legendary mathematician, throughout the 20th century, he basically proposes, I think the number is like a little over 1 ,200 different like little problems. These are the Erdos problems. People talk a lot about these as like goals for AI to solve. And you've heard like over time, there's been kind of like small iterative kind of solutions to a lot of these problems. Yeah. Sort of like collaborative, a mathematician working alongside an AI model.

16:30John Coogan:And sometimes... Or an easy one, just getting... There's like a main kind of place where all of the solutions go. So sometimes people will find, like AI will like find a different paper that wasn't actually put on the website. And then they're like, oh, AI solved it. But it's honestly true. But this is kind of the first time we've really seen kind of a big step change. Like, this is actually a new solution. This is using, like, you know, kind of novel ideas here. It doesn't exist in papers out there already. So this was problem number 90. So I can kind of read the question, then I can explain what it means.

17:01So it's, does every set of n distinct points in the real plane contain at most n to the 1 plus 0 of 1 over log log n many pairs which are 1 apart? Okay, so, like, what does that mean? Basically, we have the real plane, 2D, and we have a bunch of points on it. What is basically, how many pairs of those will be basically one unit apart? And what's the max number? How do we basically organize those points such that we have the max number of them? It's not just a grid? No. So you would think that, but I can basically explain why. So let's kind of formalize this better. So we have U of N, and this is basically the largest number of unit distance pairs among N points in the plane.

17:41Okay, so basically, we're thinking about how do we solve this. Naively, it's like, okay, what if we just take all the points, we have n points, and we just put them in a line, and unit distance apart, right? So it looks something like this. So for this example, we have four points, but it doesn't matter, it's just n. So one, two, three, four, how many pairs are there? There's three, right? Okay, yeah. So basically, this scales with n minus one. So you could have a billion points, and there's nine, n, n, n, whatever, n minus one. Okay, so now if we put it in a grid, what happens, right? So if we have a square grid here, there's nine points here, and then how many pairs are there?

18:18I believe there's 12. And basically as this number scales up, it's still linear, so it's 2n. Basically, if you do a billion points, it's two billion pairs. Two billion pairs. Yeah.

18:29John Coogan:The pair is specifically that line. Correct. Yes, because it's one unit distance, right? Because diagonal doesn't count. And you specifically don't, diagonal doesn't count because it's not one unit length. Yeah, it's not a unit distance length. Got it. So then, okay, what's the next thing we can do? The next kind of configuration is what's called the lattice construction. Okay. And so if we can pull up a picture of it, it's this kind of crazy-looking grid that has all these super, like, intricate lines in between. You can see it on the – this is from the OpenAI blog. If we can pull it up here.

18:59John Coogan:Oh, I think I saw this. So this is what it looks like. Okay. So if you can, like, zoom in on any of these points, you see that, you know, it looks like a grid, right? Yeah. There's not just kind of pairs at the edges, right? There's like way more. Okay. So this scales at n to the 1 plus 0, 1 over log log n, right? This is basically the best kind of example that we know works. Yes, but not a proof. We know we can find this, but is this the upper bound? We don't know. So this is basically the lower bound. So then the question is like, we have the lower bound, which is the best one we found. This is the most number of pairs.

19:39And then we theorized that the high bound, upper bound, scales with n to the 4 thirds. And then so Erdos, the original conjecture that he thought that the upper bound is still going to be less than n to the 1 plus O of 1. So this means O of 1, it's like as n scales to infinity, right? So O of 1 basically scales to 0. It goes to 0 as n goes to infinity. Basically, OpenAI figured out that this is not true and that there actually are some n's for which this kind of max number of pairs is greater than the original Airdosh conjecture. So for infinitely many n, this is not for every single n, right?

20:18So it's not like five points or whatever, but there are infinitely many n's for which this is true. Okay. That it's greater than n to the one plus some constant. Interesting. Okay, so that was basically the big thing, right? This is like, you know, decades old problem, right? This is an incredible thing. Terence Tao is like, wow, this is incredible. But yeah, that's basically the overview of the problem. But yeah, it's very exciting because this is not like a math model. This is just an internal model, general reasoning. You could say it's like generally intelligent. Yes, I think you could say that.

20:55And then I think it's interesting because from like public perception, And it seems like this didn't take that many tokens. This was not millions of dollars of inference time. It was maybe something like hundreds to thousands of dollars of inference compute spend. That's very interesting because we were talking

21:14John Coogan:about Gwern's conjecture about novel ideas coming from just brute forcing different connections between things. And this is more token efficient. Yeah, this is not just taking some solution to a different Erdos problem and just spamming it on all 1 ,200 of the problems. and oh, one of them works. This is like kind of a new novel idea. Like maybe this solution is, the way that they found this, it's like super complex. You can read the proof. It's like 18 pages long. I don't really know what it means. But like there's a lot of the mathematicians are saying, okay, this actually could be useful to a lot of other problems.