ROUNDUP: Anthropic Pays Elon $45B, SpaceX Files $1.75T IPO, Claude Mythos Hacks Apple

Three of the biggest deals in AI history all hit in the last 48 hours. SpaceX filed for a $1.75 trillion IPO, the biggest ever. Anthropic signed a deal to pay Elon $1.25 billion a month to train Claude on his GPUs, $45 billion total through 2029. And Anthropic is now projected to be profitable by next month, beating OpenAI to it.
We break down the S1 filing, the 30% retail float, the insider unlock structure, and why Antonio Gracias is about to become absurdly wealthy. Then we get into Cursor's $60 billion acquisition by SpaceX, why Michael Truell is about to be the youngest billionaire alive, and how Composer 2.5 (built on an open-source Kimi base) is beating Opus 4.7 and GPT 5.5 at a fraction of the cost.
Plus: Claude Mythos cracking Apple's M5 chip in five days for $35,000, China's GPU race, OpenAI's rumored IPO filing, and Karpathy joining Anthropic.

------
🌌 LIMITLESS HQ ⬇️

NEWSLETTER:    https://limitlessft.substack.com/
FOLLOW ON X:   https://x.com/LimitlessFT
SPOTIFY:             https://open.spotify.com/show/5oV29YUL8AzzwXkxEXlRMQ
APPLE:                 https://podcasts.apple.com/us/podcast/limitless-podcast/id1813210890
RSS FEED:           https://limitlessft.substack.com/

------
TIMESTAMPS

00:00 SpaceX Partners With Anthropic
06:43 Anthropic's Shocking News
08:55 SpaceX IPO Filing Here!
14:12 The New Youngest Billionaire
17:48 Claude Mythos Hacked Apple
21:30 China Summit Failure?
24:34 OpenAI IPO FIling This Week??
25:25 Andrej Joins Anthropic

------
RESOURCES

Josh: https://x.com/JoshKale

Ejaaz: https://x.com/cryptopunk7213

------
Not financial or tax advice. See our investment disclosures here:
https://www.bankless.com/disclosures⁠

Creators and Guests

Host

Ejaaz Ahamadeen

Host

Josh Kale

What is Limitless: An AI Podcast?

Exploring the frontiers of Technology and AI

Ejaaz:
[0:00] In the last 48 hours, three of the biggest deals in the history of AI just happened. SpaceX officially filed for the biggest IPO. Anthropic signed an unexpected deal to pay Elon Musk $1.25 billion per month to use his GPUs to train Claude, two of the biggest rivals in AI teaming up. And finally, Anthropic is now projected to be profitable by the end of next month, the first AI lab to do so, beating OpenAI. Now, what most people don't realize is that these three stories are very closely interlinked. SpaceX is eyeing an eye-watering $2 trillion IPO as soon as next month. They need to book billions of dollars of revenue to look good in front of investors. Ananthropic, on the other hand, needs all the compute they can get their hands on to beat OpenAI. Now, the common thread between these two rivals is Elon Musk and Dario Modey share a mutual disdain for their enemy, OpenAI and Sam Altman. And this unlikely partnership gives them a shot at taking the number one spot away from them. We're going to get into all the economics of the deal and the S1 filings in

Ejaaz:
[0:56] today's episode and much more.

Josh:
[0:58] This is a pretty huge deal and we were aware of the Colossus 1 partnership between SpaceX and Anthropic. We knew that they were going to take the entirety of Colossus 1. What we learned here through the actual S1 filing is that the deal extends even further than that and this is the recent development. Within the S1 filing was this tiny little paragraph that said not only are they taking Colossus 1 but they're also going to lease part of Colossus 2 like you mentioned for 1.25 billion dollars all the way through May 2029 Where if you do the math on that, that's a $45 billion commitment to SpaceX from Anthropic just prior to SpaceX going public. So this is huge for the SpaceX revenue. Anthropic is rumored to actually be profitable soon. We're going to get into that a little bit later. But the S1 filing was really fascinating. When you think about what Anthropic needs versus what it gets, right? They're competing directly with OpenAI. The constraint is compute.

Josh:
[1:48] Colossus 1 has 220,000 NVIDIA GPUs. And this is a mix of a few types, H100s, H200s, all pretty cutting edge GPUs. And that delivers 300 megawatts of power. The fact that they now have access to Colossus 2 on top of that is a huge competitive advantage and one that I don't think can be overstated. And it seems like it makes sense for both people, right? Like Elon's going on a roadshow. He's getting ready to take spacex public he wants the market to know that spacex is an incredibly strong company they're delivering revenue and they just booked 45 billion of it over the next couple of years anthropic needs this because not only does it create a lot of compute usage that's available for them but it creates this really powerful alliance that it's like there's the funny saying that the enemy of my enemy is my friend well if anthropic and spacex are both competing against open ai why not work together and that's exactly what they're doing

Josh:
[2:38] here with this deal it's a pretty big deal.

Ejaaz:
[2:39] It's funny. I watched a recent interview with Anthropic CFO with Krishna Rao, and he was asked, like, how do you spend most of your time, like, presuming that the answer was going to be all on the finances of the upcoming IPO that Anthropic's going to do? And the answer with one line, which was, I spend 70% of my time on compute, how to acquire more compute, how to get more compute, how to distribute all more compute. And he was the guy that was behind this deal. And the reason is quite obvious. Anthropic needs all the compute that they can and get their hands on to train an inference Claude. Now, the deal that we spoke about last week was on Colossus One, as you mentioned. That was purely for inference. So Anthropoc gets a lot of requests for people who want to use Claude, and they need to file those requests and serve those requests. That is what Colossus One deal was for. Now, these new GPUs in Colossus.

Ejaaz:
[3:24] Two specifically, they're all H100s and H200s, as well as a bunch of like cutting edge GB200s and 300s. These are like the bleeding edge NVIDIA stack, but it's homogenous, which means that the entire stack is perfected for training specifically. And that is something that Claude needs to get more compute for. Like there's two different architecture styles and this specific deal that was signed yesterday for $1.25 billion or $15 billion a year ends up being really

Ejaaz:
[3:51] advantageous to training the next version of Claw, so Mythos or whatever comes after that. Now, some important nuances here. If you look at this excerpt from the S1 filing, you'll notice that right after, there is a termination clause that either party could sign within 90 days' notice. So really, the guaranteed amount of money that SpaceX is making for this at this particular moment is around $3.75 billion, and each party has the right to waive access. So if Elon Musk decides, I want to use this compute instead to train Grok, he's able to kind of sign a waiver or whatever, and let that go. And if Anthropic decides, I've actually spun up my own data centers, I don't need access to this anymore, they can also do that.

Ejaaz:
[4:28] So there's like, it's kind of like a, we're committing, but we're not really committing.

Ejaaz:
[4:32] And the other side of this is, this is the second major deal, or rather third major deal that Elon Musk has signed to become an AI compute provider. And he even has a tweet here where he confirms this. He says, we're in discussions with other companies to do the same. We've recently expanded our offering to include AI compute as a service at significant scale and he's doing this for a very specific reason which is like If you're going to value a company at $2 trillion for an IPO as soon as like next month, you need to have revenue on the books. Now, with this deal, they're getting paid $15 billion a year from Anthropic alone. That is 80% of SpaceX's entire revenue from last year, from 2025. So it looks pretty good from an investor perspective that this will be the case. But he also did a sneaky thing as well, Josh, that we mentioned last week, which is he got Anthropic to confirm that they're exploring AI data centers in space, which has been SpaceX's entire thesis around a $2 trillion valuation for their IPO. So if Anthropic, the biggest AI lab in the entire world, is committing to using some of these AI space data centers, it not only confirms that it's a feasible thing and not just like a fantasy kind of like goal, but two, that they're going to have one of the biggest customers in the entire world buying those loads. and that's good for the idea.

Josh:
[5:45] I love the idea of SpaceX being the compute provider. Like when we think about what each company is best at, it is very clear that SpaceX, XAI and X, that whole bundle is best at building hardware. It's best at actually deploying GPUs, creating difficult things at scale. And that's a really unique skill set that Elon has in particular that other companies don't. And I think that's the difference maker. So I hope that this remains true. But part of me thinks that it won't always be the case Because when SpaceX is competing against these other companies, which they are clearly doing with the acquisition of Cursor, which we'll talk about in a little bit, but also just the kind of forward strides that they're making with XAI and their $10 trillion or $10 trillion parameter model, there's going to be a lot of kind of competition. And I wonder if that three-month option is more an option for Elon to pull it

Josh:
[6:34] back than it is for Anthropik to stop spending. And it seems like that's the case. Now, we also have Anthropic updates, which are interesting because Anthropic now is projected to make

Ejaaz:
[6:45] $10.9.

Josh:
[6:46] Billion in the second quarter of this year, which is double the estimates. It seems like every time we talk about Anthropic, the market cap is growing, the ARR is doubling. But more importantly, they're projected to make a profit of $550 million. Now, this is shocking because none of these hyperscaling AI companies are projected to make a profit outside of the large established ones like Google and Microsoft and those companies. But for Anthropic to say that they're making a profit seems, one, amazing, but two, a little interesting because, Ijez, you mentioned to me before we were recording, well, if you're making $550 million in profit, why wouldn't you just spend it on more compute?

Ejaaz:
[7:21] So my take on this is they may indeed turn a profit by the end of next month, and they'll be the first AI lab to do so. But by the end of the year, they're still going to be net negative. They're not going to turn a profit. If I had to guess, so Dario's estimate is they're going to become sustainably profitable in 2028. I think that gets moved up to 20 sometime in 2027. But this year, no way. It's too much of an opportunity cost to not spend the money on compute and make sure you have a convincing lead over every other single AI lab. Now, it's important to set some context here. You see $10.9 billion in three months, and you look at Anthropic Success, and you're like, oh, yeah, that makes sense. At the start of the year, January 1st, Dario gave his projection of what the ARR expectations were going to be for Anthropic, and it was $10 billion by the end of the year. And now he's making more than that in three months. So it's just insane growth. They're on track for $100 billion ARR by the end of the year. Someone projected this out, by the way. If they just stay at the consistent rate that they are at now, increasing ARR, by the end of next year, so by the end of 2027, they'll be generating more ARR than Google themselves. So a lot of people are questioning the valuations of the round that they're raising right now. If you're earning more money than one of the biggest Mag7 companies ever, this kind of justifies it pretty significantly. So I'm excited to see whether this actually ends up being the case. Anthropic's been on a generational one right now. But I do think they're going to be net negative at the end of the year. They're going to spend it on compute.

Josh:
[8:49] All right, well, we'll see. One thing that we do have a little bit more certainty on is the SpaceX numbers, because the S1 is officially here. We have the SpaceX S1. It's a 308-page document.

Ejaaz:
[9:00] And we read every single page, don't we, Josh?

Josh:
[9:03] Summarized by our favorite LLM, because that is a lot of reading to do overnight. But we have the key highlights. We have the things that are important. So I think one of the most important things for people who are listening to this today retail investors, people who want access to the SpaceX S1

Josh:
[9:19] We bring you good news. The traditional IPO offers about 10% of their total share float to the public. SpaceX is tripling that. They're offering 30% of all SpaceX stock to public markets. How is this going to be distributed? Through your local brokerage account. If you have a brokerage with Robinhood or TD Ameritrade or whatever it is, they will have shares available for retail. And I think that's the most important thing that you could take away from this is that you'll have the ability to actually participate thanks to this very generous offering of 30% of the share float. Now, in terms of valuations... 1.75 trillion is the number. That's what it's going to start at. And they're looking to raise $75 billion. This creates basically the largest IPO in history by like a pretty large margin, maybe a factor of two. The largest before this was Saudi Ramco. So it's a huge raise in money in terms of actual capital they're going to bring in, in terms of investors that they're going to bring in. And then there's also this interesting detail of the selling plan, which I understand is different than traditional. So there are a lot of insiders at SpaceX, a lot of people who have become very, very wealthy off of SpaceX stock, but they're not all going to be able to sell it on day one. There are some interesting parameters here that I know each as you were looking at earlier, maybe you could cover on like, are we going to get dumped on as

Josh:
[10:34] retail investors or are there some safeguards against this?

Ejaaz:
[10:37] Okay, so typically the way that this works is you have a major IPO and a bunch of the early investors are subject to the same cliff terms, which is a four a cliff. So typically, you wait six months to a year until a portion of your shares that you acquired through investment becomes unlocked and you're able to sell them. Now, SpaceX is pursuing a very unorthodox approach where a bunch of investors, and they haven't specified who, but it's definitely not Elon, will get a portion of shares unlocked after the first 30 days. So after the first month. Now, if by the end of the next month after that, so two months later, if the SpaceX IPO price is above 30% of its original IPO price, so if it's at $100, it's above $130, they have the eligibility to execute and sell another percentage of their shares, and so on and so forth. And then there's an unlock series over a series of like 40 consequential days. So every 40 days, they're able to unlock more and more and more. Now.

Ejaaz:
[11:40] Big question I had is, why are you doing this? It seems like, you know, you have a bunch of investors that are making like billions and billions of dollars from this, and so like you're giving them an easier exit plan, and maybe there's some kind of reasoning behind that, I'm not quite sure. And then the other side of this is, why is it happening so soon? Like typically, like you don't get to see like big cliffs only until after like six months. This will like kind of like stagger it, which means that like you won't have like a sudden like windfall and like a dump in the stock price, and maybe that's the ultimate reason why they're doing it, But also like, why are you giving it so soon is the question that I have. And I was digging into this, Josh, and I realized that there was a bunch of concentrated investors that have made quite a significant back from this. And one of them was this guy called Antonio Garcia, I believe. And he basically invested in SpaceX through, I think it was 18 different vehicles, which gives him a total aggregate market ownership of 7.3% of SpaceX. So if you can imagine, I can't even do the math on this, but if it values up like a $2 to $5 trillion valuation over the next year, that is a lot of money and you probably do want to exit some of that. It's crazy.

Josh:
[12:50] Yeah, I'm sure. Well, also, I don't know if that's true. It's funny. For anyone who's read Elon's biography, the Walter Isaacson one, Antonio Gracias is his name, actually. He's a very core story or core person to the SpaceX story because he was there from the very early days. He was there for all the times that Elon was dealing with issues. He's part of one story that I loved and I remember that over the holiday break they needed to move compute so they actually hopped in a private jet they went to the data center and they started ripping out the compute for Twitter when they first bought it so there's Antonio is this like core person who is kind of pivotal in a lot of Elon's companies. He's kind of his right-hand man. He's part of the crack team of people that gets deployed if there's ever a serious problem that needs fixing. And he now has 7.3%. From what I understand, a lot of the people who own a lot of this equity will not really be inclined to sell. I can't, I mean, Antonio's a very wealthy guy. He is very loyal to Elon. And I can't suspect he would be selling a lot of it. It's a really funny story, but it is going to make a tremendous amount of people rich. When you think about any of the employees, how many millionaires are going to be minted from this? Deca millionaires are going to be minted from this? It may be higher than any IPO in history. So the amount of wealth that we're going to see kind of showing itself in public markets is going to be pretty large. And there's one person in particular that may be the biggest winner of this. And it's a 25-year-old kid,

Ejaaz:
[14:10] Which is really insane.

Josh:
[14:12] For those people that aren't familiar, there's this guy named Michael Truel, I believe is how you pronounce his name. and he's the founder of Cursor. Now, Cursor, as we know from a few weeks ago, was optionally acquired by SpaceX, meaning they invested a certain amount of money into the company with the option to purchase Cursor for a set amount. That set amount is $60 billion. And it seems now as if we've received pretty much close to confirmation that this deal is going to go through. So the CEO of a $60 billion company, who's 25 years old, is about to get a pretty sweet payday. And more importantly, SpaceX and the XAI team is going to acquire a pretty badass harness for the AI. And I think this is where a lot of that value comes from For those not familiar with harnesses, it's kind of like you can imagine it, the surrounding infrastructure around an LLM that includes memory, operating systems, and it turns a raw model like ChatGPT into something like Codex that's much larger and much more powerful. So that's kind of what they've been planning on. And it seems like Cursor themselves as a company has been doing pretty freaking great recently.

Ejaaz:
[15:14] It is my favorite comeback story ever. Cursor, for the longest time, two years exactly, were criticized as the most fallible business ever. They were called an AI wrapper, a wrapper around all the Frontier LLMs, and they're going to get replaced as soon as Claude creates like a better harness. What ended up happening is Cursor's moat, which is like agent orchestration, like making sure all your prompts make sense, routing the right prompts in the correct way, is all kind of infrastructure or IDE that it's kind of like muscle memory that surrounds your AI model, is the moat itself. And Sam Altman, in fact, said about two weeks ago that the harness and the AI model, the foundational model itself, are one and the same. And they pay effectively the same amount of time and effort to training and building both. Now, what Elon Musk is doing is very smart. He has signed a deal, officiated yesterday or whenever, that says, within 30 days of the SpaceX IPO, I will acquire Cursor for $60 billion. That's like a document that is being signed. And so, like you said, that makes Michael Troll probably the youngest billionaire ever. But the point is.

Ejaaz:
[16:19] He now gets to acquire that harness and use it on Grok to make it a frontier AI coding model. So I say this in maybe this tweet or like some of my other tweets, that Cursor is effectively becoming a frontier AI lab would be one of the few AI labs that can take a shot at joining Anthropic and OpenAI at the table as a number one product. And I have the evidence to prove it, actually. If you look at the recent benchmarks on Cursor, which fairly rates all the top models, because everyone uses the top models on their platform, you'll notice that Composer 2.5, which is a model that they released this week, that, by the way, was trained on a old open-source model, Kimike 2.5, not even the latest, not even 2.6. An open-source model using their harness ended up being not only equivalent to the quality of Opus 4.7 and GPT 5.5 extra high, but also a fraction of the cost, eight times cheaper than OpenAI, 12 times cheaper than Anthropic. And if you look down here, sorry, I don't mean to dunk on you, but we did an episode on them yesterday. Gemini 3.5 flash from Google. Billions and billions of dollars on this. Ended up ranking not only more expensive, but much, much less. So Cursor's having an absolute win, and I'm proud of them, to be honest. It's very cool. Hey, you know what.

Josh:
[17:35] Model's not on this list that would top it?

Ejaaz:
[17:38] Claude Mythos.

Josh:
[17:39] Oh, yes. Remember that one? The one that got released to the private only? Not too dangerous to be released to the public. Well, we have some updates on Claude Mythos because people have been getting active. And they've been making some pretty serious progress as it relates to cybersecurity and exploiting, this time in the case of Apple Computer. Now, I love Apple. You love Apple. We're all using MacBooks. M5 is the most recent cutting-edge chip, one of the best chips in the world for a laptop right now. I don't think money can buy anything more powerful. Claude Mythos hacked it. Claude Mythos was able to exploit Apple's brand new silicon chip, and it only took five days. And if I'm remembering this correctly, it took only like $35,000. It wasn't a tremendous amount of money. And for reference, if you are someone who has a zero-day exploit, this is an exploit that hasn't been discovered by Apple. And if you have something like that for an iPhone, those exploits sell for tens of millions of dollars. These are the crown jewels in the dark gray markets for hacking people because Apple devices are so abundant and so readily accessible for people who are generally people who you would want to attack. So these exploits sell for a tremendous amount of money. The fact that a research firm was able to generate a brand new zero day in five days on the newest hardware for $35,000 seems like a pretty big deal. And, you know, originally we were talking about mythos and how it was kind of a bummer that they didn't release it. and we kind of assumed that it was because they were just compute constrained

Ejaaz:
[19:05] While that might be true.

Josh:
[19:06] It also seems that Mythos actually is as powerful as they say it was.

Ejaaz:
[19:10] Yeah, so it's actually not true about the compute constraint. And it came confirmed from the head of their Frontier Red team, their security team, which basically battle tests a lot of their new models. He confirmed, he said, the reason why we haven't released Mythos is because it walks the walk. It is very dangerous and could be used as a threat. And so we're giving it to anyone and everyone to defend their security systems. Now, Apple, very famously, has never had kind of an operating system that has been largely hacked or exploited in that way. And MIE is the defense system which we're referencing here that Claude Mythos found an exploit for. I want to get into a few details about this because it's a very unique story. So basically, Apple spent $2 billion collectively over the last five years building this defense system. And the irony is Mythos didn't actually exploit this system itself. It found a completely new attack vector which circumvented the entire thing so it found a new attack vector that humans themselves hadn't thought of it was a version of like poisoning the data that went through the memory which meant that it kind of like opened up a gap for them to be able to kind of like pull specific kernels it gets very technically complex but the point is it took five days three people on a team and thirty five thousand dollars worth of api costs Now, it was such an important exploit that the team printed the exploit onto 55 pages and walked it to the Apple HQ in San Francisco. That's so crazy.

Ejaaz:
[20:36] Because they couldn't risk putting it on a thumb drive or an email attachment and emailing it to them. That's how important it was. We're going analog just to kind of deliver this latest thing. So the point I want to make here is that the narrative is kind of flip-flop between Mythos. It's like, is it good? Is it not good? GPT 5.5 seems to be as good at coding and can do the same things. The answer is simply no. There's a reason why the government has put a stop on it. There's a reason why a lot of these companies and partners have aired some of their case studies and said, listen, this thing has found vulnerabilities that no other model has ever found. And so it walks the walk. And I just want to kind of point out that like we are probably close to AGI than we give credit for.

Ejaaz:
[21:11] It's just appearing in different vectors and cybersecurity is the first one for it to do so.

Josh:
[21:14] Okay, so we're going to speed run the remaining topics of this, starting with the China summit. Now, IJS, I know you've been following the Chinese summit a lot. There's a lot of updates that we have. If you'll remember last week, Elon, Jensen and Donald Trump were all on Air Force One flying together, along with a bunch of prominent business leaders. Basically the leaders in every major economic sector in the United States, they all converged into China to meet with Xi Jinping and the Chinese Communist Party to see what types of deals they could work out broadly throughout the economy, but also specifically with AI. And just this week, we have some updates on how that went.

Ejaaz:
[21:47] Yes. Okay. So this is a bit of a heartbreaking story, but also it should be a concerning one, especially for the rivalry between China and the US. So one of the biggest pointers on the agenda when America visited was, let's get China approved to buy American GPUs. Now, previously, America banned China from buying these GPUs purely because they didn't want China to get ahead of the AI race. And so they've banned them for roughly like the last, I want to say six months.

Ejaaz:
[22:18] Now, Jensen Huang of NVIDIA was hurting from this because China accounted for roughly 20% of profit, which immediately got wiped out when Trump banned the entire thing. So he went on this trip with Trump. He was going to like negotiate and figure this out. And let me tell you, the news was really good. Breaking news. We have like unlocked the ability to sell GPUs to China, except worry hit almost immediately, which was Trump said, we've approved it, but China hasn't officially bought any of the NVIDIA chips that we're offering to sell them. And then just yesterday, Jetson Huang confirmed that NVIDIA has largely conceded China's AI chip market to Huawei, which is a Chinese company that since those six months that we've banned them, have created their own bleeding edge GPUs, which have now replaced the need for NVIDIA GPUs. And the China government specifically put in a mandate, a doctrine, which said any super intelligent Chinese AI lab cannot use or operate on NVIDIA's GPU stack. They must use Chinese GPUs. And that has been a forcing function to get China up to date. Now, the reason why this is very harrowing and worrying is.

Ejaaz:
[23:29] Previously, America was ahead of the race for GPU infrastructure and having frontier models because they could train on their frontier GPU infrastructure. America still has the lead here, but China has caught up aggressively, whereas the gap was pretty wide. It's now very, very close, and China has the scale. They have all the lithium and substrate resources to build all this. They're rebuilding their own ASML, and they have all the energy in the world, in fact, three times more than America, to power all these GPUs. So it's a concerning position to be in, especially since America is still catching up on energy and building data centers and all the stuff that China already has. So that's why it's a bit of a sad story, to be honest.

Josh:
[24:08] Yeah, it's tough. It's going to be interesting to see the further second order effects that come from this.

Josh:
[24:12] I hope they can figure it out. We have two more updates remaining. The first is an IPO that might surprise people on the timing. OpenAI may have an IPO filed as early as Friday when you're listening to this, actually. There may be an S1 for OpenAI, which is surprising because it feels like for SpaceX, we've kind of known it was coming for the last year or two. OpenAI, this is pretty soon for them. And you have to imagine there is this rush to get to the starting line because there is a tremendous amount of money that these IPOs are going to absorb. And if you get there sooner, odds are you will have more of an opportunity to absorb that money. OpenAI is clearly taking advantage of this and they are seemingly going to try to go public very soon. If we get an S1 filing on Friday, that'd be pretty crazy. But what's exciting from this is we'll see a lot of the internals, we'll see and have a much better understanding of how OpenAI works, what the priorities are, what the internal deal structures are. That's going to be really exciting. And then the final bit of news is on the other lab, the other main one that we talk about all the time, Anthropik.

Josh:
[25:09] You'll note that this post has 26.1 million views. That's like a tremendous amount of views for a post on someone announcing their new place of employment. So for those who aren't familiar, Andre Carpathy, he is one of the godfathers of AI,

Ejaaz:
[25:26] Co-founder of OpenAI.

Josh:
[25:27] He basically led the full self-driving program at Tesla for many years and then took a break to work on education, to work on his own little startup, but is now announcing the return and he is coming back to Anthropik. And from what I understand, he will be working at the R&D department, working on how to get Claude to make a better version of Claude and tighten that loop and improve things. So when I saw this news, I got very excited. We see the first supply is from Boris Churney. He's the guy who created Claude Code. This seems like a huge win for Anthropic, and I can't see it any other way. Because last week, I mean, when you think about last week, Elon gave them his pledge of support by saying, okay, Anthropic's actually doing this for good. Anthropic actually wants to do the right thing. And now this week, we have Andre Karpathy joining. This is like a huge couple of weeks for Anthropic.

Ejaaz:
[26:12] There's a lot of crap that is spouted in the AI rivalry and the entire sector. One of the best litmus tests is what Andre Karpathy says about your company. And there is no greater approval than Andrei Karpathy ditching what he's building to go work at your company. He's working in the R&D department, as you said, specifically on reinforcement learning. Why I'm so bullish about this is he already built and made a prototype of this, the autonomous research project that he had, where he had a bunch of models basically kind of build a better version of itself whilst he slept. And the discovery that he had through running a few loops of this was the autonomous AI agent had discovered new ways of training itself that Andre Carpathy himself couldn't have thought of. And he's been in this game for over a decade at the top of his field. So I'm very bullish for Anthropic in general. They've had an amazing week with signing this new compute deal, potentially turning profitable by the end of next month and hiring one of the best guys in the entire world, my opinion only. It's been a great week for them. And obviously it's going to be a very big month in terms of like IPO filings and potentially a SpaceX IPO next month. So that is all the news. For this week, three of the biggest deals signed this week alone, and it's going to come into effect over the next couple of months. I can't get my head around the amount of capital that is going to be absorbed into this market that is going to come online through these weird unlocks from SpaceX. It is going to be a crazy time in the market. We probably need to do an investment episode at some point.

Josh:
[27:42] Josh. Excitement is guaranteed no matter what. There is going to be a lot of volatility, a lot of news, a lot of chaos. And I think this line from Andre's post, sums it up the best where he just says, I think the next few years at the frontier of LLMs will be especially formative. And that's years. That's not months. That's not weeks. This is going to play out over a long period of time. There's going to be a lot of volatility along the way, but the rate of acceleration just seems to continue to speed up. And that's what he's hellbent on doing. That's what all of the news is kind of focused on doing is just increasing the rate of acceleration, increasing the rate of intelligence, the amount of tokens we're able to generate. And it's all really exciting. So that is a week. If you've made it this far. Congratulations. You're caught up on all of the news. Thank you so much for watching. If you enjoyed the episode, don't