NEWSMAX

I put a ton of money in stablecoins. I haven't lost anything. I mean, I haven't made a lot. But they've done really well compared to Bitcoin. They've done really well compared to Bitcoin. Outperformed. But you have a bunch of startup players, you have big institutions. It feels like I'm wondering where there's white space, where you think there still could be white space. Given that we know there's so much growth that's going to happen, that part feels inevitable. But where does the value actually flow and what are the opportunities for startups? So I think everybody at this point is bullish on stable coins. Right. So Secretary Besson has said he expects stablecoins to hit 3 trillion by the end of the decade. Right now they're about 300 billion. So that's like a 10x in the span of, call it four to five years now. I think that's pretty optimistic. I don't know that we're going to hit that kind of growth rate, but the numbers are clearly growing and they're growing every single year, like 60, 70% year over year. Now how do you make money on that? So one way is to own a stablecoin issuer. I think that's pretty tough at this point. Right. You can see it's basically a duopoly between Circle and Tether. Circle is already public, Tether is private. But they're supposedly raising around at 500 billion. There's not a lot of meat on the bone if you're investing at 500 billion into a stablecoin issuer. And the economics are really good at being a stablecoin issuer, but it's basically being a bank. Right? And it's the modern financial institution, except they have 100% net interest margin, which is. Which is great. Are there going to be new stablecoin issuers who show up now? If there are, they're probably going to be G SIPs, right? They're not going to be some new startup that comes up with a new way to launch the same product, which is effectively just a narrow bank. I think the place where you want to be investing as a VC is more in the interstitial layers, which is how. How is this stuff going to move around the world and get into the hands of a user, wherever they are? So one of the investments that we've made recently is a company called Rain, which basically they do these stablecoin cards, right? And they allow you. Let's say you're in some country where you, you're using some fintech app that's allowing you to get access to stable coins. Let's say you're in Argentina, you're in, you know, you're in Nigeria, you're somewhere where you really want stablecoins, you want dollars. You don't even know that they're stable coins. A lot of these apps, the users don't even know that there's crypto underneath the rails. They just know that, you know, they send the Naira, they get USD and they can send their USD overseas to somebody else. The problem with a lot of these apps is you can't pay anybody who's not in the app, you can't pay anybody who doesn't directly take crypto or they're not sitting on the same fintech as you. And so these cards, you can basically issue cards against your stablecoin balance. And the moment you swipe your card, whether you enter it into an app or you tap to pay or whatever, it debits stablecoins and settles directly on Visa Net using stablecoins. Visa has now seen these are one of the fastest growing use cases for Visa around the world. And they're just growing like gangbusters. You know, it's like 20% month over month growth. This, I think is going to be the way in which you see stablecoins go global is that they just get more and more integrated into the normal payments workflows for people. So how, what, what do you, what do you look at? Do you, do you see much geo, broad geopolitical risk to stablecoins? If I'm a kind of.

To like I'm in love with a stripper at the club. Unless like we're not that far away. Yeah, that's good. Everyone knows that just being in the streaming business, making all your money from streaming is rough. I think the is it. It's like a billion streams on Spotify or Apple Music gets you like low single digit millions in revenue that you're then splitting up with tons of different people. So not. Not super significant. And so the answer is do do constantly be touring, doing big tours. Are there artists that have insane streaming numbers that don't actually crush it when they go out touring? Like is that. That so. So that's interesting because I think you have. Streaming has created its own tier of superstars. If you think about NBA YoungBoy, the most streamed artist on YouTube in maybe three of the last five years. Two or three of the last five years? Yeah. Are you serious? Oh, 100% NBA YoungBoy. Look, NBA Young Boy is the most popular and meaningful rapper in America for the last five years. I had no idea in my head he's still like SoundCloud. No. Haven't you seen his earnings? He's outpacing the S&P 500. He had the number one. No, no. He had the number one debut solo rapper tour re by Revenue. The tour that Dak just completed last year, I think he did 70 million in ticket sales. Never broke again. He's never broke again. He called a shot. I love it. You will never be broke again. Put some of that in CDs. Someone who streaming created a marketplace and then went out into the world and received all the riches that came with it. But then you have other people who. You know to your point earlier about Miami xo, is this guy a star or not? Like who knows? Like is he going to be on one hit wonder bills ten years from now? Maybe. But you have people like say even Playboi Carti who like Contour well but isn't really trying to be like a public facing celebrity. It's like he's a little bit in the shadows. Everybody takes their streaming success differently and. But Youngboy is an example of someone who literally saw that and then went out and sold a million tickets off of that, you know, or something along those lines. I went to three dates on that tour. It was the most enlivening tour I've been to in years. What did you feel like you needed to go to three dates to do your job properly? I mean after the second, were you like, was it just like it's fun? I want to go. I chase joy, man. I chase pleasure. I love. I love it. I love it. The chat says he also has 13 kids. He may be broke again. Thank you so much for coming.

To any algorithm as you can. But two, when you think of how streaming platforms serve you music, the entire goal of a streaming platform is to make sure you don't press stop. It's to make sure you don't go away. And so not only did that affect the length of albums, because 30 songs is a longer play time than 10 songs, it also affected the structure of music. Think about Post Malone in the mid to late 2010s, right? What do you think? When you think Post Malone, you think that's a little bit hip hop, it's a little bit R and B, it's a little bit pop nowadays, that's a little bit country. But when you think of the kind of overall sonic framework of the song, it's real blurry, right? It's smeary. He's singing in this very stretched way. The beats aren't like. It's not like 90s rap beats where it's like, on the one. It's not that it's a lot blurrier. Streaming did that because streaming wanted to trick you into thinking that a song was never ending. And so you can press track one on a Post Malone album, and before you've even woken up, you're on track seven. The structure of the music, fundamental. It's so interesting because I remember, I don't know, being. Being young and like, buying an album. And then like, when you buy an album, it's such a different, like, for whatever, 10, 12 bucks, 15, whatever, whatever, however, was priced. Like, you're actually like, listen. You're not just listening. You're, like, studying it. You want to come away being like, do I really like this? And you want to be taken on a journey. And I would immediately clock, okay, I like these songs in this context. Like, these songs when I'm driving to school, this song in the gym, you know, you're kind of like mapping. I would map the album to my life. And then at some point it was like, oh, you made 30 of the same song. Thank you. Thank you for that. And, like, sometimes it's kind of like the cool vibe and it's fun to listen to, but it's not like watching a movie that's taking you through all these different emotions. It reminds me a little bit of jam bands in that way, which is not a genre of music I care much about, or certainly not one I enjoy. But anytime I've been forced to go to a concert of that style, I'm always kind of shocked that maybe it doesn't totally matter which point in the show you're dropped in or which point in the show you're pulled out. It's kind of always happening. And the dynamism is like, oh, and maybe minute four, he did this lick, but also in minute 47 he did this other lick that kind of talks to that maybe it doesn't totally matter when you entered. And I think streaming really, especially for hip hop, really, really pushed things in that direction. And what you're seeing now, like, when I hear bazooka, I'm hearing 2016 SoundCloud records. Like, I'm hearing Saw Baby Records. Like, I'm hearing things that post Young Thug that really, really made it seem like any kid with Fruity Loop software and like a tiny bit of a melodic sense could all of a sudden have a viral hit. It's really. I mean, we are living in 2016, not just, you know, in Instagram hashtags and everything, but the fact that we've cycled back and the biggest record of 2026 so far is basically a 2016 SoundCloud hit is, I think, striking and also talks about how nostalgic.

Progress is going to feel different a few years from now, once we get used to this, than it. It did, like, 10 or 20 years ago. Yeah. How have you reacted to that line from Dwarkesh that, like, diffusion is cope and that the models aren't actually that good, and so if they were better, they would. They would diffuse much more fast, Much more. Oh. Totally. I mean, I was actually talking to Martin last night about this, but I think we are now programming at the maximum speed that. That we can program. In fact, there are so many people who make so many mistakes and walk them back, or we ship crappy products and then walk them back, or we ship buggy products and then fix them, or the speed at which we're able to respond to user feature requests and bugs. We're totally tapped out. And so I think that we're in this really weird equilibrium state with how productive we can be in some of these use cases. And it's not super clear, at least to me, that a model that's 5% smarter at writing C code is going to dramatically change the amount of productivity we're able to create. Bull case for fdes, basically. Is that right? Yeah. It's interesting. It's like, you can do so much more, but there's still value and focus. Right. It's like, you know, a company that raises, like, Brain Trust raises a big round. You've got a lot of resources available. That doesn't mean, okay, we should launch 10 products at once. It's like, okay, you still have to, like, you know, focus that energy in. And. And, you know, weirdly enough, we actually raised a smaller round than we could have. I think we. We are trying to be very careful about not getting. Yeah, I'm sorry. Yeah. Trying to be careful about not getting caught up in. In the craziness. Like, we know exactly how much money we need to hit our revenue target for the next few years, and we raise that much money. No, Very, very smart. We're messing. Yeah. No, you're giving us time to get a bigger gong. Yeah, yeah, we. Get it. You're like. This one's. We'll be ready. Keep it warm.

The baseline for progress is going to feel different a few years from now, once we get used to this, than it. It did, like, 10 or 20 years ago. Yeah. How have you reacted to that line from Dwarkesh that, like, diffusion is cope and that the models aren't actually that good, and so if they were better, they would. They would diffuse much more fast. Much more. Oh. Totally. I mean, I was actually talking to Martin last night about this, but I think we are now programming at the maximum speed that we can program. Okay. In fact, there are so many people who make so many mistakes and walk them back, or we ship crappy products and then walk them back, or we ship buggy products and then fix them, or the speed at which we're able to respond to user feature requests and bugs is. We're totally tapped out. Right. And so I think that we're in this really weird equilibrium state with how productive we can be in some of these use cases. And it's not super clear, at least to me, that a model that's 5% smarter at writing C code is going to dramatically change the amount of productivity we're able to create. Bull case for fdes, basically. Is that right? Yeah. It's interesting. It's like, you can do so much more, but there's still value and focus. Right. It's like, you know, a company that raises, like, Brain Trust raises a big round. You've got a lot of resources available. That doesn't mean, okay, we should launch 10 products at once. It's like, okay, you still have to, like, you know, focus that energy in. And, you know, weirdly enough, we actually raised a smaller round than we could have. I think we. We are trying to be very careful about not getting. Yeah, I'm sorry. Yeah. Trying to be careful about not getting caught up in the craziness. Like, we know exactly how much money we need to hit our revenue target for the next two years, and we raise that much money. No. Very, very smart. We're messing. Yeah. No, you're giving us time to get a bigger gong. Yeah, yeah, we. Get it. You're. Like, this one's not. We'll be ready.

Multbook.com still kicking. Where are you? What's the. Yeah, it felt like prediction markets were the product that kind of broke out fully maybe last cycle. Some of these cycles blend together as different products get traction. Where are you most excited to invest? The new fund that you guys haven't necessarily invested before. We're still in this kind of moment as various laws work their way through Washington trying to figure actually get clarity on some of these things. But what are you most excited about over the kind of fun life Cycle for Fund 4? So there's the core bread and butter of crypto, which is just the financial applications. Actually I recently got into an online debate with Chris Dixon who he was kind of coming out giving this big expose about oh well, all the non financial applications of crypto is just too early and we still got to give them time to bake and maybe they're going to come back. I sort of took the other side as I think it is largely a death knell for the non financial applications of crypto. Crypto is about money, it's about finance, always has been from the very beginning, whether it's bitcoin, Ethereum, being programmable money to ICOs which are fundraising, to defi, which has finance in the name. Almost everything that's ever worked in crypto has been financial in nature. And everybody who's tried to make non financial things plug into crypto has basically failed. Now I think what falls under that umbrella is pretty big and I actually do think AI does fall into that umbrella, but it's more the ways in which agents are going to be interacting with money that very clearly. I mean you can see it right now in moat book. You go in moat book and you can see that agents are trying to find ways to pay each other for things and to get each other to do things for them. It's very primitive right now, but you can see where it's going. You can draw the line out about if you're an agent, if I have an agent, you have an agent and you live in a different country. Or maybe I don't even know who you are. I don't need to know who you are. If you're somebody else on notebook, it's very difficult for me to pay you. With a traditional financial system, it's not really designed to have non human recipients of money. We don't really know how the laws are going to work, how taxes are going to work, how money laundering is going to work. Whereas crypto doesn't ask any of those questions. Crypto was really designed for machines more than it was designed for humans. And so I think we're going to see this become an increasing part of the story of how AI agents are going to become more and more autonomous and self driving financially is by giving them crypto and allowing them to just interact with other agents and other systems using crypto Rails. So I'm very intrigued by that. But the core trend of crypto is just more and more of the same. Interesting. How do you invest in stablecoins as a category going forward?

Want to Multistream? Go to restream.com Micron is spending 200 billion. So congratulations for saying the biggest number, Micron. Micron is spending 200 billion to break the AI memory bottleneck. For decades, memory chips were low margin commodity products. Now the industry can't make enough to satisfy data centers hunger. Just like this one company is like, yeah, we're going to spend twice as much as India as the Adani Group. Boise, Idaho. Each afternoon at around 430, the earth here shakes from a series of controlled explosions as engineers blast through basalt bedrock to flatten out the ground underneath a gigantic new semiconductor factory. Let's give it up for Idaho. Yeah, we don't talk about Idaho enough. Yeah. Micron Technology is the largest American maker of memory chips. The tiny slices of silicon that store and transfer data and help power everything from smartphones and car computers to laptops and data centers. Micron is rushing to add manufacturing capacity to avert the biggest supply crunch the memory industry has seen in more than 40 years. In Boise, where the company is based, Micron is spending 50 billion to more than double the size of its 450 acre campus, including the construction of two new chip factories. The first fabs inaugural wafers are expected to roll off the factory line in mid-2027. So this is their first FAB or the one that they're spending the 50 billion on right now. I think the new one. Okay, so that's pretty quick. Yeah. Did you hear that the PS6, the PlayStation 6 is now delayed because of memory shortages. 2029, 2028, 2029. It's still a rumor. Maybe they'll, you know, maybe this solves it and they can get it out earlier. But pretty, pretty big delay to 2029. They really don't refresh the game. You just created a trillion gamers. No, seriously, I think adding insult to. Injury to the gamers might actually, gamers might rise up. They might be an important voting block. A lot of them are, a lot of them are of age to vote and a lot of them don't would rather have new gaming hardware than necessarily AI slopping the feed. They're like, yeah, I can't afford the new PC that I wanted. What do you think? Yeah, I don't know. I mean I feel like this says a lot about how good the PS5 is. Right. Because they can like afford to just like postpone the psx. Like what games? Oh, now you don't want technological progress. Wow. I wanted to go to the data centers. I don't I don't care about like the next like game graphics have like, have they gotten that much better in the past, like five years? Like yeah, maybe. But it's like yeah, for the actual gameplay. Is it that important if like the actual pixels are. Yeah, yeah. Realistically a lot of this stuff should be moving to the cloud soon if it's not already. Yeah. Like it was the Meta Quest VR. I forget. Oh yeah. Xbox Connect, Xbox Edition. Yeah, yeah. And it was all in the cloud and it was like totally fine. It was low latency. Yeah. And then if you're running in the cloud you can upgrade the hardware and in theory you should be able to run like a gen AI upresing pass to make it more photoreal and I feel like that's going to be where more of the the juice is squeezed out of the graphics than just continuing on the traditional path of more pixels, more ray tracing. It'll be make a really beautifully designed video game that works really well. Really tight deterministic interaction so it's satisfying and then give it a layer. We gotta have a RAM trader on the show. Really? Somebody that's in the thick of deal making in this space. There's a couple folks over at Semianlysis that cover it. One of them just went on odd lots of. Let's do it. Get him on the show. Let's do it. Let me tell you about graphite code review for the age of AI. Graphite helps teams on GitHub ship higher quality software faster. That's not all near Syracuse. Micron just broke down, broke ground. They just broke down in Syracuse. They just broke ground on $100 billion fab complex that represents the state of New York's largest ever private investment. Late last year, Micron announced a $9.6 billion fab in Hiroshima, Japan, while competitor SK Hynix announced in January that it would build a $13 billion fab in South Korea in addition to a $4 billion manufacturing complex it is building in Indiana. This is interesting. I feel like this is acceleration in capex from the memory makers. But this, I wonder how TSMC responds to this because the big criticism of TSMC is that they are increasing capex, but the rate at which they are increasing capex is decelerating. And so they're not really keeping up with the exponential growth in compute that every AI lab, every data center builder, every memory maker is pushing towards. And I wonder what do they know? What do they know? Or TSMC is being so conservative. I mean they might just have the AI industry in a chokehold. And so if there is a bump bottleneck, they can just raise prices, but they also might screw everything up by not investing enough. And then we get to some sort of solid bottleneck for a while and we can't actually roll out more. Well, let's get someone on from semianalysis tomorrow to talk about memory moving on. Lucas Shaw was on a tear over the weekend reporting on the Warner Brothers Paramount conversations. He says this morning Warner Brothers is going to resume talks with Paramount after two months of rejecting them playing mind games. The company still says it's committed to Netflix, but needs to find out just how much the Ellisons will offer. He originally reported on this Sunday, but it's being confirmed today. Again, we kind of knew this was going to happen if the Ellisons had been saying we're making a. We're giving you a big number, but it's not our biggest number. It's not our best and final. So no surprise here. Paramount has now just eked out a lead on who will successfully take over Warner Brothers. Over on Kalshi. They're sitting at 49% chance with Netflix at 37% and then none before July 2027 at this. Such an interesting flippening here. I was talking about this flip happened. Executive this weekend and he was like, yeah, I think there's just too many people that are against Netflix for a variety of reasons that it'll reopen the Paramount conversations. And specifically I was like, but YouTube. YouTube? And he was like, no one buys that argument. Like everyone thinks about the entertainment industry as its own thing. And then social media tech, TikTok, Instagram, YouTube as a separate thing. And no one is making that comp. It's just falling down. YouTube is not a, you know, having having. We asked Ashley Vance, how do you feel about having kind of one, one less big buyer in the buyer pool for documentaries? And he was like, not great. Already bad. This would be worse. But YouTube is not a buyer of IP. They host content, serve it to people. So it's a tough argument to make, even though they've been absolutely on an insane tear from an overall watch time standpoint. Yeah, hold on. Let me tell you about MongoDB. What's the only thing faster than the AI market your business on MongoDB? Don't just build AI, own the data platform platform that powers it. AJ says Grace Blackwell is a beautiful name for a baby girl. I totally agree. I mean, isn't that because it's actually the name of a girl? Right? Like, isn't Grace like the name, like, they're like the chips are named after real people. Who's. Well, there's like Ada Lovelace. I want to say Trevor. Grace Hopper. Grace Hopper, Yeah, Grace Hopper. So it was. And David. Harold Blackwell. David Blackwell. I always think Trevor Blackwell because he's a YC partner, but he did not invent the Blackwell. But Grace Blackwell is a beautiful name. Vera Rubin, too. Vera Rubin is also the name of a female mathematician I believe, working. The only person I can find named Grace Blackwell is an audio designer and composer based in the uk. So gonna have a rough go on the SEO front. Yeah. Let's flip over to Claude. Bottom. Kent Dodd says names the thing Claude bought. Claude asks for a rename. Renames to openclaw. OpenAI buys it. Legendary. Legendary. Couple weeks I was looking back on, like, when did we talk to Peter? It was like two weeks ago. That was when? Feels like two months ago. It feels like two months ago. Things are moving very, very fast now. No confirmation on buying. It's an open source project. They're keeping it open source. There's a whole bunch of different. Dave Morin reading is going to step in to, I believe, run the foundation that will, that will kind of steward the open source project. And then Peter's obviously joining OpenAI. This wasn't, this wasn't. This wasn't super surprising to me. When you looked at the potential places that he could land, it felt like there was. He had talked about like, hey, I'm kind of losing money on this. And he. Our takeaway from the conversation is this is a guy that, that just wants to keep shipping, get, get this product into the hands of as many people as possible. Yep. And when you looked at the kind of buyer pool, you know, he was, he was coming to the west coast. You can assume that he was making the rounds. Meta went on Lex and was like, yeah, I'm talking to OpenAI and Meta, which is like a crazy thing to say. And Meta always felt like a somewhat of a long shot because even though I think I'm sure he's done very well from this. Yeah. It didn't feel like dollars were his number one priority. It's very possible that, that Zuck would have been willing to, to overpay or pay whatever price needed to get him on board. Zuck had just also acquired Manus, which, like, feels like his, like agents team. Right. And Manus has already responded by rolling out some open claw like functionality. And so that, that didn't make sense. I think people were expecting anthropic to make a move. Part of that is like Anthropic's just had so much mind share and been on an insane run the last month or so. And so I think people and the name. Right. Peter also at multiple points said Anthropic didn't send their lawyers, they sent an individual. But it got turned into this meme that Anthropic had like really come and like really made it made his life difficult. I think they were like, hey, this is going to be confusing to consumers. And then the other thing is Peter's over and over and over talked about how much he loves Codex. Like he talked about that on our show. He's been very vocal about it. Like he likes building with Codex. And so I think when you understand all these different factors where not entirely financially motivated. I'm sure that was a factor, but not entirely financial financially motivated. So saying the biggest number is not going to get him to make a move. OpenAI with somewhere around a billion consumers that use these products, that are using products in the ecosystem today, that creates an exciting opportunity for him to come in and actually roll this out to as many people as possible. And you put all these factors together and him landing there makes a lot of sense. Yes. Let me tell you about Gemini 3 Pro, Google's most intelligent model yet. State of the art reasoning, next level vibe coding, deep multimodal understanding. Alex Hormozy chimed in on the open claw bot discourse. What did he say? He says, I'll take this day off to figure out this whole open claw thing. Every entrepreneur on President's Day weekend, we've talked about this on the show before. Long weekends are really good for AI progress and AI diffusion because people have a few less meetings and they say, hey, maybe I can mess around and figure this thing out. It really is remarkable how quickly claudebot openclaw broke through the discourse. I was at dinner at a steakhouse last Wednesday and I overheard a conversation with just a couple guys in suits who I don't think work in tech, talking about different models and how they wanted to use openclaw and they were thinking about getting Mac Minis. Like it was like it really did break through. Beyond our lawyer, our lawyer who doesn't work at a. He works at an entertainment focused law firm. He got a Mac Mini, set it up and so yeah, it's very mainstream. I did see some people reporting that the Mac Mini had sold out in their local area too. So it's not, it's not fully sold out by any means, but you look at the traction on GitHub and you look at just the overall demand and how kind of like mainstream. At least the, the, the concept of this has gone. Yeah. And Apple's overall production numbers on the device. I mean, I think we can, I think we could enter a world where, where at least the ones, the more in demand models are sold out. Yeah. Petition for 3 day weekends to speed up AGI timelines probably would work for sure. Fumble gate. Fumblegate did anthropic fumble Open Claw. Darius says I don't think about you at all. Probably overstated. It doesn't seem like they, it was like key to their strategy. They're sort of just focused on their own thing. I don't know. It doesn't seem like, oh, they completely fumbled. Yeah. Part of this was the original naming Claude Bot. Right. But when you look at how many different projects Peter shipped, it's not like he was thinking at every moment, oh, I have to name this because it's going to get a million GitHub stars in a month. Yeah. He was, he was just like ripping products, building really quickly. Yeah. And it wasn't even Claude wasn't the default. Yeah. No in, in Open Claw ever. Yeah. Yeah. Codex was actually the number one. And there are like 10 LLMs that you can choose from when you, when you actually set it up. It very much seems like he's just having fun with the name. I mean the fact that he was calling it Moltbot for a while is. Just tells you that was a great, that was a great, that was a great 24 hours. 24 hours. But, but, but seriously, like he, I, I don't think he was like, oh, I need a branding agency and I need to think of the name that will go forever. It's just like, yeah, this is like all a lobster meme. It's fun. Wes Winder says lol. Meta couldn't acquire Open Claw so they pulled a classic Zuck move. Just clone the whole thing. Then it added as a feature to an existing product. Surprised they didn't try to shove this into Instagram. So Manus announced Manus agents yesterday. They move quick. Your personal Manus Manus Manus now inside your chats. Long term memory full Manus power. Create videos, slides, websites, images from one message. Your tools connect at Gmail, Calendar notion and more. Available now in Telegram, I would assume. Like kind of surprised they wouldn't roll this out into WhatsApp. Yeah, they don't even own Telegram. Available now on. Wait. That is so confusing. I think it's possible that WhatsApp didn't have the. Didn't have some functionality that they needed to actually do this integration. And they're moving so quickly. They seem to have turned this. I don't think. I think like, Wes is maybe overstating a little bit like the classic Zuck move. You can imagine, like, Manus would have done this probably independently. 100% you were calling this movie. Manus is a good acquisition because if. You want super intelligence for like, Manus is great. Yes. And so Zuck has been great at buying a product that has a lot of potential, that has some, like, customer traction and like scaling up massively. Yeah. And even just for the most mundane Instagram usage where you could say, oh, I want to look across my audience and see who's the most engaged or understand what trends are happening in my vertical, or respond to every comment with a heart. If it's positive. Anything that you'd want as an agent to go around on these social platforms and do, let alone go buy something for you. Having a team that works on agents makes a ton of sense, but it's just funny that they're rolling it out. Mattea says, I think that's connected to WhatsApp deplatforming other AI companies. Remember ChatGPT had like a huge integration in WhatsApp and they got. And they got booted. That's crazy. Hari shares a little bit of lore about Peter Steinberger. Says he bootstrapped PSPDFKit over 10 years ago. It's the gold standard PDF library used by Box, Apple and even DocuSign. That's crazy. Insanely hard tech. If you know, you know, it's very hard to work on PDFs. I guess he's worth a couple hundred mil from that. Everyone's trying to guess the number, but nothing's been confirmed. If he goes to OpenAI, he might make more from OpenClaw, which is three. Peter says he signed the contract today with software my company built years ago. Now it's nutrient docs. Nutrient still best for anything PDF. It's like I'm PDF maxing. Let's take a look at this. Harnesses won't matter in less than three years, says Ahmad. Yes, there's money to make at the moment, but keep in mind that this is. This is the play of two years ago and you should be playing for what the models will be capable of in two years from today. Instead, Claude plays Pokemon. Kind of proves this. Opus 4.5 and 4.6 has the same insanely bad harness. No babysitting like Gemini and GPT and 4.6 is like 1,197 hours ahead of 4.5, so. Oh, it's how far they get. Yeah. I've seen some people trying to back a bunch of like open claw startups and given that just feels like a potential bloodbath, I think especially, especially knowing that OpenAI will be able to build, I would assume build the best possible kind of wrapper around everything that OpenClaw's built so far. Again, I would assume that the three major labs, DeepMind, Anthropic and OpenAI all have serious orchestrators very soon and they will acquire talent and acquire products and build products and license stuff and figure out all sorts of integrations, do all the business things necessary to make it happen. But as I wrote in the year of orchestration or orchestrators like this is coming, people are managing multiple agents already and tools that make that easier, tools that make that more reliable and more efficient. Like that's going to be a big focus for this year and everyone's going to be focused on. And so the idea of like, oh, I should just like fork this open source project and like bring it to enterprises, like that's going to be rough when OpenAI has 1000 for deployed engineers going into every enterprise that they already have contracts with. Yeah. Tyler, what do you think? Yeah, I was going to say, like, I think it's a similar thing to the hardware. So like Gemini is developing the TPU with the architecture of the model in mind and they're developing the architecture with the hardware in mind. So it's kind of like these things are helping each other. So you're going to see the same thing in the harness, where the harness is made with the model in mind, obviously. But then you're going to start to see as these things as agents become more and more popular. OpenAI already has, there's the normal GPT 513 and then there's 513 codecs. Right. So the models are like built with the, with the harness in mind. Yeah. So I think if you're not the one building the model and you're just adding your own harness. Right. The model's not going to be like customized for your harness. So if OpenAI Anthropic are building their own harness, they're going to do it with the model in mind. It's going to be much better than having some third party thing that you build, right? I would think so, yeah. No, that's totally reasonable harness for horses. Bobby. Kaufman says yeah, it really is. The horses don't stop. Harnesses don't stop. Stop. Will Brown says honestly crazy that OpenClaws sold for 1 billion. Like he's really the first Solo $5 billion founder. Time will tell if it's worth 15 billion that OpenAI spent on the acquisition, but it's pretty wild that you can just vibe code an open source project and make 40 billion in a couple months now. It really, really nails it because everyone jumped immediately to a billion immediately off of nothing. Off of nothing. Off of like one rumor from a thinking machines person that went to Meta and we don't even know what the earnout schedule for that is and how much that happens. But it's very, very funny. Who knows? We gotta give some credit to Simp for Satoshi. I brought this up. I brought this up before. Simp has been working on truffles for a handful of years now. I've been by his studio in Venice. It's a very cool setup and team over there. But he posted a little bit about this. Where is it? And Tinybox, George Hotz's project is also apparently going to ship a more consumer product. Maybe. I think they're around $10,000 right now. They're probably going to bring that down to a few thousand dollars or maybe even a few hundred dollars. You get a lot with a Mac Mini, but we'll see where the price is land after the memory shortage takes hold and whatnot. But lots of people are doing this. Alex Cohen breaks it down for Gen Z If you're wondering what happened today, Claude was mogging OpenAI for weeks. Then this Jim Cell dev ships Claude bot which was the fastest growing open source thing ever. Absolute looks max for the whole ecosystem. Anthropic tries to dairy goon him with legal Dev renames to OpenClaw. OpenAI slides in like a void pulling chad with acquisition interest. OpenClaw gets acquired by OpenAI. Now Anthropic is getting gest dragoon by the Entire timeline and OpenAI is Giga maxing off their fumble. Open Anthropic could have just let them cook him cook. Instead they went full moid and got out framed by the Jester Maxers at OpenAI 8000 likes the looks max lingo is really, it feels like hilarious. I do wonder the half life. I feel like it's. I think it's over. It's got to be. It's got to be towards the end in this boom. But the rise of the kickstreamers is certainly the story of the year, certainly the story of the year. Dave Morin shared more context on what the future of openclaw will look like. So Dave Morin, who you might know from offline ventures slow. He's been on our show path. He is now on the board of openclaw and he says he's been working on the openclaw foundation structure for weeks. A home for thinkers and hackers that choose and those who want to own their data. He's honored to serve as the founding independent board member. This community built something extraordinary. Our job is to protect it. Open source forever. Excited to share more soon, Raul says let's go. There was some funny pushback about. Someone was like, oh, it's so unfair that Peter's reaping all this reward from these work like it's an open source project. There were a lot of committers and he shared the number of commits and it's just like him and like 90% of the work. And then like a few of his buddies who are all in the chat, in the reply like, yeah, like we love this, this is great. And a number of them are going to OpenAI too. He's like the headline pickup but. And then a number of the other committers and then those AI models. Yeah, and then, and then the other key contributors started committing after. It was like fully, totally, totally. Hundreds of thousands of people and it. Was building on top of it. It was, it was like pure. I don't know, there's no term, I don't know if there's a term for it, but it was like, it was like third party, like angst or something. It was like people were feeling angst on behalf of people who felt no angst. Like the people on the team were all like, this is awesome, we're all on board. And then other people were like they got screwed. And it's like, I don't know. Not quite. Intern is sharing what your 2027 team off site will look like. Taking the Mac Minis somewhere nice. Taking them out? Yeah, we'll see. And here's the post you were talking about from Ashen. I didn't believe it, but Mac Mini studios are actually sold out in most places. Walmart, Apple Store, Amazon, Best Buy, Micro Center. Apple has a month wait now too for most AI type beat models. 48 gigs plus of Ram. Funniest part is that there are still 24 gig. What's up 48. They said it was possible. Oh, that they would sell out. Yeah, yeah. I went to the Apple store and they were fully in stock. But you know, the timeline moved on and people, people went more aggressively into the Mac Mini. Let me tell you about Vanta Automate compliance and security. Vanta is the leading AI trust management platform started today. We have some breaking news out of New York. Oh yes, what happened? Josh Kushner has announced Thrive 10 exceeding 10 billion. Thrive 10 compromises, 1 billion designated for early stage investments and 9 billion designated for growth stage investments. He said we do not view this as a milestone, but as a commitment to the long work ahead. We view Thrive as a company. Our product is partnership. The willingness to commit deeply to a small number of founders and to stand with them through momentum and adversity. This is the discipline we bring to our work and the responsibility we accept. When founders partner Thrive, we do not hedge. Concentration demands loyalty to the founders and missions we back in this moment. Exposure alone is not a strategy. Judgment without commitment is not enough. Advantage will accrue to those who choose deliberately commit deeply and endure through difficult moments. Thrive was founded to be an enabling technology for the world we want to see. We are deeply aware that we are not the main character. The founders that we are fortunate enough to partner with are the artists. Our role is to help create the conditions where great work can come to life. We take a long view grounded in the belief that category defining companies tend to create structural compounding advantages over long arcs. This fund reflects the continuity of our approach and the ways our work has deepened alongside the founders we support. We are grateful for the trust our limited partners place in us and for the opportunity to work alongside those who are building with purpose, integrity and courage. Hit that gong. Love to see it. Numerology 510 and 10 billion. I like that. Yeah. Despite having zero exposure or anything to tvpn, he's been nothing but kind and helpful to us on the journey. And it's. Yeah. Testament to the work of the whole Thrive team. And I will say their merch is phenomenal. Really good. I was wearing the. You were in the pajamas. The pajamas. No way. I was wearing the pajamas this weekend. Yeah. They got some good stuff. Creative stuff. Like different. Different stuff. Unexpected drops, which I love. Yep. And we have to. We have to read this off. Of course. Just so good. Selah says Joshua Kushner strode billionaire Lee across his room. His contrarian high concentration vibes engulfed Rick Rubin's bohemian monk retreat. Joshua's high conviction and mystery filled the room. I know what this is referring to. This is a reference to the Colossus profile. Which is. Which is. Which really got a Lot of people upset. A lot of people were not like this. This doesn't count as writing how it was. It transported me to Malibu. It was fantastic. I thought it was great. Let me tell you about Lambda Lambda is the super intelligence cloud, building AI supercomputers for training and inferences that scale from one GPU to hundreds of thousands. Let's head over to Tyler Cowan, Tyler Cowen. He is giving us some more economic data on AI productivity. Quote, you see tech and AI everywhere but the productivity statistics. That's what people say. He says, how many times have I heard versions of that claim? Eric picks up the telephone in the Financial Times. He says while initial reports suggested a year of steady labor expansion in the United States, the new figures reveal that total pay payroll growth was revised downward by approximately 400,000 jobs. Crucially, this downward revision occurred while real GDP remained robust, including a 3.7% growth rate in the fourth quarter. This decoupling, maintaining high output with significantly lower labor input is a hallmark of productivity growth. My own updated analysis suggests a US productivity increase of roughly 2.7% for 2025. That is a near doubling from the sluggish 1.4% annual average that characterized the past decade. It's fine to suggest this, suggest caution in interpreting such statistics, but they hardly push the other way, says Tyler. And yeah, it's just a very fascinating that we're potentially the end of stagnation. In seeing productivity growth relevant to this conversation. You were looking at employment data out of the Philippines. Philippines, yeah. So I've been very cautious about the AI will take all the jobs narrative. When will that happen? Is it happening yet? Is there Jevons paradox to the type of work that we're doing. How much software do we want? You know, we have the ability to write way more lines of code. How many more lines of code do we actually want? It feels like there's a lot of opportunity. It feels like the labor market is weak, but it hasn't collapsed. A lot of people still have jobs. There's a lot to unpack. And so I was interested to know, are we seeing a massive spike in unemployment in the Philippines or in India where a lot of white collar labor is outsourced, whether it's answering the phones or doing business process outsourcing. A lot of companies have back offices in. I was talking to somebody who works in architecture and they said that a lot of the architectural drawings that happen happen overseas. The same things happens in the big three management consulting firms. You want a lot of management consultants will sort of sketch out a slide and then they will send it over to a team offshore that actually turns that into the final PowerPoint product. And you can see how if you have a sketch and you just need to turn it into a real chart, that's a textbook use case of artificial intelligence. Whether you just style transfer it with something like nanobanana or you go and do a deep research report and then basically code the actual chart. And all the models are very capable of generating charts based on data. That's actually how I generated the chart of the Philippines unemployment rate. And I saw from both the Philippines and India there was not a major spike in unemployment. Maybe it's coming. You know, we've heard from the ground the job market's not good and maybe the jobs are getting worse, but there's at least there's not high unemployment yet, both in the United States and abroad. So it's something to keep an eye on. But I don't know, I've just been like looking for more solid data points around labor displacement because it's such a huge narrative. I don't know. We'll see. Let me tell you about Railway. Oops. There we go. Railway. Railway is the all in one intelligent cloud provider. Use your favorite agent to deploy web apps, servers, databases and more. While Railway automatically takes care of scaling, monitoring and security. What did Claude do? What did Claude do? Pentagon has said that Anthropic will pay a price. Of course there was reporting last week that Claude was leveraged in some way during the Maduro planning. The planning of the Maduro raid. I was imagining in my head Dario as Walter White in the suv just being like watching the logs and seeing Pete Hegseth running a deep research report on Maduro. Who is Nicholas Maduro? He's just like, no, no, don't do it. Yeah. Very, very unclear how, how it was used, but a lot of pushback. You know, Palmer, Palmer was pushing back pretty hard. Let's see if I can pull up what he actually said so I don't botch it. The Wall Street Journal headline is Pentagon used Anthropics clawed in Maduro, Venezuela Raid use of the model through a contract with Palantir highlights growing role of AI in the Pentagon. Now I remember seeing a clip from Sham Sankar at Palantir talking about how every query that the government runs is actually run through three or four different LLMs and then synthesize. They sort of put their own model router on top of the other models. Because if you're doing Fed wrapper, basically I guess it's basically a wrapper because you sort of want every possible, you know, piece of information. All the LLMs have different parts of the Internet, different training corpuses, different opinions, different. All sorts of different things. So you put all those together. The mission to capture Maduro and his wife included bombing several sites in Caracas last month. Anthropic's usage guidelines prohibit CLAUDE from being used to facilitate violence, develop weapons or conduct surveillance. We cannot comment on whether Claude or any other AI model was used for any specific operation, classified or otherwise, said an Anthropic spokesman. Any use of claude, whether in the private sector or across government, is required to comply with our usage policies which govern how CLAUDE can be deployed. We work closely with our partners to ensure compliance. The deployment of CLAUDE occurred through Anthropic's partnership with data company Palantir Technologies, whose tools are commonly used by the Defense Department and federal law enforcement. Following the raid and employee at Anthropic asked a counterpart at Palantir how CLAUDE was used in the operation, according to people familiar with the matter. An Anthropic spokesman said it hadn't. It. It hasn't discussed the use of CLAUDE for specific operations with any industry partners, including Palantir, outside of routine discussions on strictly technical matters. Anthropic is committed to using Frontier AI to support in support of US national security. Okay, so we'll see where this goes. Anthropic's concerns about how CLAUDE can be used by the Pentagon have pushed administration officials to consider canceling the contract worth up to 200 million, which is a drop in the bucket at 14 billion ARR, but still probably has larger implications for their relationship with the government broadly. Yeah, there was. Someone was saying. I don't know who needs to hear this but punishing private companies because their CEOs don't share your politics is really bad capitalism. Palmer said this is not a reasonable characterization of what is happening. It isn't a matter of punishing companies for not sharing political views. It is a rational response to a vendor trying to control the government via terms of service and the products they power. I hope you realize. Oh yeah, people are going on. Okay, wait. So Disclose TV shared it as well? Pentagon used. And what is this? Someone launched a Federal Reserve simulator. Is there anything more on the. No, we can move on. This is more important. Somebody Bill Mead Gaming just released the Federal Reserve Simulator in Steam. If you like flight simulators, you're probably going to love this product is going. To go to zero turn based people aren't going to Be going to work. They're going to be taking days off to play this. A turn based economic simulation game where the player assumes the role as head of the United States for Federal Reserve. This is awesome. Tyler fired up this. I would actually play this. Yeah. Fire it up. Fire it up. Download it. I want your review. I like Garrett Jones here. Promising. And he's a professor of economics at George Mason. He's actually serious. I feel like this would be an amazing game to play. In the reviews I'm reading on Steam, they say I learned more about monetary policy in 15 minutes on this SIM than I did in five years of university. That's actually. Please do investment banking simulator next. Yeah, I was thinking about the LBO. Simulator would go so hard. Let me tell you about Cisco. First unlock infrastructure critical infrastructure for the AI era. Unlock seamless real time experiences and new value at Cisco. So I was thinking about Sholto's Age of Empires real time strategy game based on the AI era and I was wondering if it's easier to ship basically a skin or a mod for Age of Empires because the actual fundamental mechanics are pretty tight and you actually don't want to mess with those because the game's so balanced and it works so well. What you really want is just instead of archery units and horses, you want robots and Terminators and you want a sci fi theme on top of it. And I feel like that would be something that an agent could work very well at. Just export the image and the description and the words of every single unit and item in the game. Export all the textures and then upload each texture to nanobanana one by one and say this is a tree. What does the cyberpunk tree look like? What does the Diamond Age tree look like? What is the 2026 tree or building look like? And then it just iterates through all of those to create effectively a mod or a, or a skin for the game. But if you do want to change the actual underlying structure, you got to go deeper. But that's, that's a lot of tokens. So we'll see, we'll. We'll see where it goes. We got to keep pressing him on his game. We got, he's got to ship it. You got to ship it was. It was. You got to beat the. Beat the vibe coding allegations right? The never ask somebody what. Never ask a Claude code maxi what they've. What they've shipped. He's got a shit. I want to show you this Instagram video because it had me dying, laughing and so if we can play this, we're gonna get demonetized, but we'll talk over it. So he Vibe coded a calculator. Can we zoom in? I don't know. You Vibe coded a calculator and you type in your numbers, you type in 5 times 9 and then it says unlock results. $299. One time payment. You pay and then it gives you payment successful. The answer is 45. This is the future of Vibe coding. I loved it. Anyway, I wonder, did he actually ship the product? Because this got millions of views. You'd think someone would just be like, I want to know if the payment rails work. Let me see. Quickly, let me tell you about Fin AI, the number one AI agent for customer service. If you want AI to handle your customer support, go to Fin AI, SAG, AFTRA. What happened? Put out a statement on CDance 2.0 and it's not a comment, it's a statement. The Chinese, now the Chinese have been quivering in fear ever since SAG came after them. What'd they say? ZAG stands for the studios in condemning the blatant infringement enabled by ByteDance's new AI video model, Sea Dance 2.0. The infringem includes the unauthorized use of our members voice and likenesses. This is unacceptable and undercuts the ability of human talent to earn a livelihood. It is kind of interesting that just in this statement they're admitting to saying like it's so good, you're, you're going to make it impossible for our members to earn a living. Which doesn't actually. It says undercuts, undercuts. Doesn't say eliminates under troops. See Dance 2.0 disregards law, ethics, industry standards and basic principles of consent. Responsible, Responsible AI Development demands responsibility. That is non existent here. Completely correct. Some of the Sea Dance videos are insanely infringing. It's just like, wow, it's Larry David. And it looks exactly like Larry David. Do we license this? Let's figure it out. Beginning of the end says growing. Daniel? Yeah, we have the Larry David video, but I don't think we want to play it. It's very, it's very uncouth. Yeah, we'll skip that one. Disney, as expected, sent a cease and desist letter to Bytedance over Sea Dance 2.0. I wonder. It's crazy. I wonder how Bytedance will actually react to this pushback. Obviously they expected it. Yeah, they know that they're not abiding by a number of different US laws. Whether or not they care is another thing. Yeah, I wonder how. Like, what will the. Will the equilibrium be here? Like, they don't allow usage in America, but everyone goes and gets VPNs. Is CDance 2.0 open source? Because if you can just run it, it's like so hard to play whack a mole and actually get the. And actually get this taken down in any meaningful way. Like, the cat's pretty much out of the bag. If this is something you can just download. I don't think it's open source. Okay, so it's running on open source. But again, like, in some ways, a lot of this stuff will need to be solved by meta and YouTube. So if somebody makes a video with a famous actor and they're like, just monetizing that, I would assume that YouTube should be, in the fullness of time, forced to demonetize that content or just give the revenue back to the actual IP holders. Yeah, you think that you'd be able to. And then meta, Meta. The other thing is meta. Like if you. If you make an AI version of Andrew Huberman and you get in a fresh ad account, you can probably start spending money before for the human lab team finds out what you're doing. I would disagree. I think Rob's on top of it. I think he's goaded. But anyone else, any other team would be cooked. I don't know. I mean, he might respond faster than the others, but this is certainly happening. He is superhuman. And this kind of thing has been happening to Joe Rogan for years at this point with just like very poor deep fakes. But yeah, I watched a video on Corridor Crew about how to identify the latest. The latest AI slop. And it was very interesting. All the different techniques that are used now, there's some really obvious ones where it's just clearly AI video from end to end. And so if you look in the background, the wood grain on the wooden poles or the walls changes from shot to shot. And that's like a tell that it's not consistent. But there were some where there was an AI influencer that was taking real photos and then just adding this AI influencer into the photo. So it'd be like Joe Rogan with Tony Hinchcliffe replaced Tony Hinchcliffe with this AI influencer girl. And then you look in the background and you're like, there's no hallucinations. Like, everything looks perfect because it is. It's just the base. It's just the base image. And then you're just adding this one character. There were a bunch of Interesting things. There was also a video that everyone was calling AI that was just a dog painting. And it just turns out the dog can paint, which is awesome. But it was a little bit of a magic trick because the dog couldn't actually paint a picture, but it could sort of pick up the paintbrush and, like, draw strokes. But the dog was next to the dog's owner. And the dog's owner. That's not art. It is absolutely art. But the dog's owner was actually a great artist and would switch paintings with the dog. So the dog would make one, like, big, messy brush. They would switch paintings and then she would turn that messy brush into a tree and, like, adapt. So they adapt the time lapse? Yeah, well, yeah, there were just some cuts and so they would switch. And then whatever the dog did would sort of serve as, like, the interpretation and the case for the artist to, like, take it to the finish line. And so you wind up with these, like, two beautiful paintings that, like, the dog and the artist sort of collabed on, which is kind of cool. Yeah. Gavin purcell says if ByteDance doesn't restrict or nerf Sea Dance, this is going to escalate to a political conversation. Unlike the Sora 2 launch, because OpenAI restricted likeness ahead of time. Sea Dance shows just how clearly every major movie star is generatable by AI. So this is interesting if. If you're already like an A list, massive superstar, I think you see some stuff like this and you're actually like, great, I'm going to be able to shoot a movie in a week from la. I'm not going to have to travel to these insane exotic locations and spend a week in the desert filming all these clips. So if you're like a Timothee Chalamet, this is maybe like, yes, you're worried for the overall industry, but at the same time you're thinking, okay, my name and likeness is now infinitely scalable. I can still, like, restrict the supply to some degree. Right. I'm not going to tell any movie studio, hey, you can make a movie with me, whatever. You're still going to kind of restrict it and have pricing power. But if you're a, the question becomes new, new talent that's emerging, trying to build their brand. At what point do studios say, we're just going to make a character, we're going to make a new actor out of thin air, place them across different movies, build them up over time. You could imagine. I don't think a company like CAA would do this because all their talent would Be like, what are you doing? Like, you're. You're taking our job. But I could imagine a group trying to make like a little Michaela style actor that you build up over time. 1. One thing that we'll find out is how much does the actual actor's real life matter in the context of their career? Like, if Timothee Chalamet is dating Kylie Jenner, does that, like, increase his appeal on the big screen? Yeah, and I would say, yes, probably, right? There's so much fixation on the lives of all this talent. You could sort of fake a decent amount of that. You could fake paparazzi photos, you could fake, you know, vacation photos. You could create a whole world around a fake person potentially. But if people find out, will that break the illusion? Will they be not into it? But it's not real drama. People like the drama. They like following stars. I'm still. I'm still just pretty optimistic about, like, AI as a tool in Hollywood. Did you see that fully, fully AI generated movie about the woman who goes in the cybertruck that went viral this weekend? Wasn't it, like, only three minutes? I think, yeah, exactly. It was only three minutes. But it was consistent from one shot to another, which is impressive. And somebody said, like, not enough people are talking about why this, how much this sucks. And Ben Stiller chimed in, was like, because it's bad. People don't talk about things that are bad. And it was. It was sort of interesting. I think there's still a lot of interesting. Like, I was thinking about, like, imagine if Kevin Feige. Is that how you say his name? He's the creator of the Avengers. Don't ask me. Don't ask you. Anyway, I don't know Christopher Nolan or James Cameron. If James Cameron got up and was like, I'm introducing Avatar. It's a great movie. And yes, we use cgi. It's like, no, the CGI is, like, just a tool that he uses. It's like, you don't say, like, yeah, we're doing Avengers, and guess what? We use green screen. And that's a selling point. It's like, it'll always be a tool. And so I would assume that there will be amazing directors and creators in Hollywood that use AI as a tool in certain areas, but they won't. Like, you'll know it's arrived when they're not making a big deal out of it. Like, it should just stand on its own. Right now, things go viral because you say, this is 100% AI generated and you're like, okay, I want to know how good is this AI stuff? But no one's just like, here's my green screen test. And they're like, okay, awesome. We know green screen works. We know it's a thing. It happens. Yeah. When I see the Sea Dance content, I think, okay, if movie budgets stay the same, I think we can potentially get a lot more great movies. And that's specifically because how much of these budgets goes to. Let's say somebody's making, like, a movie, like, 300. And we need to get, you know, a thousand people out in a field. 300 people, but, yeah, who's counting? We need to get a lot of people out in a field and spend days, weeks, you know, recreating all these different scenes. Yeah. And now that budget can go towards. Yeah, the famous Henry Cavill. Superman. Right? They finished shooting Superman. He moved on to, I think, a man from uncle. He grew a mustache. And they were like, we got to shoot another Superman scene. And he's like, contractually, I cannot shave my mustache. And they were like, fine, we'll do it in post. And they have him come on, and they CGI'd like a shaved lip on the top of his mustache. And it looked really bad. And it was like, people called it out. I actually think a lot of people saw the movie and they were like, that's fine. It's pretty quick. But the VFX artists were like, this is bad. But with AI, obviously, that's much, much more doable these days. So anyway, console. Console builds AI agents that automate 70% of it. HR finance support, giving employees instant resolution to access requests and password resets. Andrew Curran gave us a shout out. Very nice of Andrew Curran. He said, one of the reasons Sea Dance 2 has impressed everyone is that it does the same thing that Sora did make very detailed video gens from very simple prompts that there's probably some prompt hydration going on. There's reasoning in the LLM that then prompts the model. So you type in one sentence. It hydrates that into a lot to work with for the model. In fact, most of the Sea Dance clips that went viral were made from one or two sentence props. We never did find out exactly what's going on under the hood with Sora. TVPN asked about it when Sam was on, and I think he means Bill Peebles from the Sora team, and he says, thank you, guys. So thank you for shouting us out there. But what they chose, but they chose not to answer this specifically, so they dodged our Question? No. We asked the hard question, give us the ip. They said no. This is the funny thing about hard questions in tech interviews. You're like, okay, so tell me how many parameters in the model? And they're like, that's a great question. We're really investing in advancing the frontier. I saw what you did there. But it's understandable with CDance 2. We don't have a paper yet, but there is one out for CDance 1 and it says that they do use an LLM to convert the user prompt into a much more informationally dense and structured style of the prompt that the video model was trained on. They use a quen to do this. This is what most of us guess was going on with Sora. A version of GPT5 maybe fit for this, was rewriting our prompts and adding details. Vanilla 5x would already be great at this out of the box. So it doesn't even need to be fine tuned, really. If this is how Sora really works, then everything does make more sense and there's a whole bunch more context from Andrew Curran's post that you can go check out. So Bojan Tungus is saying, I can't wait to use Sea Dance to fix the last season of Game of Thrones. Go do it. Go do it. I think that the last season of Game of Thrones was not. I mean, I thought it was good, but also I think that everyone says, oh, I wish it would end this way. I don't know. I think if you did that, you would actually be like, ah, it's harder to land this plane than you think. There are a lot of things. It'd be interesting when people are watching a series in real time. And then in advance of the final episode, let's say there's a week they're like, okay, I'm just gonna make the version that I want. Yeah, and I'll put it out. No. People on Reddit had written like full endings of Game of Thrones and some of them were really convincing. Some of them I was looking forward to. And it went a different direction. But, you know, it is what it is. Let's pull up the linear lineup so we can tell you about the guests that are coming on the show today, because we have one in the Restream waiting room already. I'll let you read this while I tell you that LINEAR is the system for modern software development. 70% of enterprise workspaces on LINEAR are using agents. We have John Karamica, New York Times music critic and buyer of Fake TVPN merch coming on the show. We have Spencer Skates from Amplitude. We have Haseeb from Dragonfly. We have Celine from Loyal, Anchor from Brain Trust and Reed.

Do it. Are we. In the Cournot equilibrium? Do you agree with that characterization? How long does that last? Walk me through how you think the AI lab economics might change over the next few years. I think it's really simple. So basically when new models are coming out, people forget about open source and they forget about economics. And that's because it changes fundamentally what you can do with a model in the first place. Programming is a good example. The workflow today is completely different than it was a year ago. And if you tried to use the models from a year ago today, you wouldn't be able to do crazy stuff like Cloudbot or whatever that people are building nowadays. When models don't change at that speed, then what happens is people optimize the performance on the use cases that they have. And so in between major model releases, and I think we're in one of these periods right now, there are new models coming out, but they're not fundamentally changing what's possible or not possible. You see a lot of interest in open source models and I think that's very exciting right now. Although there are not that many companies using open source models, if you look at usage on our platform, almost half the usage, like token usage is coming from open source models from a very small number of companies that have figured out how to optimize use cases really well. And so I think it's going to be really interesting. Okay, so I feel like.

US app. But before that, basically today, unless you were on the US app, you were using it on chain. If you look at increasingly a lot of the apps that are really working in crypto are more fintechy and we're seeing a lot of fintech investors coming up against us in a lot of these deals. And then of course, like you mentioned, crypto is increasingly bleeding over with AI. There was just this acquisition of OpenClaw, I think it was yesterday announced that OpenAI is acquiring them. The Multbook drama if you go on multiple book today. So I still look at Multbook every. Day. And most of the highest ranked things on multiple crypto are all crypto related. It's all, almost all. But is that, is that because people are trying to create. Is that because people are creating projects? There's, there's a lot of, there's a lot of bullshit. Going on right now in they're like you don't even need users now. You just create a million users for your project. That's right. That's right. Look, it's the infinite variety of people will always find ways to try to make money from anything. And crypto, it both represents the best in humankind and the worst in humankind. Part of the reason why it'll never go away, no matter how unseemly or distasteful or just insane it becomes, it just never dies. So it's something that I was just reflecting on because we're now eight years in to Dragonfly. It's been a long time, been a lot of cycle. We're now clearly in a trough of this cycle. Sentiment in Cryptoland is extremely low right now. But on the other hand, the actual adoption stablecoins you see all these financial institutions coming into the space. Quote, multiple of your sponsors are crypto sponsors and they're seeing their underlying traction go up even as prices are going in the other direction. So I feel really good and it is bleeding into everything. Quick update on multiple.

And widgets and text responses that's formatted that tells you all about it. Yeah. Talk about the SaaS apocalypse. Are you worried about people vibe coding their own in house version? Like what are you seeing on the ground? What do you think long term moats and software look like? Is anything? Yeah. And I think there's like two things like what happens to the core business and then how do you create net new products or this new kind of category of company that has entirely new kind of needs around, understanding how their products get used and all that kind of thing. Yeah. So first on SaaS, I think the key lesson is the software you've already built that is no longer remote because to your point, whether it's someone vibe coding it or someone going after as a company, they can replicate it pretty quickly. And so the new key moat in my mind is it's a little trite, but speed of innovation, you have to be pushing the bleeding edge of what capabilities are and that's what customers and if you do that and you create a company that can do that, that's what customers will buy. What I think the SaaS apocalypse has actually gotten right is if you look at the median SaaS company, whether it's a few hundred million ARR like us or whether it's in the Billions, the median SaaS company, their innovation has actually slowed to a standstill. I don't know if you guys have ever been inside of these, but it's crazy how little that they ship in terms of net new products. They'll put some nice branding on it. They'll be like, oh, introducing our new salesforce AI agents. But then it's like, okay, this is the same thing you guys had last year. What are we doing here? And so I think the difference is speed of using the bleeding edge of these model capabilities is all that matters. And you're even seeing it in foundational model companies. It's right like okay, Opus was the hottest thing a week ago, then it was Codex, literally launched the same day and said, hey, we're even better and we're faster and we're on cerebrus trips and all this stuff. And so I think it just matters the rate of improvement. The analogy I've heard for software that I actually like by one of my a former Amplitude investor is it's like sushi. And so buyers are always going to want the best thing. And so if you're keeping up with innovating the best thing, you will be able to charge a premium. And so it's fine that the gas station at the 7:11 at the Gas station now sells sushi. That's not going to put Jiro and Japan out of business. And so for us, the lesson is, okay, there is no moat anymore from what we've built. There is only a moat if we're able to deliver the most bleeding edge capability. Yeah. How do you think about.

After the memory shortage takes hold and whatnot. But lots of people are doing this. Alex Cohen breaks it down for Gen Z if you're wondering what happened today, Claude was mogging OpenAI for weeks. Then this Jim Cell dev ships Claudebot which was the fastest growing open source thing ever. Absolute looks max for the whole ecosystem. Anthropic tries to dairy goon him with legal Dev renames to OpenClaw. OpenAI slides in like a void pulling chad with acquisition interest. Openclaw gets acquired by OpenAI now Anthropic is getting jester gooned by the entire and OpenAI is giga maxing off their fumble open Anthropic could have just let them cook him cook. Instead they went full moid and got out framed by the Jester Maxers at OpenAI 8000 likes the looks max lingo is really, it feels like hilarious. I do wonder the half life I feel like it's I think it's over. It's got to be it's got to be towards the end of this boom. But the rise of of the kick streamers is certainly the story of the year. Certainly the story of the year. Dave Morin shared more context on what the future of openclaw will look like. So Dave Morin, who you might know from offline, ventures slow. He's been on our show path.

Willing to spend. We're often the largest spend in someone's stack. Wow. Talk to me about how the analytics are changing in the age of AI. Are you rolling out tools that will build an IPython notebook, write pandas, do SQL queries? I imagine that the previous era was very much like, great, you got the data together, now hand it over to me and my data scientist will write a bunch of Python to actually interpret it. How much of that is changing and happening in house now? Oh, it's completely changing. So the key issue is that you had to be familiar with SQL, you had to be familiar with data taxonomies. You have to be familiar with how to do a funnel analysis, like all these pieces of specialized knowledge. And so today we launched an AI agentic analytics platform with amplitude, where you can just type any free form question and then get an answer back. And it'll do the deep work of figuring out how to translate it to a query, how to use the right tools, how to map it to your taxonomy. So you don't need to know any of that. You can just say, what are my daus? Or what's the worst step in my conversion funnel? And then how do I improve it? It'll come back with a whole notebook and analysis of results, all with an amplitude. You can also connect it via Slack. So you can just connect our Slack bot, have it hit our MCP server, and then it comes back with this very rich dashboard and widgets and text responses that's formatted. That tells you all about it. Yeah, talk about the saspocalypse.

We'll talk to you soon. Have the rest of your day. Congrats on the progress. Goodbye. Let me tell you about Sentry. Sentry shows developers what's broken and helps them fix it fast. That's why 150,000 organizations use it to keep their apps working. Let's have Ankur from Brain Trust. Brain Trust. He's in the restream. Let's bring him in. I was chatting with him earlier. How are you doing? What's happening? What's up, guys? Not too much. First time on the show. Quick introduction. I want to hear the news and then I want to talk about the economics of AI labs, but I want to talk about your news first. So please introduce yourself. Awesome. Yeah, thanks so much for having me. I'm anker, founder and CEO of BrainTrust. We build observability for AI products. So companies like, you know, our Overlord or Overlord here, Ramp Notion, Instacart, Dropbox, et cetera, that are building really great AI products, they all use BrainTrust to help with that. We're announcing our Series B today. We just raised $80 million and helps us build. Sorry to interrupt. Congratulation. We're very excited. You said that helps you build. Yeah, it helps us ship more stuff and hire more people. Yeah, very simple. So, yeah, walk me through why someone's picking you, how they're integrating you, what the business model looks like, and what the best case scenario of AI observability looks like. Yeah, I think what's interesting about AI is when you build a product, you actually have no idea what's going to happen when you ship it. So if you're building a traditional ui, you can pretend to be Steve Jobs, look at the ui, play with it, form an opinion, and then you can kind of guess what's going to happen when people actually use it. With AI, you have literally no idea what's going to happen. And so being able to look at how people actually use your product and then capture the cases where it works well and the cases where it doesn't and test them is critically important. And that's basically the data flywheel that we power for all these companies. And then you have. Yeah, digging in a little bit deeper, you have like, same application, different models integrating as well that I imagine you're also comparing. So, like, what does best in class look like? If you're a software company that's integrating.

Hi. Let me tell you about Sentry. Sentry shows developers what's broken and helps them fix it fast. That's why 150,000 organizations use it to keep their apps working. Let's have Ankur from. Brain Trust. Brain Trust, he's in the Restream. Let's bring him in. Chatting with him earlier. How are you doing? What's happening? What's up, guys? Not too much. First time on the show. Quick introduction. I want to hear the news and then I want to talk about the economics of AI labs. But I want to talk about your news first. So please introduce yourself. Awesome. Yeah, thanks so much for having me. I'm Ankur, founder and CEO of braintrust. We build observability for AI products. So companies like our overlord here, Ramp notion, Instacart, Dropbox, etc. That are building really great AI products, they all use Braintrust to help with that. We're announcing our Series B today. We just raised $80 million and helps us build. Sorry to interrupt. Congratulations. We're very excited. You. You said that helps you build? Yeah, it helps us ship more stuff and hire more people. Yeah. Very simple. So, yeah, walk me through why someone's picking you, how they're integrating you, what the business model looks like, and what the best case scenario of AI observability looks like. Yeah. I think what's interesting about AI is when you build a product, you actually have no idea what's going to happen when you ship it. So if. You'Re building a traditional ui, you can pretend to be Steve Jobs, look at the ui, play with it, form an opinion, and then you can kind of guess what's going to happen when people actually use it. With AI, you have literally no idea what's going to happen. And so being able to look at how people actually use your.

Keep their apps working. Let's have Ankur from. BrainTrust Brain Trust. He's in the restream. Let's bring him in. Just chatting with him earlier. How are you doing? What's happening? What's up, guys? Not too much. First time on the show. Quick introduction. I want to hear the news and then I want to talk about the economics of AI labs. But I want to talk about your news first. So please introduce yourself. Awesome. Yeah, thanks so much for having me. I'm Ankur, founder and CEO of braintrust. We build observability for AI products. So companies like, you know, our Overlord, or Overlord here, Ramp notion, Instacart, Dropbox, etc. That are building really great AI products, they all use Braintrust to help with that. We're announcing our Series B today. We just raised $80 million and helps us build. Sorry to interrupt. Congratulations. We're very excited. You said that helps. That you said that helps you build. Yeah, it helps us ship more stuff and hire more people. Yeah, Very. Simp. So, yeah, walk me through why someone's picking you, how they're integrating you, what the business model looks like, and what the best case scenario of AI observability looks like. Yeah, I think what's going on.

By mid March, I really wouldn't be surprised. Yeah, sounds right. I want to, like, where this goes from a Just like the cynical machine that takes place and, like, turns this into a repeatable formula. Is there. Do you think there will be a new movement where a mainstream big artist hires a comedian to write some lyrics and then they do specifically architect the virality, where they have some UGC creator with a guitar doing an acoustic version on day one, and they hire that guy to do the dance outside the car dealership. And so it's like, much more orchestrated to, like, create that viral boom. What. I would say is things like that. Are already happening. Like, I know people in. Los Angeles. It may not be the literal. The exact literal thing you just said, but I know people in Los Angeles who work in production and songwriting. A lot of songwriting rooms are three or four people. Oftentimes you'll have one of those people who have fluency in viral content or fluency and memeable Internet content. And the gap between, like, I'm a comedian, I'm a content creator, I'm a musician. We used to think of those as three different jobs. Yeah, they're not three different jobs anymore. It's basically just one job, which is getting attention. That's the main job. So I think what you'll see is actually there are a lot of people even who bill themselves as songwriters, but who have been posting TikTok content for 4, 5, 6 years already in their 20 years old. Are they a songwriter or are they a TikTok maker? You know, are they engineers of IR? Are they recipients of virality, or are they engineers of virality? I just don't think those lines are as meaningful as they used to be. But, yeah, things like what you're saying, they're definitely already happening. What are the.

Cheaper and extremely scalable. There is another fun clip that we should watch from A Beautiful Mind. Jordi, have you seen A Beautiful Mind? It won the Oscar for best picture. I believe it's about the mathematician John Nash. Have you seen A Beautiful Mind? I've not. Wow. Unk status over there. You would have. I was walking on the beach with Senra on, I think, Saturday, and we walked by an incredibly famous. One of the top movie directors of the last probably 10 years. Really? And Sunra was like, do you see that? And I was like, see what? It's a guy with a dog. That's hilarious. I feel like that your beach tours really star studded lately. This is a different from the previous one you mentioned, correct? Yes. Wow. Yes. That's remarkable. Well, let's pull up the clip from. Incoming, gentlemen. It's from A Beautiful Mice. Nash, you might want to stop shuffling your papers for five seconds. Is that Eric Lyman? Yes, it's Eric Lyman, gentlemen, in the. In the ramp biopic. We got. We got. We got our cast right here. Oh, this is the original. Like, looks maximum. Anyone else feels she should be moving in slow motion? Will she want a large wedding, you think? Should we say swords, gentlemen? Pistols at dawn? Have you remembered nothing? Recall the lessons of Adam Smith, the father of modern economics. In competition, individual ambition serves the common good. Exactly. Every man for himself, gentlemen. And those who strike out or stuck with their friends. I'm not gonna strike out. You can lead a blonde to water, but you can't make a drink. I don't think you said that. All right, nobody move. She's looking over here, right? She's looking at Nash. Oh, God. All right, he may have the upper hand now, but wait until he opens his mouth. I think this. This is very, very stylized. What are you talking about? And completely apocryphal. Like, he definitely thought of this theory, but not at a bar. We block each other. Not a single one of us is gonna get her. So then we go for her friends, but they will all give us the cold shoulder because nobody likes to be second choice. What if no one goes for the blonde? We don't get in each other's way, and we don't insult the other girls. That's the only way we win. That's the only way we all get laid. So he's describing the prisoner's dilemma, Adam Smith said, where everyone must work together. This result comes from everyone in the group doing what's best for himself. Right? That's what he said, right? Incomplete. Incomplete. Incomplete. Okay, incomplete. Because the best result would come from everyone in the group doing what's best for himself and the group. Ash, this is some way for you to get the blonde on your own. You can go to hell. Governing dynamics, gentlemen. Governing dynamics. Adam Smith, what's wrong with you? Yep, there we go. Careful, careful. Thank you. Tis so good. Anyway, very fun. Dave says just joined the stream. We watching a movie. Yeah, movie day. Movie day. We should bring back. Bring back movie days in school. Yes. I mean, that was iconic. Rainy day substitution Gladiator in Latin class. That's the best. Really Quickly, let me tell you about Vibe co. We're DTC brands, B2B startups and AI companies. Advertisements.

But no one's just like, here's my green screen test. And they're like, okay, awesome. We know green screen works. We know it's a thing. It happens. Yeah. When I see the sea dance content, I think, okay, if movie budgets stay the same, I think we can potentially get a lot more great movies. And that's specifically because how much of these budgets goes to. Let's say somebody's making, like, a movie, like 300. And we need to get, you know, a thousand people out in a field. 300 people, but, yeah, who's counting? Who's counting? We need to get a lot of people out in a field and spend days, weeks, you know, recreating all these different scenes. Yeah. And now that budget can go towards. Yeah. The famous Henry Cowell. Superman. They finished shooting Superman. He moved on to, I think, a man from uncle. He grew a mustache. And they were like, we gotta shoot another Superman scene. And he's like, contractually, I cannot shave my mustache. And they were like, fine, we'll do it in post. And they have him come on and they CGI'd like a shaved lip on the top of his mustache. And it looked really bad. And it was like. People called it. Out. I actually think a lot of people saw the movie and they were like, eh, it's fine. It's pretty quick. But the VFX artists were like, this is bad. But with AI, obviously that's much, much more doable these days. So anyway, console. Console builds AI agents that automate 70% of it. HR finance support, giving employees instant.

Like two beautiful paintings that, like, the dog and the artist sort of collabed on, which is kind of cool. Yeah. Gavin purcell says if ByteDance doesn't restrict or nerf Sea Dance, this is going to escalate to a political conversation. Unlike the Sora 2 launch, because OpenAI restricted likeness ahead of time. Sea Dance shows just how clearly every major movie star is generatable by AI. So this is interesting. If you're already like an A list, massive superstar, I think you see some stuff like this and you're actually like, great, I'm gonna be able to shoot a movie in a week from la. I'm not gonna have to travel to these insane exotic locations and spend a week in the desert filming all these clips. So if you're like a Timothee Chalamet, this is maybe like, yes, you're worried for the overall industry, but at the same time you're thinking, okay, my name and likeness is now infinitely scalable. I can still like, restrict the supply to some degree. Right? I'm not going to tell any, any movie studio, hey, you can make a movie with me, whatever. You're still going to kind of restrict it and have pricing power. But if you're a, the question becomes new, new talent that's emerging, trying to build their brand. At what point do studios say, we're just going to make a character, we're going to make a new actor out of thin air, place them across different movies, build them up over time. You could imagine, I don't think a company like CAA would do this because all their talent would be like, what are you doing? Like, you're taking our job. But I could imagine a group trying to make like a little Mikayla style actor that you build up over time. One thing that we'll find out is how much does the actual actor's real life matter in the context of their career? Like, if Timothee Chalamet is dating Kylie Jenner, does that like increase his appeal on the big screen? Yeah. And I would say yes, probably. Right? There's so much fixation on the lives of all this talent. You could sort of fake a decent amount of that. You could fake paparazzi photos, you could fake, you know, vacation photos. You could create a whole world around a fake person potentially. But if people find out, will that break the illusion? Will they be not into it? But it's not real drama. People like the drama. They like following stars. I'm still just pretty optimistic about, like, AI as a tool in Hollywood. Did you see that? Fully, fully AI generated movie about the woman who goes in the cybertruck that went viral this weekend. Wasn't it like, only three minutes, I think? Yeah, exactly. It was only three minutes, but it was consistent from one shot to another, which is impressive. And somebody said not enough people are talking about how much this sucks. And Ben.

Cloud. So anyway, fun to dig into some little Econ 101 or 102 depending, but let's go to the timeline. Buco Capital says I thought Dwarkesh had a good point that software engineering is the only job where the full context needed to do the job is available to an AI agent via the code base. And I didn't think Dari had a good answer for why automating other jobs will be as easy. This got a bunch of a lot of people kind of reacting, kind of disagreeing generally that all of the full context needed to do the job is available. But I do think it's something we need to figure out. Yeah, we were debating this because there was a post that was just sort of like a wojack reaction that was just making fun of this and it wasn't clear if they were saying that they were agreeing or disagreeing. But basically my take was well, it's possible that a lot of the full context needed to do the job of a lot of different white collar jobs is in fact logged. It's just logged in the final product, which is like a deck or a spreadsheet or a decision, and then a whole bunch of emails, a whole bunch of slacks and then a whole bunch of zoom calls that's recorded. And so yes, if you're running a business where a lot of work gets done in smoky bars late at night and back alley deal making sure that's going to be harder to automate. But in the world where it's someone sitting in front of a computer and there's a screen recorder running, you should be able to pull up most of the context at the same time. You can't just snap your fingers and go back and get every decision that was made in the 80s that allowed coca Cola to become a dominant soda maker. But you can with Linux. You literally can with Linux. People on calls just being like knowing the calls being recorded and used to train something to replace them, they're just like I'll tell you offline. I'm not in speak speaking this secret. In record golf this weekend. You want to play golf really quickly? Bigma ship the big.

Triple legs. Let's just roll, right? Market clearing order inbound. You're surrounded by journalists. Thank you. Strike1. Strike 2. Activate Go retriever mode. Market clearing order inbound. Five clear. You're watching TVPN. Today is Tuesday, February 17, 2026. We are live from the TVPN Ultradome. The temple of technology, the fortress of finance, the capital of capital. That's right. Let me tell you about ramp.com time is money save. Both easy use, corporate cards, bill pay, accounting, and a whole lot more all in one place. It is the year of the year. It's the year of the Fire Horse. And we get to celebrate it twice. Because I think we. I remember we talked about it at the beginning of the year. But the Chinese New Year did not start until today. It's Lunar New Year. Right. Worshipers burned large incense sticks on Monday outside a temple in Hong Kong to mark the Lunar New Year. There we go. Which falls on Tuesday. People around Asia celebrate the start of. The Year of the Horse in the Ultradome. It's the year of the Horse. Every year. Every year. I think so. It is. It is the year of the Horse. Look at this. We were early. We were early and right on horses. Well, everyone is talking about the Cournot Equilibrium. At least Dario Amade and Dwarkesh Patel are you and me and some folks on the timeline. We're going back and forth and basically trying to get to this question. I feel like we should give the context on this, on the titling strategy, because we call the run of shit. We'll title an essay. Why is no one talking about the Cournot Equilibrium? Because this one guy had this super viral like a year or two ago and he's. And the title was why is no one talking about Marc Andreessen? We were laughing about it so much. Because he's one of the most talked about. Oh yeah. Investors in venture. He's ever minus list constantly. He's written essays. Everyone's talking about, like been viral a million. And he's someone. So many podcasts. Yeah. He's someone that everyone in the industry has an opinion on already. Totally. He's not like a Midas lister. Yeah, yeah. Up there. Not a lot of people. It's like everyone's talking about. But in this case, really, I mean, truly, I do think more people should be talking about the Cournot Equilibrium or at least. Years ago. His name's Antoine Cournot. Idea is that if there's only a few players in a given market, you can think about any Specific market lemonade stands or whatever. And they aren't competing on price, they will compete on supply and they'll try to predict what their competitors are doing and then respond accordingly. And this is really, really relevant to the AI lab discussion because you can tell that even though all the leaders of the AI lab say I don't think about the competition, I don't talk about the competition, all use general terms, they're all obsessed with what everyone else is doing and they think about it constantly, very clearly. And if someone's buy 10 billion of compute over here, they're gonna counter with 8 over there, try and jump to 12 and everyone's sort of keying off of each other. Microsoft pauses, AWS goes all in. There's all these horse races. It's why semianalysis exists and provides great cross functional data. Daniel in the chat says state actors at work again. Apparently we're having technical difficulties today. We're back, we're back in every way, but we're working on it. Okay. And so yeah, there's this big question outside of tech. There's this discussion that I see that's always funny to me where people would be like, oh, the price to earnings ratio. Ryan says they don't want you to talk about the price. They don't want us to talk about it. They don't want us to talk about it for sure. I like the Chiron too. AI labs locked in corno battle and the so outside of tech there's this discussion. I saw Emerson say the price to earnings ratio for OpenAI and Anthropic is just simply too high. And I was like earnings. These companies are losing money. They don't have a price to earnings ratio. Divide by zero. This is going to blow your mind. Yeah, it's so much worse than you think. They're not making any money. They're deeply unprofitable. These are the most unprofitable companies in human history, I think. But at the same time there is an economic rationality behind all of this. And what is that economic rationality? Well, when you're running a AI lab, you actually have two businesses that are sort of hiding within the P and L. And so the first you, you know, you share this brand, you share the data centers, you share compute. But there's a whole bunch of risks and rewards to the various pieces of the business. So first you have training the models, so you make an upfront investment in a training run that creates a particular model generation and then that asset depreciates as the frontier moves and then when a new and better model is released, everyone moves over and you stop reaping the benefits of that and the value of that model goes much, much lower. How much money is OpenAI making from GPT 3 point. Zero. Spent example? Because I believe that model costly. Your variable costs so GPUs, power, engineering overhead, and then your revenue is subscriptions, API usage and enterprise contracts. And so when you just look at inference, you see positive contribution margin. And we can see that because we can compare the cost to inference. A model of the GPT5 class size or the Opus 4.5 size, you can see what does it look like to run an open source version of that model on commodity hardware? Well, it's way, way cheaper than what you pay to Anthropic or OpenAI. So they must have good margins. And everyone sort of agrees this point, that inference margins are in fact healthy. The question is how do you balance those two pieces and when do you risk over investing? And that's sort of this Cournot game of chicken that everyone's playing now. The Cournot equilibrium comes when a small number of labs, an oligopoly, effectively choose supply at the frontier level and then the market clears at a high price for frontier access. So choosing supply in this case means how many data centers get built, how many GPUs get ordered, but also how much low latency capacity is allocated to the top tier. So right now OpenAI just did the Cerebras deal. There's Claude Fast and there's a whole bunch of different modes that will deliver faster inference and how many of those fast queries you get. How much of the best chips are allocated to a particular tier that you're paying for is an economic question for the labs. So on the true frontier there aren't great substitutes and so price stays high based on customers willingness to pay for frontier access. So you can just think about it in more simpler terms like there's a ton of developers and knowledge workers who are happy to pay hundreds of dollars a month or more, but they always want the best available model. This is most people in executive roles in startups, right? It's like, yeah, I got my $200 a month subscription, I'll pay 250 or 100 or whatever, couple hundred bucks and it just makes me better at my job. I just do whatever I need to do. But don't give me the old thing. I want the best. I want to know that the hallucination rate is as low as possible. I want to know that when it builds me an economic model, it's doing it as best as it possibly can. So I need to spend less time. 1% of the time it makes a career ending mistake. Maybe. I. No, I really do. I mean, there are some crazy, crazy, like, possible outcomes. We haven't heard too many nightmare stories, but they certainly have to be out there. Like the true hallucinations. You need checks and balances. Yeah, yeah. But increasingly. Yeah, increasingly less. Everyone's gone through the experience of somebody building a model. Yeah. And then being like, I don't know why, but it's wrong. Yeah. Yeah. And then it's usually a battle. Yeah. I was, I was actually. You got to train that instinct that can just clock this even if you have. Have no, if it just like, if it feels off. I was talking to a buddy about using AI tools who works in like high yield debt, sales and trading and issuance. And he was like, yeah, like the models just like, aren't that good. And I was like, really? And he was like, yeah, they don't understand like Q4. 76 of the tax code and how it applies. And I was like, okay, yeah, actually that seems like something that they might not be great at. I was like. Because I built a toy model which just like project out the cash flows, discount them back, and it did a great job. And I was like, cool. And he was like, no, I'm like 25 levels above that and like deep in all these different codes and like, it's not there. Good work, little bro. It'll get there. But it's just like, you know, the base models are not just going to be able to like one shot that. What do you think, Tyler? Fire it back. You would imagine that like something as like if it's just like text code, even if it's super complex, you can just put that in the context and it just does a very good job. Yes, but there's like finding specific facts from a super big document is like, that's one of the things that models are best at. Yes, but there's. There's a lot of. There's a lot of. You're laughing at the chat. Ethan says computer fix tvpn use Gunna and Grok. Gunna could fix us. We're back. We're back. Yes, but there are nuances to the tax code and to a lot of legal documents that. Where the law says one thing, but it's implemented in a different way or it's not enforced, or some regulatory organization has enforcement discretion or there's some understanding Even if you just go into like training on, you know, ip, it's like if you have really good lawyers, you can figure out, okay, well they'll come to the table, we'll be able to negotiate this, this will be the price, this will be the settlement. So even though the model might say like, don't do that, like you have a chance to do it, there's a whole bunch of different reasons why. But yes, I generally agree with you. This is like a six month away thing. But it's just like the current literature. It's the current thing, but with like good, you know. Yes. And maybe some more harnessing and more, some and better context. Anyway, the question is like, where does this all go? What is the natural state of equilibrium in the longer term? Right now they're locked in Cournot where there's a whole bunch of companies, individuals that want to buy the best AI inference possible. They want the best tokens, they want the best deep research reports, they want the best code and they're willing to pay for it. But we're supply constrained. And so OpenAI brings a certain amount of GPUs and tokens and supply to the market. And then Anthropic also does. And the whole Cornot thing is that they're all keying off of each other and saying, okay, if they're offering this much, I'm going to offer that much, to be clear. So having a product, not just an API business, gives you leverage because at some point the models are smart enough where you don't need to train them, you don't need to train a model that is 4% better because people are still coming to your application and having a good product experience, right? Yes. So historically one of the critiques to Anthropic's business was that they have to just be on this constant, constant fly sort of hamster wheel of training the best model. Because the majority of their business is this API business, they're not an aggregator yet swap it out for a smarter model. That said, they have Claude code now, which gives them some more leverage over the market. And the really interesting thing is that Dario is now talking about being near the end of the exponential or maybe producing like the final models because we've talked to a few people about this, but it's very unclear if it's possible to create like a super intelligence that's like 5,000 IQ. It might just be they get good at all knowledge work and they can answer all tasks, but it's like the digital guy. And so at that point, it does commoditize and you drop out of Cournot equilibrium and you become more. Customers are more aggressive about switching to cheaper models to cut costs because the frontier is now commoditized in the entire backlog. Everyone is at the frontier, basically. And so in that scenario, you switch over to Bertrand competition, which doesn't really mean that profits go to zero, but there is more competition and it looks a lot more like the hyperscaler cloud market, which is, I think, what people have been sort of signaling towards. And also it sort of explains why a lot of the VC firms are getting in multiple companies because they don't think it's going to be winner take all anymore. They think it's going to be much more oligopolistic for the long term. And there will be competition between the major three or four labs that will. And it will be much more about how can you marshal enough supply, create a huge barrier to entry. Like you and I could start an AWS competitor tomorrow, but it's going to be extremely expensive to bring up data centers that just serve web apps everywhere, let alone AI stuff, right? Building all those data centers. You're thinking what I'm thinking. You're thinking AWS people are saying, when does TPW go? I was hanging with my buddy Ben on Sunday and he was, we both live in Malibu. He was thinking of just getting some chips and setting up Malibu Inference. There we go. Just the name alone sounds like you could get at least, at least one Malibu Inference. That's really funny. Would be a beautiful name for a NEO cloud. Yeah, it's funny. And so there is a question about in the final models, will there be differentiation? What does differentiation look like? There's a potential for differentiation around. Okay, you get a model that's really good at bio or math or engineering or code or writing. That's possible. Another possibility is that there's just differentiation on on quality, latency, safety, enterprise tooling and you wind up with a small number of vertically integrated firms. And so in that scenario, I feel like OpenAI's recent moves make a lot of sense. Cerebras offers differentiated product around speed Peter from OpenClaw joining makes a lot of sense around building orchestration products that can route to particular models. So the next generation of the Claude code router that can route you to specific APIs and act as a model router on top for the agentic programmer AI assistant use case you already see the model router happening in the ChatGPT consumer app. We haven't really Seen that in the desktop CLI app. And so there's something interesting there. And then frontier. OpenAI frontier, that for deployed engineer thing that digs deeper into the enterprise and probably gets them, you know, more hooks in there. More, more pricing power. The chat's going off today, Ryan says the Cournot Company of Malibu. Let's do it. I love it. Let me tell you about CrowdStrike. Your business is AI. Their business is securing it. CrowdStrike secures AI and stops breaches. Let's. Let's play the clip of Dwarkesh Patel and Dario Amadei discussing the economics of AI labs because I think it's informative. And this is where the whole Cornot current thing came from, because Dario dropped that phrase. So when you said, why is no one talking about it? It was literally everyone that listened to the biggest podcast of the week. But I'm explaining it. Let's play it industry here. Like, we have a, you know, let's just imagine we're in like an economics textbook. We have a small number of firms. Each can invest a limited amount in, you know, or, or like each can invest some fraction, fraction in R and D. They have some marginal cost to serve the margins on that, the profit margin, the gross profit margins on that marginal cost are like very high. Because, because, because inference is efficient. There's some competition, but the models are also differentiated. There's some, there's some, you know, companies will compete to push their research budgets up. But like, because there's a small number of players, you know, we have the, what is it called? The Cournot equilibrium, I think is what the firm equilibrium is. The point is it doesn't equilibrate to perfect competition with zero margins. If there's like three firms, if there's three firms in the economy, all are kind of independently behaving rationally. It doesn't equilibrate to zero. Help me understand that, because right now. We do have three leading firms and they're not making profit. And so. That's a good question. Yeah. What is changing? Yeah. So again, the gross margins right now are very positive. What's happening is a combination of two things. One is we're still in the exponential scale up phase of compute. So basically what that means is we're training. Like a model gets trained, it costs, let's say a model got trained, that costs a billion dollars last year, and then this year it produced $4 billion of revenue and cost $1 billion to inference from. So, you know, again, I'm using stylized number Here. But you know, 75. These are my numbers. I'm just picking random numbers. Gross margins and you know, this 25% tax. So that model as a whole makes $2 billion, but at the same time, we're spending $10 billion to train the next model because there's an exponential scale up. And so the company loses money. Each model makes money, but the company loses money. The equilibrium I'm talking about is an equilibrium where we have the country of geniuses. We have the country of geniuses in the data center. But that model training scale up has equilibrated more. Maybe it's still going up. We're still trying to predict the demand, but it's more leveled out. Let me tell you about turbopuffer, serverless vector and full text search built from first principles and Object storage. Fast 10x cheaper and extremely scalable. There is another fun clip that we should watch from A Beautiful Mind. Jordy, have you seen A Beautiful Mind? It won the Oscar for best picture. I believe it's about the mathematician John Nash. Have you seen A Beautiful Mind? I have not. Wow. Unk status over there. You would have. You would have. I was, I was walking on the beach with Senra on I think, Saturday, and we walked by an incredibly famous. One of the top movie directors of the last probably 10 years. Really? And Senra was like, do you see that? And I was like, see what? It's a guy with a dog. That's hilarious. I feel like that your beach tours has been really star studded lately. This is a different from the previous one you mentioned, correct? Yes. Wow. Yes. That's remarkable. Well, let's pull up the clip from. Incoming. Gentlemen, This room of beautiful mice. Nash, you might want to stop shuffling your papers for five seconds. Is that Eric Lyman? Yes, it's Eric Lyman in the, in the ramp biopic. We got, we got. We got our cast right here. Oh, this is the original. Like looks maxing movie should be moving in slow motion. Well, she want a large wedding, you think? Should we say swords, gentlemen? Pistols at dawn? Have you remembered nothing? Recall the lessons of Adam Smith, the father of modern economics. In. In competition, individual ambition serves the common good. Exactly. Every man for himself, gentlemen. And those who strike out are stuck with their friends. I'm not gonna strike out. You can lead a blonde to water, but you can't make a drink. I don't think he said that. All right, nobody move. She's looking over him, right? She's looking at Nash. Oh, God. All right, he may have the upper hand now, but Wait until he opens his mouth. Remember the last episode? That was in the history book. I think this. This is very, very stylized and completely apocryphal. Like, he definitely thought of this theory, but not at a bar. We block each other. Not a single one of us is going to get her. So then we go for her friends, but they will all give us the cold shoulder because nobody likes to be second choice. What if no one goes for the blood? We don't get in each other's way, and we don't insult the other girls. That's the only way we win. That's the only way we all get laid. So he's describing the prisoner's dilemma, Adam Smith said, where everyone must work together. This was all contradicted from everyone in the group doing what's best for himself. Right? That's what he said. Incomplete. Incomplete. Incomplete. Okay, Incomplete. Because the best result would come from everyone in the group doing what's best for himself and the group. Ashley, this is some way for you to get the blonde on your own. You can go to hell. Governing dynamics, gentlemen. Governing dynamics. Adam Smith. Who's wrong? There we go. Careful, careful. Thank you. Tis so good. Anyway, very fun. Dave says just joined the stream. We watching a movie. Yeah, movie day. We should bring back. Bring back movie days in school. Yes. I mean, that was iconic. Rainy day substitute. Watching Gladiator in Latin class. That's the best. Really quickly, let me tell you about Vibe Co, where DTC brands, B2B startups, and AI companies advertise on streaming TV, pick channels, target audience, and measure sales. Just like on Meta. Anyway, very. Yeah, very fun. Nash. Equilibrium. Lots of game theory going on in the AI wars right now. Everyone's trying to figure out how far to push it. There's a fair amount of risk. There's still the Cournot game of chicken around. Who will invest the most in advancing the frontier. But the end state looks a lot more durable than pure model commoditization and the perfectly competitive situation that many were predicting a few years ago. So if you go Back to like 2023, a lot of people were saying, like, the models will commoditize and there will be no value there, like, no profits, because, like, the deep seq moment. And it'll all be run on commodity hardware. That doesn't seem that likely. It seems like the labs will turn into sort of new hyperscalers. There will be, you know, increased competition, but still very, very good businesses a la cloud. So, anyway, fun to dig into some little econ 101 or 102 depending. But let's go to the timeline. Buco Capital says, I thought Dwarkesh had a good point that software engineering is the only job where the full context needed to do the job is available to an AI agent via the code base. And I didn't think Daria had a good answer for why automating other jobs will be as easy. This got a bunch of a lot of people kind of reacting, kind of disagreeing generally that all of the full context needed to do the job is available. But I do think it's something we need to figure out. Yeah, we were debating this because there was a post that was just sort of like a wojack reaction that was just making fun of this and it wasn't clear if they were saying that they were agreeing or disagreeing. But basically my take was, well, it's possible that a lot of the full context needed to do the job of a lot of different white collar jobs is in fact logged. It's just logged in the final product, which is like a deck or spreadsheet or a decision and then a whole bunch of emails, a whole bunch of slacks and then a whole bunch of zoom calls that's recorded. And so yes, if you're running a business where a lot of work gets done in smoky bars late at night and you know, back alley deal making sure that's going to be harder to automate. But in the world where it's someone sitting in front of a computer and there's a screen recorder running, like, you should be able to pull up most of the context at the same time. You can't just snap your fingers and go back and get every decision that was made in the 80s that allowed coca Cola to become a dominant soda maker. But you can, you literally can with Linux on calls just being like knowing the calls being recorded and used to train something to replace them, they're just like, I'll tell you offline, I'm not in speak speaking this secret. Yeah, golf this weekend. You want to play golf really quickly. Figma ship the best version, not the first one. With FIGMA introducing Claude code to Figma. Explore more options, push ideas further. Figma but this was funny. Bucho said. Basically a non zero chance of Joe Wiesenthal victory where software engineers write themselves out of a job first and everyone else has the boring parts of their job automated. Joe of course was joking, just saying, well, of course software engineering is being automated. So easy. It's so easy. Yes, the debate around the posture. Dwarkesh Dwarkesh his posture was. He's been at the gym a lot. He looks fantastic. And I love this sweater. The crew neck works really well. The pushed up sleeves is a particular choice. Didn't translate into that. Chad Wojack. But he looks fantastic here. And the Aeron chair is certainly helping with the posture. A lot of fun on the timeline, looking at the. The looks mogging or whatever. The looks maxing. I don't even know. Frame mogging. That's the frame. He kept bringing up the example of a video editor saying, yeah, but when will the models be good enough to edit videos? Well, and pick out moments. Yes. And give me two years and another 500 billion. This is. I mean, we've tried every tool there is. Yeah. They can't do it yet. Tricky. I don't know, I don't know what's. And it's not even that we're not trying the tools to replace the people on our team. We're trying to make them have higher output. Yeah. But to date, I don't know if. There'S something super sticky. I mean, it does seem like. One interesting thing is that there aren't a lot of open source, like premiere profiles. Like, I've edited a ton of videos for YouTube. There's a whole bunch of cuts in there. What I cut out, what I didn't. You could have that record, but it's not stored in GitHub. Like, it's just. You can't necessarily train on it. You can train on the final product and understand, but you don't understand what actually got left on the cutting. Cutting room floor. There's this whole concept of like, kill your darlings. Like when you're in the edit, like, you need to be cutting more. You're like, I like that shot. It's so cinematic, so cool. But does it actually advance the story? No. So you cut it down. I was watching the Matrix this weekend and there's this amazing shot of when Neo and Morpheus are going to visit the oracle. Is that one the oracle And Neo. Is that the one oracle? Yeah. And they reach for the doorknob and the doorknob has this perfect reflection and the reflection shows Neo and Morpheus. And they had to do this crazy VFX shot to hide the camera in what looks like Morpheus's coat. Because if you point a camera at a mirror, you see the camera and you don't want to see the cameraman there. That ruins the shot. And so they did all this crazy stuff to like, to like, you know, cover up the camera and I'd seen the behind the scenes and been like, wow, that's really impressive. And in my memory I thought it was like, oh, it's such an important shot. They probably lingered on that for like five seconds to really let it sink in. Like they're pulling a trick on the audience. It's beautiful. It's like half a second and they did all this work and then they knew that like, from a story perspective, from a storytelling perspective, you don't want to hang out and watch a picture of a doorknob for five seconds. And so all these decisions, like, they sort of get chronicled, but they don't get neatly organized in the way that a GitHub log does with. With pull request discussions and what happens. So it'll be. It'll be difficult. So maybe two years, another $5 billion does it. But it's coming, so we'll. We'll keep monitoring it. Let me tell you about the New York Stock Exchange. Want to change the world? Raise capital at the New York Stock Exchange. We have a special guest joining. John Caramenica is joining and I want to music. But I wanted to tee up his appearance with a little music review that I found in the newspaper today. The algorithm wants you to grieve. Rest in peace. My granny, she got hit by a bazooka. A title that functions as its own thesis statement is the kind of song that resists every framework we've built for understanding virality. It is not quite parody, not quite sincerity, not quite meme. It exists in some fourth space where emotional register becomes irrelevant and the only metric that matters is whether the thing lodges itself in your prefrontal cortex like a splinter. The melody, if we're being generous enough to call it that, is a three note loop that sounds like it was composed on a Fisher Price keyboard during a power outage. The vocal delivery is flat, almost affectless, a kind of anti performance that paradoxically demands more attention than virtuosity ever could. The lyrical content is, well, the title repeated with conviction. And yet something is happening here that's worth taking seriously, or at least worth resisting the urge to dismiss. The song collapses the distance between tragedy and comedy so completely that neither category survives the collision. It's grief as non sequitur, eulogy as punchline, memorial as munition. The bazooka isn't just absurd, it's so specifically absurd that it loops past irony and arrives somewhere strangely earnest. Nobody's grandmother has ever been hit by a bazooka. The scenario is impossible and impossibility it turns out, is its own form of tenderness, a way of saying loss is so incomprehensible that only nonsense can hold it. Or maybe it's just a guy yelling about a bazooka. The Internet doesn't require you to choose, but we'll be following this story, so thanks for. We'll get John's take on. We will. It's the song of the week and the ad platform of the week is Applovin. Profitable advertising made it Easy with Axon AI. Get access to over 1 billion daily active users and grow your business. Today, Andrew Reed says, horses don't stop, they keep going. Wait, did he actually say that? Yes. No way. Yes. In response to 2026 being the year of the horse. I love it. One of the greatest lyrics of all time. Originally to explain the joke. It's a Young Thug song and the actual lyric is hustlers don't stop, they keep going. But it sounds like horses. And so people put horses don't stop, they keep going. And they show the AI generated image of the horse bench pressing. And it's incredibly inspiring. There's a lot of Young Thug songs that are hard to fully decipher. 100%. Let's hit the size gong for this. Pennsylvania Girl Scout, 6 years old, breaks record selling 87,000 boxes of cookies. She's unstopp, unstoppable. How much is that? What's the ARR. How much is a break it down? Is it $12? That's a lot. How long does this. Six years old. Wow. What, what enabled this incredible scale? Was it potentially Shopify? Shopify is the commerce platform that grows with your business, lets you sell in seconds online, in store, on mobile, on social, on marketplaces. And now with AI estimating that it somewhere around $600,000 of sales at only six years old. Really incredible stuff. Heartwarming. That's awesome. India's Adani group to invest 100 billion in AI infrastructure. We gotta hit the gong again. Hit it again. Indian investment may boost the country's ambitions to become an AI power. India's Adani Group, an energy and logistics giant, said it would invest $100 billion to develop large scale data centers by 2035, the largest such commitment in India so far. Tyler, what do you think about the timing here? Is this going to be too late? Are the clankers going to like is 2035? How are we looking there? Is that, I mean, a singularity? Yeah, I'm very bullish on the clankers coming pretty early. So, you know, time will tell. I Guess we'll see. I cannot wait to pull up this clip. It is, it is a big number that I feel like a lot of countries have been teasing big numbers, but. This is, they're kind of mogging. Macron. Yeah, this is like a really big number. You see a bunch of like multibillion dollar deals, multibillion dollar releases. But this is like serious, serious, serious investment. So you know, good, good news for many countries. One of the biggest concerns about the looming AI era is that it will deepen existing technology divides in which tech services are developed by companies in a few countries. Some countries have also expressed concerns about how large tech firms will come consume data generated by their citizens to build their AI products. India is focused on making sure that startups and researchers use AI to solve pressing development challenges and paving the way for its businesses to become providers of global AI services. India will shape solutions not just for India but for the world, said Indian Prime Minister Modi in a social media post today. The Andani Group said the dedicated said the dedicated computing capacity it is building will support Indian large language models and ensure data generated in India is stored locally. Let's see how the stock's doing up 2% today. Is it a $28 billion company? Do I have that right? I don't know. It's like 2.5 trillion INR so I think that's about 28 billion. So this is a huge says Andani will more than likely use government leverage. Sure. Government backed data centers. You happy about that Tyler? Government backstop coming to India. Backstop. Let's go. Before we move on, let me tell you about Restream 1 livestream 30 plus destinations. If you want a multi stream go. To restream.com Micron is spending 200 billion so congratulations for saying the biggest number. Micron. Micron is spending 200 billion to break the AI memory bottleneck. For decades memory chips were low margin commodity products. Now the industry can't make enough to satisfy data centers hunger. Just this one company is like yeah, we're going to spend twice as much as India as the adani Group. Boise, Idaho. Each afternoon at around 4:30 the earth here shakes from a series of controlled explosions as engineers blast through basalt bedrock to flatten out the ground underneath a gigantic new semiconductor factory. Let's give it up for Idaho. Yeah, we don't talk about Idaho enough. Micron Technology is the largest American maker of memory chips. The tiny slices of silicon that store and transfer data and help power everything from smartphones and car computers to laptops and Data centers. Micron is rushing to add manufacturing capacity to avert the biggest supply crunch the memory industry has seen in more than 40 years. In Boise, where the company is based, Micron is spending 50 billion to more than double the size of its 450 acre campus, including the construction of two new chip factories. The first fabs inaugural wafers are expected to roll off the factory line in mid-2027. So this is their first fab or the one that they're spending the 50 billion on right now? I think the new one. Okay, so that's pretty quick. Yeah, yeah. Did you hear that the, the PS6, the PlayStation 6 is now delayed because of memory shortages. 2029, 2028, 20 or 29. It's still a rumor. Maybe they'll, you know, maybe this solves it and they can get it out earlier. But pretty, pretty big delay to 2029, they really don't refresh. You just created a trillion gamers. No, seriously. I think adding insult to injury to. The gamers might, actually gamers might rise up. They might be an important voting block. A lot of them are, A lot of them are of age to vote and a lot of them don't would rather have new gaming hardware than necessarily AI slopping the feed. They're like, yeah, I can't afford the new PC that I wanted. What do you think? Yeah, I don't know. I mean I feel like this says a lot about how good the PS5 is. Right. Because they can like afford to just like postpone the psx. Like what games? Oh, now you don't want technological progress. Wow. I wanted to go to the data centers. I don't, I don't care about like the next like game graphics. Have like, have they gotten that much better in the past, like five years? Like, yeah, maybe. But it's like. Yeah, for the actual gameplay, is it that important if like the actual pixels are. Yeah. Realistically, a lot of this stuff should be moving to the cloud soon if it's not already. Yeah, like it was the Meta Quest VR. I forget. Oh yeah. Xbox Connect, Xbox Edition. Yeah, yeah. And it was all in the cloud and it was like totally fine. It was low latency. Yeah. And then if you're running in the cloud, you can upgrade the hardware and in theory you should be able to run like a gen AI up, res in pass to make it more photo real. And I feel like that's going to be where more of the juice is squeezed out of the graphics than just continuing on the traditional path of more pixels. More ray tracing. It'll be make a really beautifully designed video game that works really well. Really tight deterministic interactions so it's satisfying and then give it a layer. We got to have a ramp trader on the show. Really. Somebody that's in the thick of deal making in the space. There's a couple folks over at Semianlysis that cover it. One of them just went on odd lots. Let's do it. Get him on the show. Let's do it. Let me tell you about graphite code review for the.

Five coded. I see multiple journalists on the horizon. Standby. Uav online. Double glaze. Triple glaze. Double kill. Five cup. Wrong. Is. Team Deathmatch. We are experts. Triple blaze. Let's just run. Right. Market clearing order inbound. Come, get up. We are surrounded by journalists. Hold your position. Strike1, Strike 2. Golden retriever mode. Market clearing order inbound. 5. You're watching TVPN. Today is Tuesday, February 17, 2026. We are live from the TVPN Ultra on the temple of technology, the fortress of finance, the capital of capital. That's right. Let me tell you about ramp.com, time is money save. Both easy use, corporate cards, bill pay, accounting and a whole lot more all in one place. It is the year of the fireworks. It's the year of the fire horse. And we get to celebrate it twice. Because I think we. I remember we talked about it at the beginning of the year. But the Chinese New Year did not start until today. It's Lunar New Year. Right. Worshipers burned large incense sticks on Monday outside a temple in Hong Kong to mark the Lunar New Year, which falls on Tuesday. People around Asia celebrate the start of. The Year of the Horse in the Ultra Dome. It's the year of the Horse. Every year. Every year. I think so. It is the year of the year. Look at this. Look at this. We were. We were early. We were early and right. Yes. On horses. Well, everyone is talking about the Cournot Equilibrium. At least Dario Amade and Dwarkesh Patel are you and me and some folks on the timeline, we're going back and forth and basically trying to get to this question. I feel like we should give the context on this, on the titling strategy. Because we call the run of shit. Like we'll title an essay. Why is no one talking about the Cournot Equilibrium? Because there's one guy had this super viral a year or two ago and he's in the title was. Why is no one talking about Marc Andreessen? We were laughing about it so much. Because he's one of the most talked about. Oh yeah. Investors in venture. He's a minus list constantly. He's written essays and everyone's talking about. Been viral million times. And he's someone so many. Yeah. He's someone that everyone in the industry has an opinion on already. Totally. He's not like a Midas lister. Yeah, yeah, yeah. Not a lot of people. It's like everyone's talking about. But in this case, really, I mean, truly, I. I do think more people should be talking about the Cournot equilibrium or at. Years ago. His name's Antoine Cournot. Idea is that if there's only a few players in a given market, you can think about any specific market, lemonade stands or whatever, and they aren't competing on price, they will compete on supply and they'll try to predict what their competitors are doing and then respond accordingly. And this is really, really relevant to the AI lab discussion because you can. But even though all the leaders of the AI lab say I don't think about the competition, I don't talk about the competition, I'll use general terms. They're all obsessed with what everyone else is doing and they think about it constantly, very clearly. And if someone's buying 10 billion of compute over here, they're gonna counter with 8 over there, try and jump to 12 and everyone's sort of keying off of each other. Microsoft pauses. AWS goes all in. There's all these horse races. It's why semi analysis exists and provides great, you know, cross functional data. Daniel in the chat says state actors at work again. Apparently we're having technical difficulties today. We're back, we're back in every way but we're working on it. Okay. And so yeah, there's this big question in outside of tech there's this discussion that I see that's always funny to me where people are be like, oh, the price to earnings ratio. Ryan says they don't want you to talk about the. They don't want us to talk about it. They don't want us to talk about it for sure. I like the chiron too. AI labs locked in cornot battle. So outside of tech there's this discussion. I saw the price to earnings ratio for OpenAI and Anthropic is just simply too high. And I was like, earnings. These companies are losing money. They don't have a price to earnings ratio. Divide by zero. This is going to blow your mind. Yeah, it's so much worse than you think. They're not making any money. They're deeply unprofitable. These are the most unprofitable companies in human history, I think. But at the same time there is an economic rationality behind all of this. And what is that economic rationality? Well, when you're running a AI lab, you actually have two businesses that are sort of hiding within the P and L. And so the first you, you know, you share this brand, you share the data centers, you share compute. But there's a whole bunch of risks and rewards to the various pieces of the business. So first you have training the models so you make an upfront investment in a training run that creates a particular model generation and then that asset depreciates as the frontier moves. And then when a new and better model is released, everyone moves over and you stop reaping the benefits of that and the value of that model goes much, much lower. How much money is OpenAI making from GPT 3.5 they. Spent example, because I believe that model costly. Your variable costs so gps, power, engineering overhead and then your revenue, subscriptions, API usage and enterprise contracts. And so when you just look at inference you see positive contribution margin. And we can see that because we can compare the cost to inference. A model of the GPT5 class size or the Opus 4.5 size, you can see what does it look like to run an open source version of that model on commodity hardware? Well, it's way, way cheaper than what you pay to Anthropic or OpenAI. So they must have good margins. And everyone sort of agrees at this point that inference margins are in fact healthy. The question is how do you balance those two pieces and when do you risk over investing? And that's sort of this Cournot game of chicken that everyone's playing now. The Cournot equilibrium comes when a small number of labs, an oligopoly, effectively choose supply at the frontier level and then the market clears at a high price for frontier access. So choosing supply in this means how many data centers get built, how many GPUs get ordered, but also how much low latency capacity is allocated to the top tier. So right now OpenAI just did the Cerebras deal. There's Claude fast and there's a whole bunch of different modes that will deliver faster inference and how many of those fast queries you get. How much of the best chips are allocated to a particular tier that you're paying for is an economic question for the labs. So on the true frontier there aren't great substitutes and so price stays high based on customers willingness to pay for frontier access. So you can just think about it in more simpler terms like there's a ton of developers and knowledge workers who are happy to pay hundreds of dollars a month or more, but they always want the best available model. This is most people in executive roles in startups, right? It's like yeah, I got my $200 a month subscription, I'll pay 250 or 100 or whatever, couple hundred bucks and it just makes me better at my job. I just do whatever I need to do. But don't give me the old thing. I want the best. I want to know that the hallucination rate is as low as possible. I want to know that when it builds me an economic model, it's doing it as best as it possibly can. So I need to spend less time. 1% of the time it makes a career ending mistake. Maybe. I. No, no, I really do. I mean, there are some crazy, crazy things, like possible outcomes. We haven't heard too many nightmare stories, but they're certainly have to be out there. Like the true hallucinations need checks and balances. Yeah, yeah. But increasingly. Yeah, increasingly less. Everyone's gone through the experience of somebody building a model and then being like, I don't know why, but it's wrong. Yeah, yeah. And then it's usually a battle. Yeah, I was, I was actually. You got to train that instinct that can just clock this even if you have. No, if it feels off. I was talking to a buddy about using AI tools who works in high yield debt, sales and trading and issuance. And he was like, yeah, the models just like, aren't that good. And I was like, really? And he was like, yeah, they don't understand like Q4. 76 of the tax code and how it applies. And I was like, okay, yeah, actually that seems like something that they might not be great at. I was like. Because I built a toy model which is like project out the cash flows, discount them back. And it did a great job. And I was like, cool. And he was like, no, I'm like 25 levels above that and like deep in all these different codes. And like, it's not good work, little bro. It'll get there, but it's just like, you know, the base models are not just going to be able to like one shot that. What do you think, Tyler? Fire it back. You would imagine that like something as like if it's just like text code, even if it's super complex, you can just put that in the context and it just does a very good job. Yes, but there's like finding specific facts from a super big document is like, that's one of the things that models are best at. Yes, but there's. There's a lot of. There's a lot of you laughing at the chat. Ethan. Computer fix tvpn. Use Gunna and grok. Gunna can fix us. We're back. We're back. It's. Yes, but there are nuances to the tax code and to a lot of legal documents that. Where the law says one thing, but it's implemented in a different way, or it's not enforced, or some regulatory organization has enforcement discretion or there's some understanding. Even if you just go into training on ip, it's like if you have really good lawyers, you can figure out, okay, well they'll come to the table, we'll be able to negotiate this, this will be the price, this will be the settlement. So even though the model might say don't do like you have a chance to do it, there's a whole bunch of different reasons why. But yes, I generally agree with you. This is like a six month away thing. Not. But it's just like the current, the. Currents, the current thing, but with like good, you know. Yes, yes. And maybe some more harnessing and more, some and better context. Anyway, the question is like where does this all go? What is the natural state of equilibrium in the longer term? Right now they're locked in Cournot where there's a whole bunch of companies, individuals that want to buy the best AI inference possible. They want the best tokens, they want the best deep research reports, they want the best code and they're willing to pay for it. But we're supply constrained. And so OpenAI brings a certain amount of GPUs and tokens and supply to the market. And then Anthropic also does. And the whole Cornot thing is that they're all keying off of each other. Yeah. And saying, okay, if they're offering this much, I'm going to offer that much. And to be clear, so having a product, not just an API business, gives you leverage because at some point the models are smart enough where you don't need to train them. You don't need to train a model that is 4% better because people are still coming to your application and having a good product experience, right? Yes. So historically one of the critiques to Anthropic's business was that they have to just be on this constant, constant fly sort of hamster wheel of training the best model because they have the majority of their business is this API business. They're not an aggregator yet. Swap it out for a smarter model. That said, they have cloud code now which gives them some more leverage over the market. And the really interesting thing is that Dario is now talking about being near the end of the exponential or maybe producing like the final models because we've talked to a few people about this, but it's very unclear if it's possible to create a super intelligence that's like 5,000 IQ. It might just be they get good at all knowledge work and they can answer all tasks, but it's like the digital guy. And so at that point it does commoditize and you drop out of Cournot equilibrium and you become more. Customers are more aggressive about switching to cheaper models to cut costs because the frontier is now commoditized in the entire backlog. Everyone is at the frontier, basically. And so in that scenario you switch over to Bertrand competition, which doesn't really mean that profits go to zero, but there is more competition and it looks a lot more like the hyperscaler cloud market, which is I think, what people have been sort of signaling towards. And also it sort of explains why a lot of the VC firms are getting in multiple companies. Because they don't think it's going to be winner take all anymore. They think it's going to be much more oligopolistic for the long term. And there will be competition between the major three or four labs that will. And it will be much more about how can you marshal enough supply, create a huge barrier to entry. Like you and I could start an AWS competitor tomorrow, but it's gonna be extremely expensive to bring up data centers that just serve web apps everywhere, let alone AI stuff, right? Building all those data centers. You're thinking what I'm thinking. You're thinking AWS people are saying, when does tpp? I was hanging with my buddy Ben on Sunday and he was, we both live in Malibu. He was thinking of just getting some chips and setting up Malibu Inference. There we go. Just the name alone sounds like you could get at least one Malibu Inference. That's really funny. Would be a good. Would be a beautiful name for a neocloud. Yeah, it's funny. And so there is a question about in the final models, will there be differentiation? What does differentiation look like? There's a potential for differentiation around. Okay. You get a model that's really good at bio or math or engineering or code or writing. That's possible. Another possibility is that there's just differentiation on quality, latency, safety, enterprise tooling. And you wind up with a small number of vertically integrated firms. And so in that scenario, I feel like OpenAI's recent moves make a lot of sense. Cerebras offers differentiated product around spin feed Peter from Open Claud joining makes a lot of sense around building orchestration products that can route to particular models. So the next generation of the Claude code router that can route you to specific APIs and act as a model router on top for the Agentic Programmer AI Assistant use case. You already see the model router happening in the ChatGPT consumer app. We haven't really seen that in the desktop CLI app. And so there's something interesting there. And then frontier. OpenAI frontier, that for deployed engineer thing that digs deeper into the enterprise and probably gets them more hooks in there, more pricing power. The chat's going off today, Ryan says the Cournot Company of Malibu. Let's do it. I love it. Let me tell you about CrowdStrike. Your business is AI, their business is securing it. CrowdStrike secures AI and stops breaches is. Let's play the clip of Dwarkesh Patel and Dari Amide discussing the economics of AI labs because I think it's informative and this is where the whole Cornot current thing came from because Dario dropped that phrase. So when you said why is no one talking about it? It was literally everyone that listened to the biggest podcast of the week. But I'm explaining it it. Let's play it. Whole industry here. Like we have a, you know, let's just imagine we're, we're in like an economics textbook. We have a small number of firms. Each can invest a limited amount in, you know, or, or, or like each can invest some fraction, fraction in R and D. They have some marginal cost to serve the margins on that, the profit mar. The gross profit margins on that marginal cost are like very high. Because, because, because inference is efficient. There's some competition, but the models are also differentiated. There's some, there's some, you know, companies will compete to push their research budgets up. But like, because there's a small number of players, you know, we have the, what is it called? The Cournot equilibrium, I think is what the, what the Cornell firm equal equilibrium is. The point is it doesn't equilibrate to perfect competition with, with, with, with, with, with zero margins. If there's like three firms, if there's three firms in the economy, all are kind of independently behaving rationally. It doesn't equilibrate to zero. Help me understand that because right now. We do have three leading firms and they're not making profit. And so. That's a good question. Yeah. What is changing? Yeah. So again, the gross margins right now are very positive. What's happening is a combination of two things. One is we're still in the exponential scale up phase of compute. So basically what that means is we're training. Like a model gets trained, it costs. Let's say a model got trained. That Costs a billion dollars last year. And then this year it produced $4 billion of revenue and cost $1 billion to inference from. So, you know, again, I'm using stylized number here, but you know, 75. These are my numbers. I'm just picking random numbers. Gross margins and you know, this 25% tax. So that model as a whole makes $2 billion, but at the same time, we're spending $10 billion to train the next model because there's an exponential scale up. And so the company loses money. Each model makes money, but the company loses money. The equilibrium I'm talking about is an equilibrium where we have the country of geniuses. We have the country of geniuses in the data center. But that model training scale up has equilibrated more. Maybe it's still going up. We're still trying to predict the demand, but it's more leveled out. Let me tell you about Turbo Puffer, serverless vector and full text search built from first principles and object storage. Fast 10x cheaper and extremely scalable. There is another fun clip that we should watch from A Beautiful Mind. Jordi, have you seen A Beautiful Mind? It won the Oscar for best picture. I believe it's about the mathematician John Nash. Have you seen A Beautiful Mind? I've not. Wow. Unk status over there. You would have. I was walking on the beach with Senra on I think, Saturday, and we walked by an incredibly famous. One of the top movie directors of the last probably 10 years. Wait, really? And Senra was like, do you see that? And I was like, see what? It's a guy with a dog. That's hilarious. I feel like that your beach tours have been really star studded lately. This is a different from the previous one you mentioned, correct? Yes. Wow. Yes. That's remarkable. Well, let's pull up the clip from. Incoming. Gentlemen, This room of beautiful mice. Nash, you might want to stop shuffling your papers for five seconds. Is that Eric Lyman? Yes, it's Eric Lyman in the. In the ramp biopic. We got. We got. We got our cast right here. Oh, this is the original like looks maxing movie feels. She should be moving in slow motion. Will she want to look? Should we say swords, gentlemen? Pistols at dawn? Have you remembered nothing? Recall the lessons of Adam Smith, the father of modern economics. In competition, individual ambition serves the common good. Exactly. Every man for himself, gentlemen. And those who strike out are stuck with their friends. I'm not gonna strike out. You can lead a blonde to water, but you can't make a drink. I don't think he said that. All right, nobody move. She's looking over, looking at Nash. Oh, God. All right, he may have the upper hand now, but wait until he opens his mouth. I think this. This is very, very stylized and completely apocryphal. Like, he definitely thought of this theory, but not at a bar. We block each other. Not a single one of us is going to get her. So then we go for her friends, but they will all give us the cold shoulder because nobody likes to be second choice. What if no one goes for the blonde? We don't get in each other's way, and we don't insult the other girls. That's the only way we went in. That's the only way we all get laid. So he's describing the prisoner's dilemma, Adam Smith said, where everyone must work together. This result comes from everyone in the group doing what's best for himself. Right? That's what he said. That's right. Incomplete, Incomplete, incomplete. Okay, incomplete. Because the best result would come from everyone in the group doing what's best for himself and the group. Ash, this is some way for you to get the blonde on your own. You can go to hell. Governing dynamics, gentlemen. Governing dynamics. Adam Smith. Who's wrong? There we go. Careful, careful. Thank you. Tis so good. Anyway, very fun. Dave says just joined the stream. We watching a movie. Yeah, movie day. Movie day. We should bring back class. Bring back movie days in school. Yes. I mean, that was iconic. Rainy day substitution Gladiator in Latin class. That was the best. Really quickly, let me tell you about Vibe Co, where DTC brands, B2B startups, and AI companies advertise on streaming TV, pick channels, target audiences, and measure sales, just like on Meta. Anyway, very. Yeah, very fun. Nash. Equilibrium. Lots of game theory going on in the. In the AI world wars. Right now, everyone's trying to figure out how far to push it. There's a fair amount of risk. There's still the Cournot game of chicken around. Who will invest the most in advancing the frontier. But the end state looks a lot more durable than pure model commoditization and the perfectly competitive situation that many were predicting a few years ago. So if you go back to, like, 2023, a lot of people were saying, like, the models will commoditize and there will be no value there. Like, no profit. Because, like, the deep seq moment, and it'll all be run on commodity hardware. That doesn't seem that likely. It seems like the labs will turn into sort of new hyperscalers. There will be increased competition, but still, Very, very good businesses a la cloud. So anyway, fun to dig into some little econ 101 or 102 depending. But let's go to the timeline. Bucho Capital says I thought Dwarkesh had a good point that software engineering is the only job where the full context need do the job is available to an AI agent via the code base. And I didn't think Daria had a good answer for why automating other jobs will be as easy. This got a bunch of a lot of people kind of reacting, kind of disagreeing generally that all of the full context needed to do the job is available. But I do think it's something we need to figure out. Yeah, we were debating this because there was a post that was just sort of like a wojack reaction that was just making fun of this and it wasn't clear if they were saying that they were agreeing or disagreeing. But basically my take was, well, it's possible that a lot of the full context needed to do the job of a lot of different white collar jobs is in fact logged. It's just logged in the final product which is like a deck or a spreadsheet or a decision and then a whole bunch of emails, a whole bunch of slacks and then a whole bunch of of zoom calls that's recorded. And so yes, if you're running a business where a lot of work gets done in smoky bars late at night and back alley deal making sure that's going to be harder to automate. But in the world where it's someone sitting in front of a computer and there's a screen recorder running, you should be able to pull up most of the context at the same time. You can't just snap your fingers and go back and get at every decision that was made in the 80s that allowed coca Cola to become a dominant soda maker. But you can, you literally can with limits people on calls just being like knowing the calls being recorded and used to train something to replace them. They're just like I'll tell you offline, I'm not speaking this secret into your record. Yeah, golf this weekend. You want to play golf really quickly. Figma ship the best version, not the first one with Figma introducing Claude code to figma explore more options, push ideas further. Figma but this was funny, Buko said. Basically a non zero chance of Joe Wiesenthal victory where software engineers write themselves out of a job first and everyone else has the boring parts of their job automated. Joe of course was joking, just saying. Well of course software engineering Is being automated. So easy. It's so easy. The debate around the posture. To our cash. To our cash. His posture was. He's been at the gym a lot. He looks fantastic. And I love this sweater. The crew neck works really well. The pushed up sleeves is a particular choice. Didn't translate into that. Chad wojack. But he looks fantastic here. And the Aeron chair is certainly helping with the posture. A lot of fun on the timeline, looking at the. The looks mogging or whatever. The looks maxing. I don't even know. Frame mogging. That's the frame. He kept bringing up the example of a video editor saying, yeah, but when will the models be good enough to edit videos? Well, and pick out moments and give me two years and another 500 billion. This is. I mean, we've tried every tool there is. Yeah. They can't do it yet. Tricky. I don't know. I don't know what's. And it's not even that we're not trying the tools to replace the people on our team. We're trying to make them have higher output. Yeah. But to date, I don't know if. There'S something super sticky. I mean, it does seem like one interesting thing is that there aren't a lot of open source, like premiere profiles. Like, I've edited a ton of videos for YouTube. There's a whole bunch of cuts in there. What I cut out, what I didn't. You could have that record, but it's not stored in GitHub. Like, it's just. You can't necessarily train on it. You can train on the final product and understand, but you don't understand what actually got left on the cutting. Cutting room floor. There's this whole concept of like kill your darlings. Like when you're in the edit, like you need to be cutting more. You're like, ah, I like that shot. It's so cinematic, so cool. But. But does it actually advance the story? No. So you cut it down. I was watching the Matrix this weekend and there's this amazing shot of when Neo and Morpheus are going to visit the Oracle. Is that one the Oracle And Neo. Is that the one Oracle? Yeah. And they reach for the doorknob and the doorknob has this perfect reflection and the reflection shows Neo and Morpheus and they had to do this crazy VFX shot to hide the camera in what looks like Morpheus coat. Because if you point a camera at a mirror, you see the camera and you don't want to see the cameraman there. That ruins the shot. And so they did all this crazy stuff to like, to like, you know, cover up the camera. And I'd seen the behind the scenes and been like, wow, that's really impressive. And in my memory I thought it was like, oh, it's such an important shot. They probably lingered on that for like five seconds to really let it sink in. Like they're pulling a trick on the audience. It's beautiful. It's like half a second. And they did all this work and then they knew that like from a story perspective, from a storytelling perspective, you don't want to hang out and watch a picture of a doorknob for five seconds. And so all these decisions, they sort of get chronicled, but they don't get neatly organized in the way that a GitHub log does with pull request discussions and what happens. So it'll be difficult. So maybe two years, another $5 billion does it. But it's coming, so we'll keep monitoring it. Let me tell you about the New York Stock Exchange. Wanna change the world? Raise capital at the New York Stock Exchange. We have a special guest joining. John Caramenica is joining and I want to music. But I wanted to tee up his appearance with a little music review that I found in the newspaper today. The algorithm wants you to grieve. Rest in peace. My granny, she got hit by a bazooka. A title that functions as its own thesis statement is the kind of song that resists every framework we've built for understanding virality. It is not quite parody, not quite sincerity, not quite meme. It exists in some fourth space where emotional register becomes irrelevant and the only metric that matters is whether the thing lodges itself in your prefrontal cortex like a splinter. The melody, if we're being generous enough to call it that, is a three note loop that sounds like it was composed on a Fisher Price keyboard during a power outage. The vocal delivery is flat, almost affect less, a kind of anti performance that paradoxically demands more attention than virtuosity ever could. The lyrical content is, well, the title repeated with conviction. And yet something is happening here that's worth taking seriously, or at least worth resisting the urge to dismiss. The song collapses the distance between tragedy and comedy so completely that neither category survives the collision. It's grief as non sequitur, eulogy as punchline, memorial as munition. The bazooka isn't just absurd, it's so specifically absurd that it loops past irony and arrives somewhere strangely earnest. Nobody's grandmother has ever been hit by a bazooka. The scenario is impossible. And impossibility, it turns out, is its own form of tenderness, a way of saying loss is so incomprehensible that only nonsense can hold it. Or maybe it's just a guy yelling about a bazooka. The Internet doesn't require you to choose, but we'll be following this story, so thanks for telling me. We'll get John's take on that. We will. It's the song of the week and the the ad platform of the week is Applovin. Profitable advertising made it Easy with Axon AI. Get access to over 1 billion daily active users and grow your business. Today, Andrew Reed says, horses don't stop, they keep going. Wait, did he actually say that? Yes. No way. Yes. In response to 2026 being the year of the horse. I love it. One of the greatest lyrics of all time, originally. To explain the joke is it a Young Thug song? And the actual lyric is hustlers don't stop, they keep going. Really? But it sounds like horses. And so people put horses don't stop, they keep going. And they show the AI generated image of the horse bench pressing. And it's incredibly inspiring. There's a lot of Young Thug songs that are hard to fully decipher. 100%. Let's hit the size gong for this. Pennsylvania Girl Scout, 6 years old, breaks record selling 87,000 boxes of cookies. She's unstopp, unstoppable. How much is that? What's the ARR? How much is a break it down? Is it $12? That's a lot. How long do this? Six years old. Wow. What enabled this incredible scale? Was it potentially Shopify? Shopify is the commerce platform that grows with your business, lets you sell in seconds online, in store, on mobile, on social, on marketplaces. And now with AI estimating that it's somewhere around $600,000 of sales. Sales. It's amazing. At only six years old. Really incredible stuff. Heartwarming. That's awesome. India's Adani group to invest 100 billion in AI infrastructure. We gotta hit the gong again. Hit it again. The Indian boost to the country's ambitions to become an AI power. India's Andani Group, an energy and logistics giant, said it would invest $100 billion to develop large scale data centers by 2035, the largest such commitment in India so far. Tyler, what do you think about the timing here? Is this going to be too late? Are the clankers going to like is 2035? How are we looking there? Is that, I mean, singularity? Yeah, I'm Very bullish on the clankers coming pretty early. So you know, time will tell I guess. We'll see. I cannot wait to pull up this clip. It is, it is a big number that I feel like a lot of countries have been teasing big numbers but this is. Yeah, they're kind of mogging. Macron. Yeah, this is like a really big number. You see a bunch of like multibillion dollar deals, multi billion dollar releases but this is like serious, serious, serious investment. So you know, good, good news for many countries. One of the biggest concerns about the looming AI era is that it will deepen existing technology divides in which tech services are developed by companies in a few countries. Some countries have also expressed concerns about how large tech firms were will consume data generated by their citizens to build their AI products. India is focused on making sure that startups and researchers use AI to solve pressing development challenges and paving the way for its businesses to become providers of global AI services. India will shape solutions not just for India but for the world said Indian Prime Minister Modi in a social media post today. The Andani Group said the dedicated said the dedicated computing capacity it is building will support Indian large language models and ensure data generated in India is stored locally. Let's see how the stock's doing up 2% today. Is it a $28 billion company? Do I have that right? I don't know. It's like 2.5 trillion INR so I think that's about 28 billion. So this is a huge investment says Andani will more than likely use government leverage. Sure. Government backed data centers. Are you happy about that Tyler? Government backstop coming to India. Backstop. Let's go. Before we move on let me tell you about Restream 1 livestream 30 plus destinations. If you want a multi stream go. To restream.com Micron is spending 200 billion so congratulations for setting the biggest number. Micron. Micron is spending 200 billion to break the AI memory bottleneck. For decades memory chips were low margin commodity price products. Now the industry can't make enough to satisfy data center's hunger. Just this one company is like yeah, we're going to spend twice as much as India as the adani Group. Boise, Idaho. Each afternoon at around 4:30 the earth here shakes from a series of controlled explosions as engineers blast through basalt bedrock to flatten out the ground underneath a gigantic new semiconductor factory. Let's give it up for Idaho. Yeah, we don't talk about Idaho enough. Micron Technology is the largest American maker of memory chips. The tiny Slices of silicon that store and transfer data and help power everything from smartphones and car computers to laptops and data centers. Micron is rushing to add manufacturing capacity to avert the biggest supply crunch the memory industry has seen in more than 40 years. In Boise, where the company is based, Micron is spending 50 billion to more than double the size of its 450 acre campus, including the construction of two new chip factories. The first fabs inaugural wafers are expected to roll off the factory line in mid-2027. So this is their first fab or the one that they're spending the 50 billion on right now? I think the new one. Okay, so that's pretty quick. Yeah. Did you hear that the PS6, the PlayStation 6 is now delayed because of memory shortages. 2029, 2028, 2029. It's still a rumor. Maybe they'll, you know, maybe this solves it and they can get it out earlier. But pretty, pretty big delay to 2029. They really don't refresh. You just created a trillion gamers. No, seriously, I think adding insult to. Injury to the gamers might actually rise up. Gamers might rise up, up. They might be an important voting block. A lot of them are, a lot of them are of age to vote and a lot of them don't would rather have new gaming hardware than, you know, necessarily. AI slop in the feed and they're like, yeah, I can't afford the new PC that I wanted. What do you think? Yeah, I don't know. I mean I feel like this says a lot about how good the PS5 is. Right. Because they can afford to just like postpone the PS6. Like what games? Oh, now you don't want technological progress. Wow. I want it to go to the data centers. I don't care about like the next like game graphics. Have like, have they gotten that much better in the past, like five years? Like yeah, maybe. But it's like for the actual gameplay, is it that important if like the actual pixels are. Yeah. Realistically, a lot of this stuff should be moving to the cloud soon if it's not already. Yeah. Like it was the Meta Quest VR. I forget. Oh yeah, Xbox Edition. Yeah, yeah. And it was all in the cloud and it was like totally fine. Low latency. Yeah. And then if you're running in the cloud, you can upgrade the hardware and in theory you should be able to run like a gen AI up resing pass to make it more photoreal. And I feel like that's going to be where more of the juice is squeezed out of the graphics than just continuing on the traditional path of like more pixels, more ray tracing. It'll be make a really beautifully designed video game that works really well. Really tight deterministic interactions so it's satisfying and then give it a layer. We got to have like a RAM trader on the show. Really somebody that's in the thick of deal making in this space. There's a couple folks over at Semianlysis that cover it. One of them just went on lots. Let's do it. Get them on the show. Let's do it. Let me tell you about graphite code review for the age of AI. Graphite helps teams on GitHub ship higher quality software faster. That's not all. Near Syracuse, Micron just broke down, broke ground, just broke down in Syracuse. They just broke ground on $100 billion fab complex that represents the state of New York's largest ever private investment. Late last year, Micron announced a $9.6 billion fab in Hiroshima, Japan. Cool. While competitor SK Hynix announced in January that it would build a $13 billion fab in South Korea in addition to a $4 billion manufacturing complex it is building in Indiana. This is interesting. I feel like this is acceleration in capex from the memory makers. But this, I wonder how TSMC responds to this, because the big criticism of TSMC is that they are increasing capex, but the rate at which they are increasing capex is decelerated. And so they're not really keeping up with the exponential growth in compute that every AI lab, every data center builder, every memory maker is pushing towards. And I wonder what do they know? What do they know? Or TSMCs being so conservative, I mean, they might just have the AI industry in a chokehold. And so if there is a bottleneck, they can just raise prices, but they also might screw everything up by not investing enough. And then we get to some sort of solid bottleneck for a while and we can't actually roll out more. Well, let's get someone on from Semianalysis tomorrow to talk about memory. Moving on. Lucas Shaw was on a tear over the weekend, reporting on the Warner Brothers Paramount conversations. He says this morning Warner Brothers is going to resume talks with Paramount after two months of rejecting them playing mind games. The company still says it's committed to Netflix, but needs to find out just how much the Ellisons will offer. He originally reported on this Sunday, but it's being confirmed today. Again, we kind of knew this was going to happen if the Ellisons had been saying, we're Making a. We're giving you a big number, but it's not our biggest number. It's not our best and final. So no surprise here. Paramount has now just eked out a lead on who will successfully take over Warner Brothers. Over on Kalshi, they're sitting at 49% chance with Netflix at 37% and then none before July 2027. At 37. This is such an interesting flippening here. I was talking this flip happened executive. This weekend and he was like, yeah, I think there's just too many people that are against Netflix for a variety of reasons that it'll reopen the Paramount conversations. And specifically I was like, but YouTube. YouTube. And he was like, no one buys that argument. Like, everyone thinks about the entertainment industry as its own thing and then social media Tech, TikTok, Instagram, YouTube as a separate thing. And no one is making that comp. It's just YouTube is not a, you know, having having. We asked Ashley Vance, how do you feel about having kind of one less big buyer in the buyer pool for documentaries? And he was like, not great. Already bad. This would be worse. But YouTube is not a buyer of IP. Yep. They host content and serve it to people. So it's a tough argument to make. Even though they're been absolutely on an insane tear from a. From an overall watch time standpoint point. Yeah. Hold on. Let me tell you about MongoDB. What's the only thing faster than the AI market your business on MongoDB? Don't just build AI. Own the data platform that powers it. AJ says Grace Blackwell is a beautiful name for a baby girl. I totally agree. I mean, isn't that. Because it's actually the name of a girl? Right. Like, isn't Grace like the name. Like, they're like the chips are named after real people. Who's. Well, there's like Ada Lovelace. I want to say Trevor. Grace Hopper. Yeah, Grace Hopper. So it was. And David. Harold Blackwell. David Blackwell. I always think Trevor Blackwell because he's a YC partner. But he did not invent the Blackwell. But Grace Blackwell is a beautiful name. Vera Rubin too. Vera Rubin is also the name of a female mathematician, I believe. Or the only person I can find named Grace Blackwell is an audio designer and composer based in the uk. So gonna have a rough go on the SEO front. Yeah. Let's flip over to Claude Bot. Kent Dodd says names the thing Claude Bot. Claude asks for a rename. Renames to Open Claw. OpenAI buys it. Legendary. Legendary. Couple weeks I was looking back on like, when did we talk to Peter? And it was like two weeks ago. That was when? Feels like two months ago. It feels like two months ago. Things are moving very, very fast now. No confirmation on buying. It's an open source project, they're keeping it open source. There's a whole bunch of different. Dave Morin, Reading is going to step in to, I believe, run the foundation that will kind of steward the open source project. And then Peter's obviously joining OpenAI. This wasn't, this wasn't, this wasn't super surprising to me. When you looked at the potential places that he could land, it felt like there was, he had talked about like, hey, I'm kind of losing money on this. And he. Our takeaway from the conversation is this is a guy that just wants to keep shipping, get this product into the hands of as many people as possible. And when you looked at the kind of buyer pool, you know, he was, he was coming to the west coast, you can assume that he was making the rounds. He went on Lex and was like, yeah, I'm talking to OpenAI and Meta, which is like a crazy thing to say. And Meta, Meta always felt like a somewhat of a long shot because even though I think I'm sure he's done very well from this, it didn't feel like dollars were his number one priority. And it's very possible that Zuck would have been willing to overpay or pay whatever price needed to get him on board. Zuck had just also acquired Manus, which feels like his agent's team. Right. And Manus has already responded by rolling out some open claw like functionality. And so that didn't make sense. I think people were expecting Anthropic to make a move. Part of that is like Anthropic's just had so much mind share and been on an insane run the last month or so. And so I think people and the name, right. Peter also at multiple points said Anthropic didn't send their lawyers, they sent an individual. But it got turned into this meme that Anthropic had like really come and like really made his life difficult. I think they were like, hey, this is going to be confusing to consumers. And then the other thing is Peter's over and over and over talked about how much he loves Kodak. Like he talked about that on our show. He's been very vocal about it. He likes building with Codex. And so I think when you understand all these different factors where not entirely financially motivated, I'm sure that was a factor. But not entirely financially motivated. So saying the biggest number is not going to get him to make a move. OpenAI with somewhere around a billion consumers that use these products, that are using products in the, in the ecosystem today, that creates an exciting opportunity for him to come in and actually roll this out to as many people as possible. And you put all these factors together and him landing there makes a lot of sense. Yes. Let me tell you about Gemini 3 Pro, Google's most intelligent model yet. State of the art reasoning, next level vibe coding and deep multimodal understanding. Alex Hormozy chimed in on the open claw bot discourse. What do you say? He says, I'll take this day off to figure out this whole open claw thing. Every entrepreneur on President's Day weekend, we've talked about this on the show before. Long weekends are really good for AI progress and AI diffusion because people have a few less meetings and they say, hey, maybe I can mess around and figure this thing out. It really is remarkable how quickly Claude Bot openclaw broke through the discourse. Like I was at dinner at a steakhouse like last Wednesday and I overheard a conversation with just a couple guys in suits who I don't think work in tech talking about different models and how they wanted to use openclaw and they were thinking about getting Mac Minis like it was like it really did break through. Beyond our lawyer, our lawyer who doesn't work at a. He works at an entertainment focused law firm. He got a Mac Mini, set it up and so yeah, it's very mainstream. I did see some people reporting that the Mac Mini had sold out in their local area too. So it's not fully sold out by any means, but you look at the traction on GitHub and you look at just the overall demand and how kind of like mainstream at least the concept of this has gone and Apple's overall production numbers on the device. I mean I think we could enter a world where at least the ones the more in demand models are sold out. Yeah. Petition for 3 day weekends to speed up AGI timelines probably would work for sure. Fumble Gate FumbleGate did anthropic fumble Open claw. Darius says, I don't think about you at all. Probably overstated it doesn't seem like they, it was like key to their strategy. They're sort of just focused on their own thing. I don't know, it doesn't seem like, oh, they completely fumbled. Yeah, Part of this was the original naming claudebot. Right. But when you look at how many different projects Peter shipped, it's not like he was thinking at every moment, oh, I have to name this because it's going to get a million GitHub stars in a month. He was just like ripping products, building really quickly. And it wasn't even. Claude wasn't the default in Openclaw ever. Yeah, yeah, Codex was actually the number one. And there are like 10 LLMs that you can choose from when you actually set it up. It very much seems like he's just having fun with the name. I mean, the fact that he was calling it Mult Bot for a while is just tells you that was a. Great, that was a great, that was a great 24 hours. 24 hours. But, but, but seriously, like he, I don't think he was like, oh, I need a branding agency and I need to think of the name that will go forever. It's just like, yeah, this is like all a lobster meme. It's fun. Wes Winder says lol. Meta couldn't acquire openclaw, so they pulled a classic Zuck move. Just clone the whole thing. Then it added as a feature to an existing product. Surprised they didn't try to shove this into Instagram. So Manus announced Manus agents yesterday. They move quick. Your personal Man. Manus. Manus. Manus now inside your chats, long term memory, full Manus power. Create videos, slides, websites, images from one message, your tools, connect, Gmail, calendar notion and more. Available now in Telegram, I would assume. I was kind of surprised they wouldn't roll this out into WhatsApp. Yeah, they don't even own Telegram. Available now on. Wait. That is so confusing. I think it's possible that WhatsApp didn't have the, didn't have some functionality that they needed to actually do this integration and they were moving so quickly. Like they, they, they seem to have turned this. I don't think. I think like, Wes is maybe overstating a little bit like the classic Zuck move. You can imagine, like Manus would have done this probably independently. 100% you were calling this. Manus is a good acquisition. Because if in the, if you want. Super intelligence for like, Manus is great. Yes. And so Zuck has been great at buying a product that has a lot of potential, that has some customer traction and like scaling up massively. Yeah. And even just for the most mundane Instagram usage where you could say, you know, oh, I want to look across my audience and see who's the most engaged or understand what trends are happening in my vertical or, you know, like, respond to every comment with a heart. If it's positive. Like anything that you'd want as an agent to go around on these social platforms and do, let alone like go buy something for you. Having, having a team that works on agents makes a ton of sense, but it's. Yeah, it's funny. It's just funny that they're rolling it out. Matea says, I think that's connected to WhatsApp de platforming other AI companies. Remember ChatGPT had like a huge integration in WhatsApp and they got. And they got booted. That's crazy. Hari shares a little bit of lore about Peter Steinberger. Says he bootstrapped PSPDFKit over 10 years ago. It's the gold standard PDF library used by Box, Apple and even DocuSign. That's crazy. Insanely hard tech. If you know, you know it's very hard to work on PDFs. I guess he's worth a couple hundred mil from that. Everyone's trying to guess the number, but nothing's been confirmed. If he goes to OpenAI, he might make more from OpenClaw, which is three. Peter says he signed the contract today with software my company built years ago. I love that. Now it's nutrient docs still best for anything PDF. It's like I'm PDF maxing. Let's take a look at this. Harnesses won't matter in less than three years, says Ahmad. Yes, there's money to make at the moment, but keep in mind that this is the play of two years ago and you should be playing for what the models will be capable of in two years from today. Instead, Claude plays Pokemon. Kind of proves this opus 4.5 and 4.6 has the same insanely bad harness. No babysitting like Gemini and GPT and 4.6 is like 1197 hours ahead of 4.5, so. Oh, it's how far they get. Yeah, I've seen some people trying to back a bunch of like open claw startups and given that just feels like a potential bloodbath, I think especially especially knowing that that OpenAI will be able to build, I would assume build the best possible kind of wrapper around everything that OpenClaw has built so far. Again, I would assume that the three major labs, DeepMind, Anthropic and OpenAI all have serious orchestrators very soon. And they will acquire talent and acquire products and build products and, and license stuff and figure out all sorts of integrations, do all the business things necessary to make it happen. But as I wrote in the Year of Orchestration or Orchestrators like This is coming. People are managing multiple agents already and tools that make that easier, tools that make that more reliable and more efficient. Like, that's gonna be a big focus for this year and everyone's gonna be focused on it. So the idea of like, oh, I should just like fork this open source project and like bring it to enterprises, like that's going to be rough when OpenAI has 1004 deployed engineers going into every enterprise that they already have contracts with. Tyler, what do you think? Yeah, I was going to say, like, I think it's a similar thing to the hardware. So like Gemini is developing the TPU with the architecture of the model in mind, and they're developing the architecture with the hardware in mind. Right. So it's kind of like these things are like helping each other. So you're going to see the same thing in the harness. Right. Where the harness is made with the model in mind, obviously. But then you're going to start to see as these things as agents become more and more popular. OpenAI already has, there's the normal GPT513 and then there's 513 codecs. Right. So the models are built with the harness in mind. So I think if you're not the one building the model and you're just adding your own harness, the model is not going to be like customized for your harness. Yeah. So if OpenAI Anthropic are building their own harness, they're going to do it with the model in mind. It's going to be like much better than having some third party thing that you build, right? I would think so, yeah. No, that's totally reasonable. Harness for horses. Horses don't stop. Harnesses don't stop. Will Brown says honestly crazy that OpenClaws sold for 1 billion. Like he's really the first Solo $5 billion founder. Time will tell if it's worth 15 billion that OpenAI spent on the acquisition, but it's pretty wild that you can just vibe code an open source force project to make 40 billion in a couple months. Now. It really, really nails it because everyone jumped immediately to a billion. Immediately off of nothing. Off of nothing. Off of like one rumor from a thinking machines person that went to Meta and we don't even know what the earnout schedule for that is and how much that happens. But it's very, very funny. Who knows? We got to give some credit to Simp for Satoshi. I brought this up before. SIMP has been working on truffles for a handful of years now. I've been by his studio in Venice. It's a very cool setup and team over there. But he posted a little bit about this. Where is it? And Tinybox. George Hotz's project is also apparently going to ship a more consumer product. Maybe. I think they're around $10,000 right now. They're probably going to bring that down to a few thousand dollars or maybe even a few hundred dollars. You get a lot with a Mac Mini. But we'll see where the prices land after the memory shortage takes hold and whatnot. But lots of people are doing this. Alex Cohen breaks it down for Gen Z. If you're wondering what happened today, Claude was mogging OpenAI for weeks. Then this Jim Cell dev ships Claude bot which was the fastest growing open source thing ever. Absolute looks max for the whole ecosystem. Anthropic tries to dairy Goon him with legal Dev renames to OpenClaw. OpenAI slides in like a void pulling chad with acquisition interest. OpenClaw gets acquired by OpenAI. Now Anthropic is getting Jester Goon by the Entire timeline and OpenAI AI is giga maxing off their fumble Open Anthropic could have just let them cook him cook. Instead they went full moid and got out framed by the Jester Maxers at OpenAI 8000 likes the looks max lingo is really, it feels like hilarious. I do wonder the half life. I feel like it's. I think it's over. It's got to be. It's got to be towards the end of this boom. But the rise of of the kick streamers is certainly the story of the year. Certainly the story of the year. Dave Morin shared more context on what the future of openclaw will look like. So Dave Morin who you might know from offline ventures slow he's been on our show path. He is now on the board of openclaw and he says he's been working on the openclaw foundation structure for weeks. A home for thinkers and hackers that choose and those who want to own their data. He's honored to serve as the founding independent board member. This community built something extraordinary. Our job is to protect it. Open source forever. Excited to share more soon, Raoul says let's go. There was some funny pushback about someone was like oh it's so unfair that Peter's like reaping all this reward from these work like it's an open source project. There were a lot of committers and he shared the number of commitments and it's just like him and like 90% of the work. And then like a few of his buddies who are all in the chat, in the reply, like, yeah, like we love this, this is great. And a number of them are going to OpenAI too. He's like the headline pickup. But. And then a number of the other committers. And then those AI models. Yeah, and then, and then the other key contributors started committing after. It was like fully, totally, totally. Hundreds of thousands of people building on top of it. It was, it was like pure. I don't know, there's no term, I don't know if there's a term for it, but it was, it was like third party angst or something. It was like people were feeling angst on behalf of people who felt no angst. The people on the team were all like, this is awesome. We're all on board. And then other people were like, they got screwed. And it's like, I don't know. Not quite. Intern is sharing what your 2027 team off site will look like. Taking the Mac Minis somewhere nice. Taking them out. Yeah, we'll see. And here's the post you were talking about from Ashen. I didn't believe it, but Mac Mini studios are actually sold out in most places. Walmart, Apple Store, Amazon, Best Buy, Micro Center. Apple has a month wait now too for most AI type beat models. 48 gigs plus of Ram. Funniest part is that there are still 24 gig. What's up 48. They said it was possible. Oh, that they would sell it. Yeah, yeah. I went to the Apple store and they were fully in stock. But you know, the timeline moved on and people, people went more aggressively into the Mac Mini. Let me tell you about Vanta Automate Compliance and Security. Vanta is the leading AI trust management platform. Started today. We have some breaking news out of New York. Oh yes, what happened? Josh Kushner has announced Thrive 10 exceeding 10 billion. Thrive 10 compromises. 1 billion designated for early stage investments and 9 billion designated for growth stage investments. He said, we do not view this as a milestone, but as a commitment to the long work ahead. We view Thrive as a company. Our product is partnership. The willingness to commit deeply to a small number of founders and to stand with them through momentum and adversity. This is the discipline we bring to our work and the responsibility we accept. When founders partner with thriving Thrive. We do not hedge. Concentration demands loyalty to the founders and missions we back in this moment. Exposure alone is not a strategy. Judgment without commitment is not enough. Advantage will accrue to those who choose deliberately, commit deeply and endure through difficult moments. Thrive was founded to be an enabling technology for the world we want to see. We are deeply aware that we are not the main character. The founders that we are fortunate enough to partner with are the artists. Our role is to help create the conditions where great work can come to life. We take a long view grounded in the belief that category defining companies tend to create structural compounding advantages over long arcs. This fund reflects the continuity of our approach and the ways our work has deepened alongside the founders we support. We are grateful for the trust our limited partners place in us and for the opportunity to work alongside those who are building with purpose, integrity and courage. Hit that gong. Love to see it. Numerologist has been 5, 10 and 10 billion. I like that. Yeah. Despite having zero exposure or anything to TVPN, he's been nothing but kind and helpful to us on the journey and it's a testament to the work of the.

Someone launched a Federal Reserve simulator. Is there anything more on the. No, we can move on. This. Is more important. Somebody. Bill Mead Gaming just released the Federal Reserve simulator in Steam. If you like flight simulators, you're probably going to love this. Productivity. Is going to go to zero. Turn based people aren't going to be going to work. They're going to be taking days off to play this. A turn based economic simulation game where the player assumes the role as head of the United States Federal Reserve. This is awesome. Tyler fired up this. I would actually play this. Yeah, fire it up. Fire it up. Download it. I want, I want your. I want your review. I like Gary Garrett Jones here. Promising. And he's a professor of economics at George Mason. He's actually serious. I feel like. I feel like this would be an amazing game to play. So in the reviews I'm reading on Steam they say I learned more about monetary policy in 15 minutes on this SIM than I did in five years of university. That's actually. Please do investment banking simulator next. Yeah, I was thinking about the LBO simulator would go so hard. Let me you tell.

Automatically takes care of scaling, monitoring and security. What did Claude do? What did Claude do? Pentagon has said that Anthropic will pay a price. Of course, this was. There was reporting last week that Claude was leveraged in some way during the Maduro planning. The planning of the Maduro raid. I was imagining in my head Dario as Walter White in the suv, just being like, watching the logs and seeing Pete Hegseth running a deep research report on Maduro. Who is Nicholas Maduro? He's just like, no, no, don't do it. Yeah, Very, very unclear how. How it was used, but a lot of pushback, you know, Palmer. Palmer was pushing back pretty hard. Let's see if I can pull up what he actually said. So, I don't know, botch it. The Wall Street Journal headline is Pentagon used Anthropics Claude in Maduro, Venezuela raid. Use of the model through a contract with Palantir highlights growing role of AI in the Pentagon. Now, I remember seeing a clip from Sham Sankar at Palantir talking about how every query that the government runs is actually run through three or four different LLMs and then synthesized. They sort of put their own model router on top of the other models because if you're doing bad rapper basically, I guess it's basically a wrapper because you sort of want every possible, you know, piece of information. All the LLMs have different parts of the Internet, different training corpuses, different opinions, different. All sorts of different things. So you put all those together. The mission to capture Maduro and his wife included bombing several sites in Caracas last month. Anthropic's usage guidelines prohibit CLAUDE from being used to facilitate violence, develop weapons or conduct surveillance. We cannot comment on whether Claude or any other AI model was used for any specific operation, classified or otherwise, said an Anthropic spokesman. Any use of claude, whether in the private sector or across government, is required to comply with our usage policies, which govern how CLAUDE can be deployed. We work closely with our partners to ensure compliance. The deployment of CLAUDE occurred through Anthropic's partnership with data company Palantir Technologies, whose tools are commonly used by the Defense Department and federal law enforcement following the raid and employee of Anthropic ASTA counterpart at Palantir how CLAUDE was used in the operation, according to people familiar with the matter. An Anthropic spokesman said it hasn't discussed the use of CLAUDE for specific operations with any industry partners, including Palantir, outside of routine discussions on strictly technical matters. Anthropic is committed to using Frontier AI to support in support of US national security. Okay so we'll see where this goes. Anthropic's concerns of how about how clogged can be used by the Pentagon have pushed administration officials to consider canceling the contract worth up to 200 million which is a drop in the bucket at 14 billion ARR but still probably has larger implications for their relationship with the government broadly. Yeah there was someone who's saying I don't know who needs to hear this but punishing private companies because their CEOs don't share your politics is really bad capitalism. Palmer said this is not a reasonable characterization of what is happening. It isn't a matter of punishing companies for not sharing political views. It is a rational response to a vendor trying to control the government via terms of service and the products they power. I hope you realize oh yeah people are going on okay wait so disclose TV shared it as well Pentagon used and what is this someone launched a Federal Reserve.

Let's hit the size gong for this. Pennsylvania Girl Scout. Six years old, breaks record selling 87,000 boxes of cookies. She's unstoppable. Unstoppable. How much is that? What's the arrow? How much is a. Break it down? Is it $12? That's a lot. How long do this? Six years old. Wow. What enabled this incredible scale? Was it potentially Shopify? Shopify is the commerce platform that grows with your business, lets you sell in seconds online, in store, on mobile, on social, on marketplaces. And now with. AI estimating that it's somewhere around $600,000 of sales at only six years old. Really incredible stuff. Heartwarming. That's awesome.

For the world we want to see. We are deeply aware that we are not the main character. The founders that we are fortunate enough to partner with are the artists. Our role is to help create the conditions where great work can come to life. We take a long view grounded in the belief that category defining companies tend to create structural compounding advantages over long arcs. This fund reflects the continuity of our approach and the ways our work has deepened alongside the founders we support. We are grateful for the trust our limited partners place in us and for the opportunity to work alongside those who are building with purpose, integrity and courage. Hit that gong. Love to see it. Numerologist 5, 10 and 10 billion. I like that. Yeah. Despite having zero exposure or anything to tvpn, he's been nothing but kind and. And helpful to us on the journey. And it's a testament to the work of the whole Thrive team. And I will say their merch is phenomenal. Really good. I was wearing the. You were wearing the pajamas? The pajamas. No way. I was wearing the pajamas this weekend. Yeah. They got some good stuff, creative stuff. Like different stuff. Unexpected drops, which I love. Yep. And we have to read this off. It's just so good. So Selah says Joshua Kushner, strode billionaire. Across his room, his contrarian high concentration vibes engulfed Rick Rubin's bohemian monk retreat. Joshua's high conviction and mystery filled the room. I know what this is referring to. This is a reference to the Colossus profile, which is. Which really got a lot of people upset. A lot of people were nodding. They're like this. This doesn't count as writing. How it was. It. It transported me to. To Malibu. It was fantastic. I thought it was great. About Lambda Lambda is the super intelligence cloud. Building AI supercomputers for training and inference that scale from one GPU to hundreds of thousands. Let's head over to Tyler Cowen, Tyler Cowan. He is giving us some more economic data on.

His appearance with a little music review that I found in the newspaper today. The algorithm wants you to grieve. Rest in peace. My granny, she got Hit by a Bazooka A title that functions as its own thesis statement is the kind of song that resists every framework we've built for understanding virality. It is not quite parody, not quite sincerity, not quite meme. It exists in some fourth space where emotional register becomes irrelevant, and the only metric that matters is whether the thing lodges itself in your prefrontal cortex like a splinter melody, if we're being generous enough to call it that, is a three note loop that sounds like it was composed on a Fisher Price keyboard during a power outage. The vocal delivery is flat, almost affectless, a kind of anti performance that paradoxically demands more attention than virtuosity ever could. The lyrical content is, well, the title repeated with conviction. And yet something is happening here that's worth taking seriously, or at least worth resisting the urge to dismiss. The song collapses the distance between tragedy and comedy so completely that neither category survives the collision. It's grief as non sequitur, eulogy as punchline, memorial as munition. The Bazooka isn't just absurd, it's so specifically absurd that it loops past irony and arrives somewhere strangely earnest. Nobody's grandmother has ever been hit by a bazooka. The scenario is impossible, and impossibility, it turns out, is its own form of tenderness, a way of saying loss is so incomprehensible that only nonsense can hold it. Or maybe it's just a guy yelling about a bazooka. The Internet doesn't require you to choose, but we'll be following this story, so thanks for we'll get we'll get John's take. We will the Song of the Week and the.