New York's proposed pied-à-terre tax on second homes over $5 million will crash the high-end real estate market by destroying demand from out-of-town buyers.

(00:00:00) All right, everybody, welcome back to the number one podcast in the world. (00:00:05) We've got the core 4 here and dare I say, The King of Atoms. (00:00:11) The King of Atoms, yes. (00:00:13) Captain Travis Kalanick is here. (00:00:16) How you doing, brother? (00:00:17) I'm pretty good. (00:00:18) Pretty good. (00:00:18) I'm sitting here doing the podcast just next door to David. (00:00:23) yes, there you go. (00:00:24) Don't reveal our locations. (00:00:26) I didn't put my address out there, dude. (00:00:28) No, that's true. (00:00:29) Don't worry, Mandami did. (00:00:31) He's outside your houses right now asking them to foreclose. (00:00:34) Are you trying to collect 3.9% or something? (00:00:37) Yeah, he decided there's more rich people left in New York, so he's looking for other things to tax. (00:00:43) Is it 3.9% a year? (00:00:45) Is it per year or is it? (00:00:46) I don't know if the percentage has been released yet, but the speculation I've seen is 3.9%, but I don't think that's final. (00:00:51) But yeah, it's a Piata tax. (00:00:53) So if you have a second home, every year. (00:00:56) Wow. (00:00:57) And by the way, it's for any home over 5 million. (00:01:01) There's no homes under 5 million in Manhattan. (00:01:03) This is not a rich person tax. (00:01:05) This is within (00:01:08) 15 miles of Midtown Manhattan, you're paying an extra tax. (00:01:12) But only, but J Cal, only if it's a piano tariff. (00:01:15) So what it means is that the most, well, the most elastic part of the market is what they're targeting for this tax. (00:01:21) So in other words, people who don't live in New York, who just have it as a second or third home, who could buy that property anywhere, are now being taxed the most. (00:01:31) So what do you think that's going to do? (00:01:32) It's going to have a massive impact on demand (00:01:36) for second homes in New York, which will crash the whole market. (00:01:39) Yes. (00:01:40) Congratulations, mon dummy. (00:01:41) But in a weird way, that'll be good for housing affordability in New York. (00:01:45) Well, that's sort of the claim, but I don't think it'll be good for it because there'll be no incentive to build more. (00:01:51) Yeah. (00:01:52) All units matter. (00:01:53) Every time you add units, people upgrade. (00:01:55) And it's not like these are going to be low-income housing, like penthouse on 57th Street or (00:02:01) in Gramercy, that's not low-income housing. (00:02:05) You'd have to break it into 7 units. (00:02:07) Makes no sense. (00:02:07) But by the way, I don't know if you guys saw the video, not to get too serious, but he's doxing. (00:02:13) a certain billionaire who owns a certain place and he's literally pointing at his home. (00:02:16) No, I said that to you, Jason. (00:02:18) He's not doxing because everybody's known for years that Ken Griffin bought that place. (00:02:22) Everybody knows that address. (00:02:23) Everybody knows that unit. (00:02:24) We all knew it. (00:02:25) was marketed widely. (00:02:27) I don't think that's really doxing. (00:02:28) He doesn't live there and everybody knew he owned it. (00:02:31) would be very different if it was a place where somebody was keeping their primary residence and you didn't know and they stood in front of the house. (00:02:38) That I would agree with you. (00:02:39) I think this one is a little bit more tenuous. (00:02:40) Okay, fair enough. (00:02:41) But what I will tell you is (00:02:43) It's a dog whistle. (00:02:45) Crazy people, and this thing's been seen by 30, 40, 50 million people now. (00:02:49) It's a dog whistle to say, that's the next UnitedHealthcare CEO. (00:02:53) And in the week that a fire, a Molotov cocktail and a bullet gets shot into Sam Altman's house, it's deadly serious. (00:02:59) And if you reversed it, I always reverse it. (00:03:02) What if a Republican sat outside of Bernie Sanders' second or third home and said, we should, this is his pia de taire, this is his summer home, we should add, (00:03:13) taxes to it. (00:03:13) And I'm sure you could look up Bernie Sanders' home pretty easily. (00:03:15) So just before you point at people's homes and say, this is the villain, be careful, folks, because then if something does happen to that person, like, you know, what happened to Sam Waltman this week, you can feel however you want about him, but nobody deserves to have their house fire bombed or shot at, period, full stop. (00:03:35) I just think that it's going to kill the demand for, maybe people who already have a home in New York, like Ken Griffin, they probably are just going to suck it up and pay the tax and keep whatever they have. (00:03:47) But if you were a person who was thinking about buying a Pia de Terre in New York, there's no way you would do it now. (00:03:55) because you don't know what the tax rate's going to be, and it's going to keep going up. (00:03:58) After a decade with interest and inflation, you've effectively almost doubled the price of your unit. (00:04:03) So if you're going to buy a $10 million unit, (00:04:07) you're probably then looking at a $20 million purchase price just to break even after about 10 or 11 years. (00:04:13) That's crazy. (00:04:14) The math doesn't work anymore. (00:04:15) What is the downstream effect of these individuals not coming to New York, going to a Knicks game, going to a restaurant? (00:04:22) No, they spend money. (00:04:23) They spend money in New York. (00:04:24) They have their birthday party. (00:04:25) They do all kinds of things. (00:04:27) I'm not a big fan of the Piet A Terre culture. (00:04:29) I own one house. (00:04:30) I don't like this whole multiple houses, you flitter flattering everywhere. (00:04:33) It ruined London. (00:04:35) Shamath, where do you vacation? (00:04:38) Shamath, where do you vacation? (00:04:39) I go to a hotel. (00:04:40) Okay. (00:04:42) So my point is this, that I think the way that it works is it's not just pied-a-terre. (00:04:47) If you own a home that you fully rent out, but that is not the primary residence of the person who is renting it, it's also a problem. (00:04:59) So this is any, even like you had a two-week rental or a one-week rental or a 30-day rental, or even a five-year rental and it was somebody else's second home, the owner still gets hit. (00:05:13) So the rental, the sort of rentals of like folks who want to come to that city or who don't wanna own in that city, but it's a second place, the second place thing goes away. (00:05:24) So maybe this is just good for hotels. (00:05:27) I think it's good for hotels, but if you look at London, it's probably the best example where there was a lot of people who used London as a place to store assets. (00:05:35) Real estate became the primary way in which they would do that. (00:05:39) It hollowed out parts of London where it's not as if it was unlivable, it's just that it was unlived. (00:05:46) Nobody was there. (00:05:47) You'd drive around Chelsea, you'd drive around (00:05:50) certain parts of London, and it was like a ghost town on like a Thursday night or a Friday night. (00:05:55) That is an issue, is the land banking. (00:05:57) It's like your Bitcoin is hollowing out a neighborhood, and there is something there. (00:06:02) But you know what? (00:06:02) The whole housing thing is complete cap and utter bullshit, because if you move to a place like Austin, where you are in Nevada or Florida, you see what happens when you allow people to build units. (00:06:16) In Austin, three years in a row, rents, (00:06:19) And housing prices have gone down while net migration has gone up. (00:06:23) But that's been good rent for Austin, you guys would say? (00:06:26) That's incredible. (00:06:27) Yeah, amazing. (00:06:28) Austin has like roughly doubled as a city over the past decade and yet the rent for, you know, whatever, a one or two-bedroom apartment's gone down. (00:06:35) That's incredible. (00:06:37) So in other words, if you let people build to satisfy the demand, you won't have this problem. (00:06:43) And then who's stopping the building? (00:06:46) It's Democratic cities. (00:06:47) It's NIMBY people. (00:06:49) And then in a Republican town, they're actually building units for an affordability. (00:06:55) So you have one group saying they care about affordability and they're doing nothing about it and they're stopping it. (00:07:00) And in another place, they're like, we're just going to let you build because it's your right to build because it's your land. (00:07:05) That's the approach in Texas. (00:07:06) It's your land. (00:07:07) You have the right to develop it. (00:07:09) Go. (00:07:09) By the way, the high end of the market in London has basically turned over and collapsed, Sacks, to your point, like they introduced the stamp tax. (00:07:16) which I think is equivalent to this tax that the New York City mayor is proposing. (00:07:21) And if you look at London as a guide, the real estate market just bid it at the high end. (00:07:26) I don't think that's going to be good for that city or even the UK as a whole. (00:07:30) There's A handful of cities where people do what you say, which is kind of park money there. (00:07:34) The reason why they do that is because they believe that city, A, has the rule of law and B, is a unique world-class city that's going to keep appreciating. (00:07:43) And if you have new management that doesn't really believe in the rule of law, that basically keeps imposing all these arbitrary taxes, that money is going to flee and find other kinds of investments. (00:07:52) People aren't going to park their money there. (00:07:54) That has to be a bad thing for the city. (00:07:56) It's like, you know, who cares if the top floor of that building where Ken Griffin lives is owned by one guy who's up there that much. (00:08:02) It's not going to affect the city that much. (00:08:04) But having all these billionaires from all over the world decide to park money in that American city has to be good for the US. (00:08:11) just like it's good for the UK. (00:08:14) And if you give that up, then again, the money will just go somewhere else. (00:08:20) Well, to your point- The other thing is they're already paying taxes on the property. (00:08:24) And because they're not there very often, they're not using city services. (00:08:29) So they're paying taxes on the property. (00:08:31) They're not using city services. (00:08:33) They're profitable to the city, yeah. (00:08:35) Think about a developer who's underwriting some new project. (00:08:39) the fact that a whale like Ken Griffin is willing to overpay in the sense of price per foot for that top floor might make that whole project pencil. (00:08:50) And this is what contributes to the vitality of New York, is there's constantly development going on. (00:08:55) There's constantly cranes. (00:08:57) And so you take out that part of the market that in effect was price insensitive and was subsidizing all these projects. (00:09:06) And I think development's going to dry up to a large degree. (00:09:09) He's paying, he must be paying 3 or 4 million in taxes every year and getting no services to your point, Travis. (00:09:14) Good point. (00:09:14) Like all profit for the city. (00:09:16) London also did something else, which is that they essentially crippled what's called non-dom status, which is the big tax arb if you're moving or parking assets in London. (00:09:25) And to your point, Sax, what did all the rich people do? (00:09:27) They just redirected themselves to Zurich, to Lugano, to Milan. (00:09:32) and they took advantage of more hospitable tax policy in other places. (00:09:37) Now, what the people in Britain would tell you, I actually met with the UK government yesterday, they don't see a meaningfully enough measurable impact yet where they think it's a five alarm fire. (00:09:47) So the real question is, it more of a slow bleed and a slow melt? (00:09:52) And at some point, it's just hollowed out and it's very hard to reverse because there's not going to be a cataclysmic acute moment where people say, oh my God, we need to reverse these policies. (00:10:01) It doesn't seem like that's in the offing. (00:10:03) LA did something a little different but similar is they introduced that 5% mansion tax. (00:10:09) It's on like all the areas except for like Beverly Hills and San Francisco has that too. (00:10:13) Yeah, San Francisco got that too. (00:10:14) I think it's actually like 6%. (00:10:16) In San Francisco, it's 6% over 25 million and it starts at 5 million. (00:10:22) So there's a transfer tax, an excess transfer tax above $5 million properties that scales up above 25 million to an extra 6% on top of your brokerage commission. (00:10:32) So when you sell a house in San Francisco, you're paying 13% now. (00:10:37) Which is why you look at the transaction volume in the LA real estate market and it's just completely dried up. (00:10:43) people aren't doing the house flipping anymore, like that kind of stuff, because the transaction costs are too high. (00:10:48) But it just shows, I mean, again, this tax was imposed retroactively. (00:10:52) For example, I had my house in SF, and then they just take a couple of rooms of it. (00:10:58) And we talked about this on the show when it happened. (00:11:02) But my point is just your property is not safe in blue states. (00:11:05) And wealthy people who have a choice of where to park their money are going to increasingly realize that. (00:11:11) And (00:11:12) they're not going to buy. (00:11:13) I think real estate in blue states is dangerous because the political class thinks that they can take a chunk of it. (00:11:20) And wealthy people are going to react to that and they're going to move their money elsewhere. (00:11:23) All right, let's go to topic number one. (00:11:26) OpenAI is apparently suffering from a bit of an identity crisis on Sunday. (00:11:31) OpenAI's chief revenue officer, Denise Dresser, sent a four-page memo to employees (00:11:37) Obviously, it leaked immediately, probably the point of it. (00:11:39) And she called out Anthropic. (00:11:41) She said their $30 billion run rate is cap, inflated by 8 billion due to a revenue share and some accounting with AI model providers. (00:11:51) Jamathi pointed that out in the last couple of weeks. (00:11:53) Also, she said Anthropic's story is built on, quote, fear, restriction, and the idea that a small group of elites should control AI. (00:12:01) Obviously, she's a fan of the pod. (00:12:03) She also laid out OpenAI's pivoting, and she said (00:12:07) that they are going hard after business customers and they want to win the agent platform layer. (00:12:14) If you remember, they hired the architect of the open source project OpenClaw. (00:12:18) They didn't acquire OpenClaw. (00:12:19) So Peter Steinberger is working at OpenAI. (00:12:23) Cynical people said, hey, maybe they want his next set of innovations to go inside of OpenAI's products as opposed to the open source one. (00:12:29) I kind of agree with that directionally. (00:12:32) There's obviously Perplexity Computer is doing really well. (00:12:34) They quadrupled their revenue. (00:12:36) And I was over at XAI earlier this week with Elon, and I can tell you he's got some very cool stuff coming. (00:12:43) And this new model, SPUD, is coming from OpenAI. (00:12:48) Here's your polymarket, 75% chance SPUD is released next week. (00:12:55) Additionally, and we'll go to the panel in just a second, on Tuesday, two days after this memo came out, obviously people write these memos to go directly to the press. (00:13:03) So they're obviously trying to undercut Anthropic's valuation and they feel that's a threat. (00:13:07) That's my take. (00:13:08) The FT cited anonymous OpenAI investors who are frustrated with the company's lack of focus. (00:13:14) Here's the anonymous quote. (00:13:16) Quote, you have ChatGPT, a 1 billion user business growing 50 to 100% a year. (00:13:22) What are you doing? (00:13:23) Talking about enterprise and code. (00:13:25) It's A deeply unfocused company. (00:13:28) And we talked about this. (00:13:29) ChatGPT's market share is going down as the number of users is going up because Gemini and Claude gaining significantly. (00:13:37) And Meta just last week released their first proprietary model. (00:13:41) And Apple doesn't have a product in market yet, but they do have a lot of users. (00:13:45) So here's your Gen. (00:13:46) AI website traffic. (00:13:50) Let's start, Chamath, with you, your thoughts on OpenAI. (00:13:55) Should they be pivoting straight into business developers and getting focused on that? (00:14:00) Or should they stay focused on the consumer where they are the verb? (00:14:03) As it relates to complex, long horizon coding tasks, what I can tell you from my team at 8090 is Codex is better generally than Anthropic. (00:14:15) And so what happens is from the more day-to-day work, (00:14:19) I think it's more reliable to use the Claude ensemble of models. (00:14:26) But when you're dealing with something that's very tricky and very complicated and a little bit more long horizon, Codex is really functional and really good. (00:14:35) So I think if you're looking at it through the lens of OpenAI, what they're probably saying is, hey, hold on a second. (00:14:42) If we allocate our resources and just double down and crush consumer, (00:14:47) That's probably 3 or 4 trillion of enterprise value. (00:14:52) And then if we slowly refocus the company with the rest of the resources and double down on Codex and do something meaningful in enterprise, we can probably capture 2 or 3 trillion there. (00:15:01) And now all of a sudden, you can paint a picture for a (00:15:05) 7, 8, $9 trillion market cap in the fullness of time, not tomorrow, obviously. (00:15:09) But Codex is really good, and there is a business to be had in both. (00:15:13) You have to separate the two businesses. (00:15:14) You can't have a lot of overlap because there's too much context switching. (00:15:18) You got to let the consumer team run, and then you got to isolate the enterprise team and let them do what they think is right. (00:15:23) Travis, there's a big debate going on amongst the investors. (00:15:27) The FT piece also questioned the $850 billion valuation of OpenAI secondary markets. (00:15:34) have now priced Anthropic higher than OpenAI for the first time. (00:15:37) This is the flippening that people predicted. (00:15:40) One investor said OpenAI would need to IPO at a valuation of 1.2 trillion for the last round to make any sense. (00:15:46) But there's no buyers currently at the $850 billion valuation that OpenAI just closed, according to Bloomberg. (00:15:56) Travis, how do you handicap this race between these two leading frontier models? (00:16:01) Growth is king right now in this world, in this segment. (00:16:06) Growth is the whole damn thing. (00:16:08) And if Anthropic is growing faster than OpenAI by a significant clip, the investors right now are going to play it forward. (00:16:18) And you start to get network effects around compute, network effects around the number of tokens you're pushing out for various customers, enterprise or consumer. (00:16:29) And ultimately, it's not great today, but how that plays into reinforcement learning and the things getting smarter over time. (00:16:37) There's so much upside to volume and scale that if they are growing faster at the same size, even if OpenAI is still growing, but if they are half the growth rate, a third the growth rate, a fifth the growth rate, I'd be worried. (00:16:58) And you saw this at Uber specifically when you accelerated away from the competitors like Lyft and DoorDash, yeah? (00:17:05) Yeah, you had to. (00:17:06) Network effects was the whole thing at the end of the day. (00:17:09) And network effects was based on scale. (00:17:11) So yes, I'm sort of like coming from my very specific experience. (00:17:15) But if you believe there's network effects with the scale of data that you have and the scale of customers and the revenue, (00:17:22) that's just cash that's coming in that you redeploy into compute. (00:17:26) And so say there's network effects of large compute, I'd be very worried if I'm OpenAI and seeing somebody growing faster at the same size. (00:17:37) So Freeberg, when you look at this race between these two giants, maybe your thoughts on the flywheel as it relates, as Travis is pointing out, (00:17:46) to advantages in compute, reinforcement learning, and then also the ability to fundraise. (00:17:52) One of the great things that Travis and the team did was as they were pulling away, they just sucked all the oxygen out of the room by using capital as a weapon. (00:18:02) So your thoughts, Dave, on this high stakes game, because there's also a point at which, and I'll just end on this, there's a point at which you could run off the cliff. (00:18:10) You raise so much money (00:18:12) and you deploy it so fast and the revenue doesn't catch up to it, and then you go public and the markets don't believe the story. (00:18:17) So your thoughts, Reber? (00:18:18) I don't know the financials of the two companies well enough. (00:18:21) Obviously, Sam has no problem raising money. (00:18:23) The guy, didn't he just close like $150 billion round or something? (00:18:26) How much is it? (00:18:27) 122 billion. (00:18:28) 122 billion. (00:18:29) Largest round ever raised in any market, I think, private or public. (00:18:33) Yeah, probably, yeah. (00:18:34) I mean, that's crazy. (00:18:34) So that doesn't seem to be an issue. (00:18:37) What I've noticed is just the pace of innovation at Anthropic is (00:18:42) from my experience, unprecedented. (00:18:44) I mean, their release cadence is extraordinary. (00:18:47) They've basically supplanted OpenClaw already with this release they did a few days ago, and then today the new Opus model got dropped. (00:18:55) So there's something about the momentum, not necessarily just in user growth, but in how they're operating this business. (00:19:03) that just seems to be head and shoulders above everyone else in the cadence of upgrades. (00:19:08) If I look back six months ago, I think we were pretty heavy on Cursor and Gemini, and now I think we're probably 90% anthropic in just the last six months at my organization. (00:19:20) There's something very powerful about the flywheel they have going on. (00:19:23) Yeah, but here's where I'd go, just real quick, Dave, is (00:19:27) mad respect on a 120 or $130 billion raise. (00:19:32) I mean, this is obviously next level, but you can use capital and investment to acquire scale and network effects associated with it. (00:19:43) But if somebody is getting that scale with revenue and let's call it contribution margin, contribution profit, efficiency will outstrip subsidy. (00:19:56) And you can't just keep raising $100 billion things forever. (00:19:59) That's where... (00:20:00) The train will stop. (00:20:02) And if Anthropic is funding theirs through revenue and other folks are funding it through investment, there's like a short-term, that's a short-term solve. (00:20:13) But the long run is whoever is scaling their actual usage and system and ultimately with contribution profit that then soaks up the need for investment. (00:20:26) That's a very scary machine if you're competing against it. (00:20:30) which is exactly sacks what the legacy Mag 7 are doing. (00:20:35) You have massive profits from Meta's core business, Google's core business that are being redeployed into infrastructure and even Tesla. (00:20:46) which has a lot of profits and SpaceX, which has a lot of cash on hand. (00:20:49) And they're building out Colossus and other assets, including Elon is working on building a fab, as folks have been talking about. (00:20:58) So is there a chance for the legacy companies, the Max 7s, to compete in this, or are we looking at OpenAI and Anthropic are one and two, and then everybody else can fight for 3rd place and the bronze, as it were? (00:21:14) I think Google's clearly in the mix. (00:21:16) I mean, DeepMind has an outstanding team, and I think Elon's still in the mix with xAI, and then you've got Meta also has the resources, and they seem to be further behind, but they're going to compete. (00:21:28) Look, let me just go back to the central premise here. (00:21:31) I agree that there's some valid criticism that OpenAI has been unfocused and should be, you know, moving forward, more focused in (00:21:43) what they do. (00:21:44) I have no idea, for example, what they're doing buying a podcast that's not us. (00:21:49) So I don't know what that was about. (00:21:52) If you were going to buy a tech podcast for a few $100 million, I mean, we were here. (00:21:57) We're here. (00:21:58) We're here. (00:21:59) Dario, reach out. (00:22:00) Small potatoes for you guys, though. (00:22:02) Small potatoes. (00:22:03) They couldn't afford you. (00:22:05) They can't afford you guys. (00:22:06) Yeah, they thought we were too expensive. (00:22:08) Little did they know, we would have sold out. (00:22:12) Punch the ticket, Daria. (00:22:15) But look, this other part of the criticism of OpenAI that they shouldn't do enterprise is totally misguided. (00:22:22) One of the reasons why they should have been more focused is to do more enterprise and get enterprise more correct. (00:22:28) Now, why do I say that? (00:22:30) To Travis's point about growth rates, (00:22:33) It's true that OpenAI and Anthropic, as of the beginning of Q2, so let's say 2 weeks ago, they were both around 30 billion of revenue. (00:22:42) And that memo from that OpenAI employee was right, that if you compare them on an apples to apples basis, that Anthropic is about 20% less because they are including revenue made by their channel partners. (00:22:54) But that doesn't matter. (00:22:56) What matters is the growth rate, again, to Travis's point. (00:22:59) And (00:23:00) Let me just put some numbers around this. (00:23:02) The OpenAI growth rate's been around 3 to 4X a year. (00:23:05) The anthropic growth rate has been around 10X a year. (00:23:09) So they went from, let's call it 1 to 10 billion of ARR last year. (00:23:14) And by the end of Q1 this year, they were already at 30 billion of, again, let's call it their revenue. (00:23:21) And they're on their way, like Brad Gersten was saying on our podcast, I think in the last couple of weeks, they're going to end this year at 80 to 100 billion, at least on the current trajectory. (00:23:31) And so you can plot their revenue on a logarithmic graph. (00:23:35) I mean, again, no one's ever seen anything like it before, where every unit on the y-axis is another, it's 10x, and it's a straight line. (00:23:43) It's crazy. (00:23:44) It's crazy, right? (00:23:45) So if it's taking Anthropic, let's say one year to 10X and it's taking OpenAI two years to achieve a 10X, then it's obvious which one's going to win. (00:23:56) Now, what is the reason for this? (00:23:57) It's because Anthropic was very focused on enterprise, specifically coding. (00:24:03) And what you're seeing is that businesses are willing to pay for coding, code tokens, (00:24:08) on a metered basis, let's call it like electricity, the more they use, the more they're willing to pay, and their usage continues to scale and scale. (00:24:17) Consumer is completely different. (00:24:19) I mean, consumer is a thing that OpenAI prioritize. (00:24:21) Consumers have a lower willingness to pay, maybe only 3 or 4% of them are willing to convert to premium in the 1st place. (00:24:27) And what they want is a $20 a month, all-you-can-eat subscription. (00:24:31) So the revenue simply doesn't scale the same way that enterprise does. (00:24:34) And so if you want to tap into the scalable revenue source in the market right now, you have to go after enterprise. (00:24:41) So again, where I would agree with the criticism of OpenAI is maybe they should have been more focused, but they need to be more focused specifically to pursue coding and enterprise. (00:24:54) And if they don't catch up soon, to Travis's point, then you could see Anthropic taking a lead here that (00:25:00) let's say over the next one or two years, could be insurmountable. (00:25:03) Just by the way, let me just say one thing is, even though Anthropics revenue has followed this graph, this exponential graph very predictably, it can't do that forever, right? (00:25:13) Like, let's say it does get to... (00:25:15) into 100 billion this year. (00:25:16) Can it really get to a trillion dollars in revenue the year after that? (00:25:21) Seems hard to believe, right? (00:25:23) And the reason is because as you hit new levels of scale, you encounter new problems. (00:25:27) I mean, you're simply going to run out of compute or electricity, data centers, infrastructure. (00:25:33) There are physical limits or there's limits in the physical world that you're going to hit. (00:25:37) And there's already some evidence that Anthropic is hitting some of those limits. (00:25:42) users were complaining, for example, that Claude was thinking less. (00:25:46) Did you guys see this? (00:25:46) That, you know, a typical Claude prompt, they seem to have cut down on the thinking time by about two-thirds. (00:25:53) Now, I saw someone tweeting today that they just released Opus 4.7, replacing Opus 4.6, and the thinking is back. (00:26:00) But, you know, maybe they're charging more for that. (00:26:02) Hard to say. (00:26:03) But they're going to hit some sort of physical limits. (00:26:05) And I do wonder if over the next year, Anthropic will (00:26:10) reconsider whether it's support for all this like doomer NIMBYism was the right call because it kind of made sense for them from a business standpoint when their competitors were building data centers and they were just getting compute from the hyperscalers. (00:26:24) But now that they're, I think, going to have to move into the game of building their own data centers, they might regret salting the earth for data centers all over the country. (00:26:34) And I wonder if that'll be the natural limit (00:26:38) of their growth is they'll be wasted on their own petard of doomer NIMBYism. (00:26:42) And they're also using the coding platform to build Anthropic itself better. (00:26:48) So that in and of itself is a reason to now coding. (00:26:51) You get the double whammy. (00:26:52) You can make a better product and you can get paid for it. (00:26:55) Yes and no. (00:26:56) Hold on. (00:26:56) Because if you look on Twitter, people are panning Claude Desktop. (00:26:59) And what they said is this is all a bunch of vibe coded slop. (00:27:02) I think we have to remember, if you keep these agents on task and they're guardrailed properly, (00:27:08) This stuff is a force multiplier. (00:27:10) The problem is nobody knows how to do this really well yet. (00:27:13) Nobody has built real products of scale, largely using agents yet. (00:27:18) Nobody knows how to give an example of how an org structure should be redefined. (00:27:23) Nobody knows how to budget properly. (00:27:25) We had the CTO of Uber say, I give up, I've hit my token budget. (00:27:29) The problems I see it are twofold. (00:27:32) The first is just to build on Sax's point. (00:27:36) All of these frontier labs have a very serious issue, which is both OpenAI and Anthropic are growing so fast. (00:27:43) that they're at a point now where they need their own infrastructure. (00:27:46) It's kind of like when you first start building any kind of company, you're just much easier renting capacity from the hyperscalers. (00:27:54) And it's a dependency. (00:27:55) Yeah. (00:27:55) But then it becomes a dependency. (00:27:56) Exactly. (00:27:57) And now when you're so big, it's actually strategically a huge mistake to not have your own compute supply. (00:28:04) Why? (00:28:05) Because if you look at who's leading, the Frontier Labs are leading, and Sachs, to your point, you mentioned this on X, you're like, there was all this doomerism, but maybe it was tied to compute capacity because when Bedrock opened up more capacity for Anthropic, all the doomerism went away. (00:28:19) You're left wondering, like, is it really tied to just the fact that they were just trying to throttle usage? (00:28:24) So if you're a Frontier Lab, you don't want to have to go through Amazon and GCP and Azure and Tin Cup for access and capacity. (00:28:31) What you'd much rather have is go straight to your customer. (00:28:34) On the other side, (00:28:36) If you're Gemini or Microsoft or Meta and you have all this compute, because I saw a stat this week, the hyperscalers control 60% of all the compute. (00:28:48) So the game theory there is if you kneecap the frontier labs, it'll give you some chance to catch up. (00:28:52) And it gives you time to catch up because no matter what the demand is on the upside, you remember, you guys remember like in social networking when Friendster was the cat's meow? (00:29:01) Yeah. (00:29:02) Cat's meow. (00:29:03) Remember what the biggest problem at Friendster was? (00:29:06) Friendster was slow as a dog. (00:29:09) Yes. (00:29:09) And what happened? (00:29:10) Then MySpace came in and took all the share. (00:29:12) Then we came in Facebook and we took all their share. (00:29:15) So there is a way where you can handicap and kneecap these companies by throttling compute access to them. (00:29:22) So A, they are forced to now go and get in the game, which is weird because look, OpenAI has tried to displace some of the Stargate spend. (00:29:31) I don't see any path except they're going to have to do it themselves. (00:29:33) And Anthropic will have to do it themselves. (00:29:35) But then separately, the other problem is when you change the subscription model in Enterprise and you say, hey, we're not going to subsidize any more tokens, what's going to happen is all these token budgets are going to go crazy. (00:29:47) And what Friedberg said is going to happen where he's like, hey guys, why are you spending all this money? (00:29:51) What are you making? (00:29:52) And you inspect the code and you're like, what is this slop? (00:29:56) And you're not going to add 30, 40, 50% OPEX. (00:30:00) to produce nothing. (00:30:02) So I think that that's an open question and that question will become more amplified over the next year as they push the cost off of them. (00:30:10) So as Travis said, no more subsidy from the capital. (00:30:13) You have to grow into it, but you're not going to support negative gross margins. (00:30:16) So you're going to pass through the token costs. (00:30:19) So I think it's a very dynamic moment right now for these. (00:30:21) And if you look at the ranking of these clusters and who has the most, (00:30:26) Right now, people have forgotten about Colossus, which Elon's been building. (00:30:30) He's expanding to 555,000 GPUs across three buildings, 18 billion in investment. (00:30:38) And then if you look at Prometheus, Meta's planned 2026, that's 150,000 GPUs. (00:30:44) So this is... (00:30:46) Well, Elon just announced a deal. (00:30:47) He announced a deal this morning with Cursor. (00:30:49) So Elon's renting a bunch of capacity. (00:30:52) So he's now getting effectively into the data center business. (00:30:54) He's going to be a hyperscale. (00:30:56) So he's going to use as much as he can for XAI and whatever's left over, he'll give to, well, in this case, he's giving to Cursor to train their model. (00:31:03) He could give. (00:31:03) So what you're saying is you might as well overbuild capacity because that way your own models will be in a privileged position and you can sell the rest to your competitors. (00:31:12) 100%. (00:31:12) But to your point about the thing I was saying on X, I actually, I think I was retweeting Marc Andreessen, who pointed out that one of the reasons why Anthropic might have wanted to hold back Mythos (00:31:24) is they simply didn't have the compute to serve it. (00:31:26) The model was huge and very expensive to serve. (00:31:29) Something like maybe even 10 or 20 times the token cost of, say, Opus. (00:31:34) They knew Opus 4.7 was coming out, right? (00:31:37) So they hold it back knowing that they don't have the compute to serve it anyway, and they save their compute for the next iteration of Opus. (00:31:45) And then by holding it back, they create this impression of scarcity and altruism, and it turns into this gigantic marketing event (00:31:53) for their product because everyone in the government's like, oh wow, they're holding it back because it's so amazing. (00:31:58) Now look, I think it may have been genuinely altruistic as well in the sense that Mythos does reveal coding vulnerabilities that people didn't know about before. (00:32:09) And it does make sense to give time to companies with large code bases to patch these dormant bugs and vulnerabilities. (00:32:17) But (00:32:18) It's looking more and more like Anthropic could not have offered that model commercially anyway, because it was just too big and expensive. (00:32:25) And they needed to create space for OPUS 4.7. (00:32:29) So it's an interesting theory on what actually happened there. (00:32:33) All right, guys, before we go to our next story, some breaking news here. (00:32:37) Here's your poly market, gentlemen. (00:32:39) I don't know if you're placing some insider bets here, Friedberg, but looks like the All In podcast. (00:32:45) Anthropic now, 37% chance of buying the All In podcast. (00:32:48) This is live. (00:32:49) At what time? (00:32:50) From Polymarket. (00:32:51) Yeah, this is by the end of the year. (00:32:53) By end of the year. (00:32:54) Okay, got it. (00:32:54) By end of the year. (00:32:55) So it's. (00:32:55) This isn't real, is it? (00:32:57) Yeah, absolutely. (00:32:57) They've been people have been trading on this. (00:32:59) is heavily traded. (00:33:00) It can't be. (00:33:02) It says $92 million of volume. (00:33:03) This is the number three largest calling market rate. (00:33:06) There's no way this is real, dude. (00:33:08) There's no way this is. (00:33:09) It's breaking news, guys. (00:33:10) I don't control the news flow. (00:33:12) This is JCal's slump. (00:33:13) This is no way this is real. (00:33:15) That's what this is real. (00:33:16) Anybody can create a prediction market, I guess. (00:33:18) Hey, what's the volume of that thing? (00:33:20) 92 million. (00:33:21) It's not real. (00:33:24) There's no way that's real, guys. (00:33:26) Come on. (00:33:26) Guys, it's all good. (00:33:27) Guys, time to upgrade the planes. (00:33:31) I just, hey guys, just I don't need the PC-24. (00:33:34) I'll take somebody's G650, whoever's got a G650, go up to the 800. (00:33:38) It's going to be trickle down to JCAL. (00:33:39) Let's do it. (00:33:40) I'm giving up my United Platinum status. (00:33:43) Trickle down economics. (00:33:44) Trickle down avionics. (00:33:45) All right, listen, story #2. (00:33:47) Trickle down avionics. (00:33:52) This is pretty good. (00:33:53) Okay, #2, Allbirds. (00:33:55) Speaking of data centers, Allbirds just pivoted from ugly sneakers, discraziade to AI, and the stock has ripped. (00:34:04) Talking about puke bubble behavior, podcasts getting bought by Frontier Models and sneaker companies pivoting to data centers, Allbirds, as you know, is the ugliest sneakers on the planet. (00:34:17) And this became a massive delusion in our industry that this company was worth billions of dollars. (00:34:24) They went public in 2021. (00:34:27) This might be one of the peak Zerk moments, raised $350 million in their IPO. (00:34:32) Sacks, were you an investor in this thing? (00:34:34) Allbirds? (00:34:36) Allbirds, back-end? (00:34:37) No. (00:34:38) You're thinking about the scooter company. (00:34:39) So the scooter company. (00:34:40) That's right. (00:34:41) Okay. (00:34:41) Don't remind me of the investments that didn't work. (00:34:45) But no. (00:34:46) Bird was crazy. (00:34:48) This reminds me of the late 90s where all you had to do was change your name to whatever.com. (00:34:54) Yeah. (00:34:55) Yes. (00:34:55) And you get a huge pop in your valuation. (00:34:58) And you could spin it out. (00:34:58) Barnes and nobles.com became, and then Barnes and Nobles. (00:35:03) And so the stock crashed and never looked back. (00:35:06) There's your stock chart. (00:35:07) They sold off all their brand assets for 39 million, about 10% of what they raised in the IPO. (00:35:13) Congratulations to whoever owns those ugly ass sneakers. (00:35:16) Why were they worth 4 billion? (00:35:17) They sold sneakers. (00:35:18) Collective delusion. (00:35:19) Collective delusion. (00:35:20) People in Silicon Valley like them and they just thought he was the next Nike. (00:35:24) Tulips. (00:35:25) It was Tulip. (00:35:26) I think that was an era in Silicon Valley where people rewarded rapid growth without really looking at gross margins or cost of goods sold. (00:35:35) People didn't really make the distinction between software and everything else, right? (00:35:40) Software, you never really had to worry about COGS or gross margin because the incremental cost of serving a customer with software is like almost zero. (00:35:50) So people in Silicon Valley weren't really trained to look at gross margin and there was this (00:35:54) rash of physical world companies that all of a sudden start getting crazy valuations. (00:35:59) Travis, quite frankly, you might have, the success of Uber might have ushered in this era of these physical world companies that started getting valued like software companies, even though obviously they didn't deserve it. (00:36:10) David, thank you so much for that shout. (00:36:12) I really appreciate it. (00:36:14) You and Airbnb. (00:36:16) Anyway, let's get back to. (00:36:17) No, hold on, hold on. (00:36:18) The error of 21 is super interesting though, guys. (00:36:21) I don't believe it was a physical versus digital thing. (00:36:24) I believe it was (00:36:26) a moment in time in the COVID, ZIRP, massive money going in. (00:36:31) We hadn't seen inflation yet. (00:36:33) It was all of that happening. (00:36:35) And the investor class was basically deciding, we are going to look two and three years forward on your current growth. (00:36:43) And it wasn't one year forward, it was like two or three years forward, and that's where these crazy valuations got weird. (00:36:49) So if you went from zero to 100, if you went from 100 to $300 million in the last year, they'll play that two, three years forward and go, oh yeah, you're totally 20 times bigger. (00:37:01) We'll pay you for that now. (00:37:03) That's where it got weird. (00:37:05) It got weird. (00:37:05) Hey, Bird Scooters was valued at 2 or 3 billion. (00:37:08) They were doing micro mobility. (00:37:09) You're just trolling them, Jay. (00:37:11) No, we're giving everybody their flowers for incredible investments. (00:37:15) And even the Mighty Sacks could trip up. (00:37:18) Did you sell those series A? (00:37:21) No, we don't do that. (00:37:22) It was a series A we did. (00:37:24) I mean, look, we got it right in the sense that it was a total phenomenon. (00:37:28) But then the cities just cracked down on it and killed it. (00:37:31) Look, if the cities had leaned into it. (00:37:33) Yeah, look, if the cities had reacted differently, (00:37:35) if they had leaned into it, if they had created, let's say, a scooter or like a small EV lane, it could have transformed cities. (00:37:43) I mean, it would have been a lot easier to get around, but instead they banned it, they limited it. (00:37:46) They didn't create designated areas for it. (00:37:49) And then the coup de grace was basically when they would take a city where Bird already had like 80% dominant market share. (00:37:57) And then they would say that, well, we're going to choose 4 operators. (00:38:01) and give them each 25% of, say, 1000 scooter allocation, that just killed the economics for everybody. (00:38:07) Yeah, there's no market anymore. (00:38:08) There's no market, yeah. (00:38:09) There is no marketplace network effect, right? (00:38:12) When they're basically just picking the winners and deciding the market. (00:38:15) Yeah, you can't compete. (00:38:17) Yeah, there's no competition. (00:38:19) You can't drive value. (00:38:20) Regulatory capture at its worst. (00:38:23) Guys, like we should watch for this in the autonomous car space too. (00:38:27) Cities may get cute and start doing things like that. (00:38:30) And what they did on scooters, they could do on cars. (00:38:34) That's a really good point. (00:38:34) Yeah, you're right. (00:38:35) Instead of just letting the market play out, they say, well, we're only going to have X number, 1000 autonomous cars. (00:38:42) New York is literally doing this right now. (00:38:45) They're doing New York and Boston. (00:38:47) And then the prices go up. (00:38:49) The innovation doesn't get realized. (00:38:51) Consumers don't benefit. (00:38:52) No one benefits. (00:38:53) Basically, you've deleted the market. (00:38:55) Yes. (00:38:57) And anyway, just to wrap up this story on (00:39:00) the shoe company. (00:39:01) They're now New Bird AI. (00:39:02) They bought 8 H100s, I think, with $50 million in a convertible note. (00:39:08) And the stock has gone up. (00:39:10) Not 8, stop. (00:39:11) 8, really? (00:39:12) No, it's not 8. (00:39:13) It's a joke, obviously. (00:39:14) The stock's now at $14 a share. (00:39:17) It's up 450% in the last week. (00:39:18) Shout out to Wall Street Bets for staying retarded more than the shorts can stay solvent. (00:39:26) Can I say something serious? (00:39:27) Yes, please. (00:39:28) There are a handful of transactions that have happened in the last few days that if you look far away, are head scratchers. (00:39:36) So this is one. (00:39:38) I don't know if you guys saw recently, but Jane Street did a billion dollar investment in essentially a neoscaler and then also did a $6 billion compute deal with them. (00:39:50) That was like a little interesting. (00:39:52) So (00:39:53) What would I like to say about this? (00:39:55) I think the thing that the capital markets are getting right is that we are massively computing.