Despite 18 months of predictions that open source would kill the frontier labs, the share of economic value is actually increasing for frontier models while commodity tokens go to the rest—there is no evidence the intelligence gap is collapsing.

(00:00:00) All right, everybody. (00:00:01) Welcome back. (00:00:02) Number one podcast in the world. (00:00:05) It's July. (00:00:06) All in episode 280. (00:00:09) Freiburg is on a little vacay. (00:00:12) We'll leave it at that. (00:00:13) And yeah, bestie Brad is here. (00:00:15) How you doing, Brad? (00:00:16) I'm doing great. (00:00:17) I'm doing vacay in maybe Idaho or somewhere. (00:00:20) J Cal. (00:00:21) Who knows? (00:00:22) Who knows? (00:00:23) Who knows? (00:00:23) It could be anywhere. (00:00:24) It could be anywhere. (00:00:25) I mean, there's lots of things. (00:00:26) He could be in plenty of places. (00:00:28) And looks like you are somewhere in the Northeast. (00:00:32) I'll leave it at that. (00:00:32) You having a little vacay for yourself this week? (00:00:36) I'm in my flag room, very patriotic room here. (00:00:38) Very nice. (00:00:39) You know, where I work on the East Coast in the summertime and spent some time in DC this week. (00:00:45) And it's been a great week. (00:00:47) Been a great week celebrating America 250. (00:00:49) Great. (00:00:49) And you're going to be out of there for the by the second week of August. (00:00:53) Yeah. (00:00:53) So I have it August 10th through the 30th. (00:00:55) I'm good. (00:00:56) Exactly. (00:00:59) Jason B&B. (00:01:01) Oh, absolutely. (00:01:03) You don't know the half of it, man. (00:01:04) I am on a summer bender. (00:01:06) I'm like, oh, where's your, where's your vacation? (00:01:08) By the way, where are you? (00:01:09) Where are you right now? (00:01:09) I am in Paris. (00:01:10) I did about 8 interviews for at the RAISE conference. (00:01:14) They'll be coming out in the all-in feed. (00:01:17) And of course, tackling from the factory. (00:01:19) Look at him. (00:01:20) You're working in the factory. (00:01:22) Chamath Palihapitiya. (00:01:24) It's going to be a hot software summer for Chamath. (00:01:27) How's your hot software summer going? (00:01:28) It's good. (00:01:29) Selling enterprise software is hard, but it's good. (00:01:31) Chamath's like, man, I was such a dick to all my CEOs in the SAS period. (00:01:38) And now you know. (00:01:39) I went to Geneva. (00:01:40) Shout out to Mark Benioff. (00:01:42) I went to Geneva. (00:01:43) He works out of Europe, sees all his European customers. (00:01:45) And he had a dinner in Geneva, which I joined. (00:01:47) And then me, Jensen, Brad Smith, Anthony Tan from Grab and a bunch of other folks. (00:01:54) were put on this UN commission for AI that Mark is the co-chairman of. (00:01:59) When you see Mark Benioff in Action Man, this guy is a master. (00:02:04) Holy shit. (00:02:04) He is the impresario of impresarios. (00:02:07) Yeah. (00:02:07) You see how he's built such a ginormous business. (00:02:10) It's impressive. (00:02:11) It's impressive. (00:02:12) What is a commission by the United Nations for AI? (00:02:16) What is their calling? (00:02:20) Open source. (00:02:21) No, but no, I mean, it's like the United Nations actually do anything. (00:02:25) They actually do anything. (00:02:27) Actually, it's so funny. (00:02:28) Yeah, I mean, Anthropic was there too. (00:02:30) One of the co-founders, Tom Brown, I think was his name. (00:02:32) Yeah. (00:02:33) That's good. (00:02:34) Was the Anthropic guy running around saying, it's the end of the world. (00:02:36) It's the end of the world. (00:02:37) No, He was very funny. (00:02:40) Tom's awesome. (00:02:41) Tom's awesome. (00:02:42) In fairness to him, he wears it on his sleeve, which is like, hey, we really believe we're doing the right thing. (00:02:50) Right. (00:02:50) And just trust us. (00:02:51) And I think the future is open source for all these countries. (00:02:54) Well, we're going to get into that. (00:02:56) That's on the docket for sure. (00:02:59) But let's start with the IPO update. (00:03:00) You know, there's a trillion dollar IPO rush to the exits. (00:03:05) And you know, this was a big topic of discussion, Brad, at the liquidity summit last month, and we'd never seen a trillion dollar IPO. (00:03:12) We had one this year already, SpaceX trading right about where it went public. (00:03:16) So it was priced, I guess, to perfection. (00:03:19) And theoretically, we're going to see two more. (00:03:22) Brad has the inside information, so I'll try to get it out of him. (00:03:25) OpenAI and Anthropic are slated to go out. (00:03:28) Let's just go quickly over what happened with SpaceX. (00:03:32) It ran up to $200 a share. (00:03:34) It's been down a bit. (00:03:35) It's at $150 a share, as I said. (00:03:36) That's right at the IPO price. (00:03:38) So it's trading at that 2 trillion. (00:03:40) Market cap, currently 7th largest company in the world. (00:03:44) And Anthropic confidentially filed on June 1st. (00:03:48) I don't know why they call this confidential filing when it immediately comes out, but I guess the information is confidential. (00:03:53) Polymarket says 65% chance Anthropic's IPO will happen this year on light volume, 360k. (00:04:00) And 2 weeks ago, Gavin Baker, another bestie, said he thinks they're going to end 2026 with over $100 billion in revenue and very profitable. (00:04:14) He said, a couple of us guessed on the program, that he thinks it would trade at 3 trillion right now if it went public. (00:04:21) Chamath, you made a great call on the pod. (00:04:23) You said, hey, good idea for Elon to get out first. (00:04:26) What are the chances here, Chamath, that these other two get out this year or maybe in, say, nine months in the first quarter of next year? (00:04:34) We'll start there. (00:04:35) Well, I think that these are all great businesses. (00:04:38) I think the question is, what is the market clearing price? (00:04:40) And I think that's more of a function of (00:04:43) how much appetite the markets have to absorb new issues and at what scale. (00:04:50) That's number one. (00:04:52) And I think that's mostly determined by price. (00:04:56) So I think Anthropic and OpenAI are probably in two different places. (00:04:59) The last time we heard from OpenAI, their cash burn was still quite high just because of the diffuse nature of their business and more reliance on consumer than enterprise. (00:05:10) I think Brad mentioned it in one of the pods that (00:05:12) Anthropic may actually be accidentally profitable. (00:05:14) I think he said something like that. (00:05:16) Yeah. (00:05:16) Let me tell you something really interesting. (00:05:18) I sat down with my CTO today and I said, how are we doing on token spend? (00:05:23) And he said the most incredible thing. (00:05:25) He said, right now, our token costs are doubling every 45 days. (00:05:31) Okay. (00:05:32) And I was like, ugh. (00:05:34) And he said, yeah. (00:05:36) And I said, well, what is the downstream productivity? (00:05:40) And he said, maybe 5% max. (00:05:43) Okay. (00:05:43) And I said, okay, so my costs are doubling every 45 days. (00:05:47) My upside is essentially flat. (00:05:51) And he said, basically. (00:05:52) And I said, well, explain why that is. (00:05:55) And he said, honestly, what we're finding out is that you need to use a lot more tokens to get to this next iteration of improvement because we've effectively already asymptoted. (00:06:09) And I said, so what should we do? (00:06:10) And he said, honestly, we have to figure this out. (00:06:12) And so we're going to take a step back and try to figure out what to do. (00:06:16) I don't know how many other companies will actually go through this reckoning now, but the point is everybody in the next three or four years will for sure go through it. (00:06:24) So I suspect that if you can get out now, you should get out now before all of that starts to seep into the water table. (00:06:34) because I think that's probably what allows you to get out at a huge price and raise a huge amount of money. (00:06:39) All right, Brad, you are well invested and well known for being invested in these two next IPOs. (00:06:45) So you probably have some good insights since you've talked to them on a regular basis. (00:06:51) Chances they get out in the next six to nine months. (00:06:55) Both of them, you'd say 100% chance, unless there's some outside event, blockade of Taiwan, some black swan event that we're not anticipating. (00:07:03) What do you think the chances are they're public when we're sitting here and I'm skiing in Hokkaido? (00:07:09) Yeah, I think it's very high. (00:07:11) But let me first say, the SpaceX IPO where we were also investors and we also bought in the IPO, I mean, it was textbook. (00:07:18) It was a hugely successful IPO. (00:07:21) They raised $75 billion at $1.75 trillion. (00:07:26) Okay, so it went out below where we are today. (00:07:28) It's up 25%. (00:07:30) And let's call it on $35 billion of forward revenue. (00:07:33) So if you think about that revenue multiple, it's trading at $2 trillion on roughly $35 billion of forward revenue. (00:07:39) It's an incredible achievement. (00:07:42) I think it was textbook. (00:07:43) I think Anthropic and OpenAI were watching very closely because frankly, we had not had an IPO of that size. (00:07:49) And to Elon's credit and to the team's credit, Brett and Gwen, they really pioneered some really smart and interesting things as part of that IPO. (00:07:58) So you heard from Gavin, Anthropic's rumored to be trending over 100 billion in revenue compared to the 35, right? (00:08:09) If they exit the year at 100, (00:08:12) That means their gap revenue next year could be well over 100. (00:08:16) So based on the SpaceX success, I think it would be a blockbuster IPO. (00:08:21) And I think SpaceX has shown them the way on things like the total raise, pricing, liquidity, inclusion into the indexes, how to do the lockup. (00:08:30) Like I think they've gone to school. (00:08:31) It was a staged release in terms of getting out of the lockup. (00:08:35) It has to hit certain milestones and some of those are time. (00:08:38) Early inclusion in the index. (00:08:41) There are only inclusion in the index. (00:08:45) Let me have you unpack that for a second. (00:08:46) Because people said, hey, maybe this feels unfair that they should be forced to buy it. (00:08:52) What's your take on that? (00:08:53) Is that just like, (00:08:55) haters going to hate or is there something to that? (00:08:58) I think there was legitimate concern, right? (00:09:00) This had never been... (00:09:02) The legitimate concern is that a company that had not been through the process of being vetted post-IPO, there's a lot of volatility. (00:09:09) You've seen that chart, Jason, that the peak to trough drawdown in the six months post-IPO is 50%. (00:09:15) We've seen a pretty big drawdown here from the peak to trough. (00:09:19) as well. (00:09:20) So you don't want to jam it into an index at the peak and then have a 30% drawdown on top of people, which often happens in IPOs because people get excited. (00:09:28) It runs ahead of itself. (00:09:29) But they didn't do that here. (00:09:31) There was fear that was going to happen. (00:09:33) So both the exchanges and the indexes, they looked at this and they made some modifications because the other side of the argument is it's so damn big and important that it needs to be part of the index. (00:09:44) Right. (00:09:45) And so the reason the rules had previously existed is (00:09:49) because most companies coming public were younger, earlier, less tested, less revenues, less profitable, all the things weren't as important in the overall scheme of things. (00:09:58) So I think that they pioneered some really smart things. (00:10:01) It's worked well. (00:10:02) It's traded well. (00:10:04) And so I think that provides a bit of a blueprint for Anthropic. (00:10:07) But just in terms of the enthusiasm, is it Altimeter or is it Fidelity or is it T-Rowe an enthusiastic buyer of Anthropic based upon the things we know today around profitability and model improvement and revenue growth, et cetera? (00:10:22) Yes, everybody would be pig piling in. (00:10:25) Everybody would be trying to get into the top of the book. (00:10:28) And the last I heard, (00:10:31) again, rumored that they would like to get out this year. (00:10:34) On OpenAI, everybody knows that Anthropic kind of passed OpenAI on a revenue trajectory, but I will tell you, OpenAI has kind of got its swagger and mojo back. (00:10:44) It's coming out, you know, just today with a whole new set of models. (00:10:47) We know GPT-6, you know, there's a lot of talk of that coming out within the next 30 days, a whole new generation of models. (00:10:55) I think their revenue has really ticked back up. (00:10:57) The most recent kind of rumors I see on Twitter is around $70 billion to exit the year. (00:11:02) So just as a reminder, 70 billion may not be over 100 billion that's rumored and anthropic, but it's still twice where the revenue of SpaceX is at. (00:11:12) So can they get out at over a trillion on that type of revenue growth being one of the two frontier (00:11:18) Premier Labs, I think the answer to that is yes. (00:11:21) I'm not sure there's a huge race between the two of them to get out first. (00:11:26) I think they'll both go out when it's time. (00:11:28) I think OpenAI has a little bit more complexity just associated with the corporate restructuring that they have to go through, et cetera. (00:11:36) So I would be surprised if they go out before Anthropic, but the fact of the matter is I don't know. (00:11:40) But today, as I sit here today, Altimeter would be a buyer at scale and at size in both of those IPOs. (00:11:48) At $3 trillion, are you a buyer or are you a, hey, it's obviously going to trade up and down and there's no rush. (00:11:55) Because you, I think, were the one who said on the pod or might have been at Liquidity Live, when I asked you point blank, hey, should retail get involved in SpaceX, what's your thoughts? (00:12:03) And you were like, hey, listen, it's a 4% float, 5% float. (00:12:07) It's going to trade up and down. (00:12:08) But a year from now, it might be trading at the same basic price. (00:12:12) It's going to be priced (00:12:14) Not to perfection, which it seems to have been, but I think your position was it's going to be priced reasonably. (00:12:18) There'll be plenty of time to get in. (00:12:19) You don't have to like, panic about getting your shares. (00:12:23) Yeah. (00:12:23) Once a company's valued at over a trillion dollars, like the get rich quick schemes are over, right? (00:12:30) Like you and I share a deep passion, Jason. (00:12:33) We got to get retail investors. (00:12:35) We got to get the citizens of the United States in on these value creating opportunities earlier, right? (00:12:40) The accredited investor laws are insane that we have in this country. (00:12:44) and keeps people from participating in these things, but it is what it is, right? (00:12:48) So they're coming public at over a trillion dollars. (00:12:50) I still think there's a lot of meat on the bone on SpaceX, on Anthropic, on OpenAI, but you're not going to have things that are, I don't expect that they're going to be priced in a way where you're going to get a 50% to 100% durable bounce out of the IPOs. (00:13:04) If so, that would mean they were probably mispriced right into the IPO. (00:13:09) But I do think that these things can be compounders. (00:13:12) They're going to compound at the rate they compound revenue. (00:13:14) And I think all of these companies are going to compound revenue at well over 30% for the next many years. (00:13:20) And 30% a year, just so people understand, this is high growth in public markets on very large revenue numbers already. (00:13:28) Growing 30% when you have 100 million revenue is one thing. (00:13:30) Growing 30% when you've got 10 billion or 100 billion, this becomes a different task. (00:13:37) So let's talk a little bit about these two companies, Chamath, and what the public's going to perceive them as. (00:13:45) ChatGPT seemed to be the public brand, the consumer brand. (00:13:52) for large language models. (00:13:54) It's the AI for people who are doing their homework or mom and dad are trying to fix the dishwasher or whatever. (00:14:00) And then Claude took the lane of, hey, we're going to be the one for corporate. (00:14:04) And it did seem like OpenAI got very distracted with Sora and Disney relationship. (00:14:11) We're going to make a puck with Johnny Ive, everything consumer. (00:14:14) Then they realized, oh, wow, the revenue seems to be an enterprise first. (00:14:19) Is that going to wind up being the big mistake when we look at it? (00:14:22) They kind of gave the Google position, the high growth position to Claude and Anthropic and they took the Yahoo position. (00:14:29) Or do you think they'll catch up on the enterprise? (00:14:32) Or maybe they should just go back to trying to be the consumer version. (00:14:35) How are these going to be positioned a year from now? (00:14:38) How's the public going to look at them? (00:14:40) The problem with enterprise revenue is at some point, the person that's spending it has to see an ROI. (00:14:47) I asked Fable (00:14:49) Five high. (00:14:50) Anthropic's new model. (00:14:52) Anthropic's new model. (00:14:53) I first asked it, what is the lift of the S&P 500 earnings per share growth since 2024 from AI? (00:15:03) And they answered, oh, it's 50%. (00:15:05) So then I looked through it and I said, well, no, you're including the money that Nvidia makes from selling chips to Amazon. (00:15:12) So I said, okay, I asked a different question, which is then what was the EPS growth (00:15:18) of the S&P 493? (00:15:20) And the answer was 9%. (00:15:21) And I said, okay, well, that's different. (00:15:24) And I said, unpack that. (00:15:26) And the overwhelming majority of that was from pricing power, sitting on top of inflation. (00:15:35) And then the other 3% was from buybacks. (00:15:41) And so the answer as far as (00:15:44) all publicly available data was that the actual ROI was somewhere between 0 and 2%. (00:15:51) So I don't know. (00:15:52) I mean, I think that enterprise looks really good. (00:15:55) The problem is that very smart investors like Brad and Gavin and others at some point will start asking companies, what's your ROI? (00:16:03) What's the actual EPS lift? (00:16:05) And if the answer is, well, I don't really know, or I'm not sure. (00:16:11) And (00:16:12) you don't necessarily have the pricing power to continue to raise prices. (00:16:18) Enterprise is probably a little bit more brittle because there are fewer buyers and they're more demanding. (00:16:25) Consumer, on the other hand, then all of a sudden becomes an incredible safe harbor because you have 10s of millions of buyers. (00:16:34) And having those two orders of magnitude more buyers at a much smaller price point inoculates you from the vicissitudes of an ROI discussion. (00:16:44) So it all really depends on what the actual ROI is of this money being spent. (00:16:50) I think that we're in the phase of just being astonished, as Brad said, about the scale of the revenue growth. (00:16:57) But at some point, you'd have to be an idiot not to ask, well, who is paying you this and can they sustain paying it to you? (00:17:05) I just don't know what the answer to that question is. (00:17:09) And at some point, it may not be now, at some point, people will have to answer that question. (00:17:16) And interestingly, you're spending $1,000,000 a year on tokens and that $1,000,000 a year is doubling and tripling and quadrupling. (00:17:23) At some point, you're going to have to show an ROI that's (00:17:26) above the risk-free rate of return. (00:17:28) Otherwise, you're going to have some angry investors on your hands. (00:17:31) And our discussion here for the last couple of weeks on the pod has centered around that. (00:17:36) And the industry has responded on the place where all the CTOs, CEOs, and capital allocators hang out, which is x.com, formerly known as Twitter. (00:17:44) Here's Praveen, the CTO of Uber. (00:17:47) And so when you ask, how are they getting (00:17:51) How are they getting the ROI out of this? (00:17:53) People are now bringing that conversation front and center, and they're explaining it on X. (00:17:58) And he talked, remember, Uber was also the one that ran through all their tokens in the first quarter. (00:18:04) So then on the other side of the business, which is legal, operations, marketing, customer support, HR, and procurement, which he lists here, he says in this, you know, today, 99% of our engineers use AI tools. (00:18:15) Okay, great, right? (00:18:16) It's everybody's, you know, doing vibe coding and has coding assistance. (00:18:20) More than 70% of pull requests are attributed to local or cloud agents. (00:18:23) Our engineers have built 2,500 agentic skills. (00:18:26) So how are we bringing agentic AI beyond engineering? (00:18:29) And what they've decided to do is essentially, he talks about these agentic pods. (00:18:34) And this to me seems directionally how this should be done, which is you find engineers (00:18:38) And you, as we talked about forward, deployed engineers, fancy way of saying, put an engineer, put them into departments and have them work with the department heads who understand systems thinking, how their process is done. (00:18:51) And he says it's making basically, long and short of it is, they're making massive, massive progress on the operational side of the business. (00:19:01) So Brad, you're pretty familiar with Uber. (00:19:04) and have been a long supporter of that. (00:19:05) This is a company that knows how to deploy technology pretty well, and they're an operations machine run by an operations machine, DARA. (00:19:12) They should report the EPS gains attributable to AI. (00:19:15) Yeah, I mean, and this first step seems like they're really being thoughtful about this. (00:19:20) First, hey, this token spent got out of control with the developers. (00:19:23) We're going to need to pause this and look at it. (00:19:25) And then second, here's how it's going to lower costs and create more efficiency. (00:19:28) So Brad, let's talk about that side of it, not just token maxing with the developers hitting the slot machine of like, okay, let's see if this pull requests and let's see if this produces the right code or not to these departments in a more strategic way. (00:19:41) It's not just the person who works in HR. (00:19:44) using quad code or perplexity or whatever, and trying to vibe code something, this is, hey, we're sending engineering in to work with your top systems architect, and we're going to try to find that ROI, yeah? (00:19:57) Yes, first, I would say... (00:20:00) Chamath is right. (00:20:01) The only question is on what timeframe. (00:20:04) There's no doubt that there's a lot of money being spent today that is in the experimental bucket, right? (00:20:10) Where I think there probably isn't direct ROI, Chamath, to your point, but I think we're so early, nobody cares. (00:20:17) I think we're so early in terms of enterprise adoption. (00:20:19) Remember, the total addressable market here is every single small, medium, large company on the planet. (00:20:25) And so we've never seen revenue growth like this because we've never seen a TAM like this. (00:20:31) And if you look at the distribution of revenues across these businesses, it's not like it's concentrated with four or five customers. (00:20:38) There are millions of customers independently, economically making the decision that is rational for them every day that it makes sense, like Praveen at Uber. (00:20:46) And of course, they're trying to find things on both right now, mostly the cost side, cost takeouts to justify the investments that they're making. (00:20:56) in tokens, but I think we're on the verge of breakthroughs in intelligence that's going to dramatically change the revenue side of the equation for a lot of these businesses, breakthroughs in life sciences, breakthroughs in product innovation, et cetera, where they could not divorce themselves from this, even if they wanted to. (00:21:14) For example, Jensen Huang has talked many times that all of his design work, all of his design work now at Nvidia is using AI to design the next generation (00:21:26) chip. (00:21:27) The machine is building the machine. (00:21:29) So you can't get rid of that, even if you wanted to. (00:21:32) And tiny intelligence advantages at the frontier where he sits are required. (00:21:39) Like there's no way, I don't think that Jensen is going to use anything but the best models that he can to build out those capabilities. (00:21:47) So I just think that we're not going to see that in the next few years. (00:21:50) You're going to see it under the hood, of course, but that occurred at Snowflake. (00:21:53) There was tons of optimization that occurred at Snowflake. (00:21:56) But their revenue continued unabated, their revenue growth continued unabated because they further penetrated use cases, further penetrated the enterprise. (00:22:05) So let me be provocative here. (00:22:07) If these guys end the year over 100 billion, I think that they're on a revenue trajectory that they could 3 to 5X again next year. (00:22:16) We've never seen anything like this. (00:22:18) Never. (00:22:18) Okay, you're saying 100 to 300, 100 to 400 billion. (00:22:22) And Jason. (00:22:23) Like you and I have talked about this. (00:22:25) Our minds were blown if a company could go from 100 million to 300 million. (00:22:31) We're talking from 100 billion to 300 billion. (00:22:35) 200 billion of incremental revenue is incomprehensible in the history of Silicon Valley, okay? (00:22:42) And just the fact that we're even in that. (00:22:43) In the history of the world. (00:22:45) Yeah, in the history of the world. (00:22:46) So the fact that we're even talking anywhere close to this tells us something different is going on here. (00:22:52) I think the thing that's (00:22:53) different is that intelligence is the largest TAM we've ever seen in the history of the world. (00:23:00) These guys are penetrating it. (00:23:02) So yes, the super sophisticated companies, the 8090 and Chamath are helping optimize their token spend that are early adopters, 100% that's occurring, but it's not really changing the trajectory that the Frontier Labs are on. (00:23:17) and one of the interesting things about this technology that's really unique, we talk about intelligence on demand. (00:23:23) When you would make a piece of software or you had some technological innovation, it would typically accrue to, I don't know, one group of people in an organization, maybe two groups of people, right? (00:23:33) Excel comes out, okay, yeah, the accounting department's having a field day with it, but it's not really affecting human resources or marketing. (00:23:39) Okay, yeah, maybe it trickles down eventually. (00:23:41) Every single person in every single organization is playing with these tools. (00:23:46) So if everybody's playing with it, everybody's trying to apply it all at the same time, it's kind of like, you got a thousand-person organization, people are spending 200 a month. (00:23:55) Okay, yeah, they double it every, you know, X number of months. (00:23:59) Okay, yeah, now they're spending $400 a month per person. (00:24:02) Okay, they're spending $5,000. (00:24:03) Well, if the average salary is 100, 150K at this organization, it's only an incremental 3, 4, 5% on top of their salary. (00:24:10) So the way I look at it is, did it make that person 3, 4, 5 times more effective at their job? (00:24:15) And I think the answer is yes. (00:24:17) So that's why there's so much token maxing going on. (00:24:20) And it's also a bottom-up type product. (00:24:23) You can just get into this product for 20 bucks a month. (00:24:26) And no CIO or CTO is like, oh no, you can't spend 20 bucks a month on your corporate card for this technology. (00:24:31) So when a bottom-up technology hits everybody at the same time, that's what would explain this revenue ramp that we're all having a hard time adjusting to. (00:24:40) It applies to every single person. (00:24:43) Like who isn't impacted by the technology is my question to you, Tramath. (00:24:46) Like in what organization you're working with 80-90, is there a department that says, yeah, the intelligence on demand, not for us. (00:24:53) We don't need it. (00:24:55) it's less about being dismissive that way. (00:24:57) It's more that regulators and other people won't necessarily allow to use it the way you want. (00:25:02) Okay, so finance, HIPAA, yeah, there's HR data. (00:25:06) You're not allowed to put that to work just yet. (00:25:09) What I'm finding is, once you start using this and getting some gains, it's very addictive. (00:25:15) And we were sitting here, Brad, I don't know, maybe in January, and I got that open claw bug. (00:25:21) And then, I started playing with this Hermes, Hermes agent, which is not a French company, by the way. (00:25:26) They just use French names. (00:25:28) It's Nouveau Research or whatever it is. (00:25:30) I started playing with that. (00:25:31) It's a very peculiar piece of software, but it's a very open piece of software. (00:25:34) So I went to Open Router. (00:25:36) I got my own keys. (00:25:37) I've been playing with GLM. (00:25:39) Then I talked a little bit about BitSensor on the program known as TAO, dollar sign T-A-O. (00:25:44) It's a crypto project. (00:25:45) Somebody who is creating a subnet (00:25:48) that is putting GLM 5.2 and other models available at really cheap prices. (00:25:53) So I all of a sudden experienced, because they gave me an API key, having my token costs go down 95%. (00:26:00) And when you have unlimited tokens as an exercise, which is going to come to everybody, eventually everybody's going to learn how to drop the price by 95%. (00:26:09) And it's going to happen as well because people like Rock with inference, this is all inference, right? (00:26:15) This is what people are using. (00:26:17) They're using inference to do this. (00:26:20) Well, inference is being impacted like three or four different ways. (00:26:23) The software is getting better, open source at the same time. (00:26:27) You're going to have distributed networks like TAO, and you're going to have better chip sets from Grok and Cerebras, et cetera. (00:26:35) All that's happening at the same time. (00:26:37) Once I got down to 95% cheaper, I started setting my agents, instead of doing daily runs, (00:26:43) to doing hourly runs. (00:26:45) Then I took my agents from doing one task and I broke them up into three agents and have them doing three different things on the hour. (00:26:52) And when you start doing hourly tasks and then you wake up in the morning and like 14 jobs have been done, you're like, wait a second, this is completely different. (00:27:01) As one example, I have it has all the all-in episodes, all the This Week in Startups episodes, and we set these cron jobs to go find what the new trends are in technology. (00:27:11) I have a trend spotting agent running every hour informing me of the top three or four trends, and I just give it words. (00:27:19) Really does change your thinking when costs go down. (00:27:21) What do you think the tokens are going to cost, Brad, in next year? (00:27:24) I mean, yeah, we've seen 90% reductions in the price of tokens for each of the last 2 1/2 years. (00:27:32) We've talked a lot about Jevons Paradox. (00:27:35) which I think you're referencing here, which is you're going to use a hell of a lot more when it happens. (00:27:39) I think the central debate right now in AI is the one that Chamath keeps pointing. (00:27:46) pointing us back in the direction of, which is for 18 months since the deepseek moment, right? (00:27:52) When the deepseek moment happened, the markets fell 40%. (00:27:55) And there was a reason for that. (00:27:57) Many started arguing that the frontier models were screwed, that open source was going to kill them, that they were closing the intelligence gap, that model routing was going to make it easier and easier to route these tasks to cheap tokens. (00:28:12) But despite all of those arguments, and now we're 18 months into this, and I had this back and forth with Gurley A lot. (00:28:18) I love open source. (00:28:19) I want all the competition in the world. (00:28:21) Let's be very clear. (00:28:22) But despite all of those arguments, the facts on the field are just the opposite. (00:28:27) The share of economic value, right? (00:28:30) There's this quote, there's this tweet this week from Jesse Zhang that we ought to pull up here. (00:28:34) You know, the economic value, the share of wallet is actually increasing to the Frontier Labs, while the share of tokens, these (00:28:42) commodity tokens is obviously going up to the other guys. (00:28:46) And I had a little back and forth this week with Nikesh on this, kind of trying to suss out why is that the case, right? (00:28:53) Because what people would have thought is, oh, cheaper, pretty damn good, 90% is good enough to do all these tasks that you're talking about, Jason, so nobody's going to use the Anthropics and the OpenAIs of the world. (00:29:05) But despite that, it looks like their share of wallet has gone up. (00:29:08) I think it's not that. (00:29:09) I think it's more that when (00:29:11) The iPhone was a novelty. (00:29:13) Everybody would keep upgrading because you expected that the new price was worth it. (00:29:17) And then at some point, there's a moment, and you can debate when it happened, where people said, you know what, I'm just going to keep the old phone because it's good enough and I just don't see the difference. (00:29:26) And I think that there's going to be a moment like that. (00:29:30) Like when I use Fable 5, (00:29:32) The problem is that it's nerfed on a bunch of things that I would normally research. (00:29:37) I was with somebody this weekend, and he was telling me about some health thing, and I put it into Fable, and it's like, I won't answer you. (00:29:42) I'm like, Okay. (00:29:44) I think that everybody will get to a point, they'll get to it at different times, where they just say, You know what? (00:29:52) It shouldn't really matter what model I'm using if I get an answer that I think is reasonable, and I can go about my day. (00:29:59) Separately, I think when the corporate CFO gets involved, that'll be an entirely different conversation altogether. (00:30:04) I think that what I can tell you after this UN commission that I joined with Benioff and Jensen and Brad Smith, there is not a single country in the world that is not trying to figure out its own sovereign AI strategy. (00:30:16) And I don't think they believe using a closed source American model is the answer. (00:30:21) And so I think we have to keep in mind there's trends. (00:30:25) One is just geographic penetration of humans, and there are still many, many, many more people that don't use it than do, which is an upside and an opportunity for everybody. (00:30:35) And then the second is there is going to be the experimentation, as you said, that needs to transition to ongoing repeatable usage. (00:30:44) And then the third is that all of that then needs to plug into the existing regulatory infrastructure that we use as societies to run the world. (00:30:52) And I think when you put all of these things together, (00:30:54) It's not clear to me who wins, except that you're going to have a lot of diversity of choice. (00:31:00) Certain countries, I can tell you after this week, have no desire to subjugate themselves to any technical risk. (00:31:09) And so they're willing to spend the money to have their own. (00:31:11) Now we can argue and debate whether that country has any chance, but they would rather take an open source model like Nvidia's actually, and stand up their own stack soup to nuts for their own people and their own companies inside of their own country. (00:31:24) And if the models are 99% as good or 95% as good, there's going to be a claim that some countries make, which is it's just good enough. (00:31:32) That's the question. (00:31:33) That's the question. (00:31:34) And then separately, there are companies who will not have the earnings growth to justify this without going on some long protracted carve out of cost. (00:31:44) And most companies, you know this, they just don't do it. (00:31:50) They don't have the (00:31:52) nerve to do it. (00:31:53) They're not capable of it. (00:31:54) You wrote that famous essay to Zuck. (00:31:57) He was pressured into finally doing it. (00:31:58) Absent to very few companies, most people just allow the problems to compound. (00:32:04) So I just don't see a world where when you get clobbered over the head, (00:32:10) You don't look at other ways of just displacing costs. (00:32:13) And if it's like Coke, Pepsi kind of a thing, and Pepsi's 1/1000 the cost of Coke, I don't know. (00:32:18) I just think it's a risk that I think has to be managed in the perception of the market participants and the underwriters. (00:32:26) To add to that, Brad, just open source is very hard to implement when compared to just firing up Claude and having Claude already approved in your organization. (00:32:36) The number of steps it took me (00:32:38) in order to, and I'm pretty familiar with technology, it took me hours to configure my new setup to get onto this bit sensor network to get open router going. (00:32:48) And to your point, Shamath, it does dynamically route now. (00:32:51) So I'm dynamically routing and I'm, you know, GLM 5.2, and then if I fall back to Claude, but which Claude am I going to fall back to? (00:32:58) And here's another piece of evidence to your point, Shamath, (00:33:01) There are some organizations that just aren't capable of this. (00:33:03) They don't have the team that does this naturally. (00:33:06) We just talked about the CTO of Uber. (00:33:08) Now let's talk about another CTO. (00:33:10) Andy Fang is the CTO of DoorDash. (00:33:13) Shout out to Stanley. (00:33:15) And so he, as you can see here in this tweet, a lot of people listening to All In over the last couple of weeks are coming out, as I said, explaining what they're doing to address this exact issue. (00:33:25) He says, hey, with our internal coding benchmarks, we're able to confidently introduce open weight models (00:33:31) to our AI code review without degrading code quality, have the Frontier model, Fable from Anthropic, to do the hardest work, delegate lower level work to Kimi 2.6, and they are now releasing their benchmark. (00:33:43) So another group... (00:33:44) and releasing their benchmarks and saying, hey, we know this is an issue. (00:33:48) The CTO has been charged, to your point, Chamath again, CFO says, hey, make sure this is profitable. (00:33:54) We get the ROI. (00:33:55) They put that on the CTO. (00:33:56) Here's another CTO from another leading tech organization that knows how to implement this. (00:34:01) Yeah. (00:34:01) The really interesting thing we have to forecast right now is what happens in an earnings miss. (00:34:08) And I think what happens in a moment where, for whatever reason, maybe there's just an externality, that there are a series of earnings misses. (00:34:16) Where are people going to look? (00:34:18) And I just think that people find it very difficult to lay off other people. (00:34:23) I think it's much, much easier to cut other costs. (00:34:26) And I think that (00:34:27) The more successful these companies get in a very quick amount of time without really proving the ROI, I just think the bigger the risk is. (00:34:35) It's complicated. (00:34:36) The game on the field when you're working with these enterprises and just trying to explain it to them is, I think that they're getting smarter quickly is what I would say. (00:34:44) I would just say, I think that Chamath Sachs absolutely right about the sovereign stacks that are going to get built around the world. (00:34:51) This is not either or. (00:34:52) We are going to have open source and we are going to have frontier intelligence. (00:34:57) The preponderance of the tokens today are already shifting toward cheaper, lower, lagging models out of OpenAI or lagging models out of Anthropic or the other Frontier Labs that are out there. (00:35:10) Obviously, we talk about those two. (00:35:12) SpaceX has released an incredible model in the last two days. (00:35:17) Meta's out with a terrific model today. (00:35:21) So Gemini is still in the hunt. (00:35:23) So there are lots of choice. (00:35:26) The meta thing was really intense because I thought, okay, we talked about the game theory, which was Mark should scorch the earth with open source. (00:35:35) I think they flubbed that play. (00:35:37) But then I think he is now said he's going to create a price war. (00:35:40) And so if you look at the tweet, (00:35:42) For the quote, there was a post, Jason, I don't know, Nick, if you can find it. (00:35:46) I got it. (00:35:47) Announcing it, he was basically like, hey guys, I'm going to give you the same quality at like 1/100 of the cost. (00:35:52) Now, again, there's a lot between here and there. (00:35:56) There's a lot of enterprise distribution that's required. (00:35:58) And you know, there's been a couple of misfires before, but I thought it was interesting that the vector of challenge was on cost. (00:36:06) Yeah, here's the Mark Zuckerberg tweet, just so I queued up for you there, Brad. (00:36:09) And he's at FinkD, that was his old handle back when he was in college, F-I-N-K-D. (00:36:14) And he's done more tweets today over this MuseSpark announcement than he's done in his history. (00:36:20) So he's getting into the X.com conversation. (00:36:23) Quote, today we're releasing MuseSpark 1.1, a strong agentic encoding model at a very low price. (00:36:29) It's available through our new Meta model API and in Meta AI. (00:36:33) So he's coming out saying, hey, we got the strongest agentic tool here. (00:36:36) Please come use it. (00:36:37) He also wants to have (00:36:39) his own, essentially, he wants to jump into not the hosting space, but he wants to provide tokens as well. (00:36:48) So again, I think we're going to have a tremendous amount of selection. (00:36:51) The competition is great for America. (00:36:54) But I think if you look at the things people are doing, let me give you an example, the premium workload. (00:37:00) Jason, you talked about summarizing a document. (00:37:02) It may take 20,000 cheap tokens to do. (00:37:05) Of course, shoot that to a lagging model or an open source model. (00:37:08) But if you're talking about replacing a software engineer for two hours, that may take 2 million expensive tokens. (00:37:14) And the consequence of using something that's 95% as good is really high. (00:37:21) right? (00:37:21) Because you have a long-running task. (00:37:23) And if the task breaks early or it breaks in the middle or breaks at the end, there's a huge cost to that. (00:37:29) You still burn the tokens, right? (00:37:31) And back to this analogy as you said, they're pulling on a slot machine and you lose. (00:37:36) And the time and the compute. (00:37:37) So if an AI agent is replacing a $200 an hour consultant, right? (00:37:43) Take that as an example. (00:37:44) So 3 consulting firms, they're competing. (00:37:47) They need the smartest consultant. (00:37:49) They're charging 200 bucks an hour. (00:37:51) The difference between spending 3 bucks on a cheap model or 15 bucks on an expensive model to replace a $200 an hour consultant, it's just irrelevance. (00:38:02) That inference cost difference is irrelevance if you're getting some (00:38:05) something that's bulletproof for 15 bucks. (00:38:08) And so I think that's what we're seeing play out. (00:38:10) The best evidence for all of this is just revenue growth, right? (00:38:13) We can sit here and speculate all day long as to revenue growth not from Anthropic and OpenAI, but from their customers. (00:38:19) I'm talking about what is Anthropic's revenue growth compared to OpenAI, compared to the open source models. (00:38:25) Millions of independent actors are choosing every single day. (00:38:29) The open source companies are growing, right? (00:38:32) But they're growing, selling something that is really, really cheap. (00:38:36) And there's room in every single market for premium products, for mid-tier products, and for commodity products. (00:38:43) And I think we see a lot of this token growth. (00:38:46) People are speculating that the intelligence gap between that commodity stuff and the frontier stuff is going to collapse to the point that people won't pay for the frontier stuff. (00:38:57) There is no evidence of that on the field today. (00:38:59) It may develop over the course of the next couple of years, but it's not on the field today. (00:39:04) and just to give people an idea, we keep mentioning what sovereigns are doing. (00:39:08) To give you the specifics on that, the UAE very famously has their own Abu Dhabi Technology Innovation Institute shipping Falcon. (00:39:16) You probably have heard about that. (00:39:18) The Saudis have Humane. (00:39:20) and they're doing their own models that are Arabic LLMs. (00:39:23) And then this week, Japan is investing $6 billion in a consortium. (00:39:28) It's called the Neoterra, the Neoterra, N-E-O-T-R-A consortium. (00:39:33) And they're doing that and skipping ahead to physical AI, i.e. (00:39:37) robotics. (00:39:38) Okay, joining the conversation here, the one, the only, Saxy Poo. (00:39:43) Sax, bringing you into the discussion, talking a little bit here about the debate that we started here on the podcast. (00:39:50) getting ROI from tokens. (00:39:52) Where are the tokens going to accrue to open source versus the frontier models? (00:39:57) Bunch of CTOs chiming in on X this week.