Frontier Labs Threatened by Kimi? Should the US Ban Chinese Open-Source Models & Stripe Buys PayPal

At some point the people spending a trillion dollars a year are going to want some apps to pay for all this. If you're growing 10x year on year and you have any kind of positive and improving gross margin, it just covers all the nut. Growth for the last two or three years has been a very attractive place to make money.

I mean, an eval is just an eval. So let's not take a bunch of folks on X who had someone in their engineering department look at some evals and write a tweet for them. Okay? Like, we're not saying something is similar in performance, maybe, but let's prove it in the field. Having said that, I mean, we can't even sign up new consumers for Kimi because it's blocked. They have so much demand since this happened, right? Demand is literally — I don't know whether it's geometric or exponential — but it's so high we can't even... We can come back to this next week when it opens up and I can use it on the consumer side even better. But I mean, I think there's a lot going on, and there's a lot on politics, and it's an aha moment and a wakeup moment. On the other hand, it's not new. It's not new, right? I mean, if you look at OpenRouter data, half the traffic's through China-created models. Even 'China models' is a confusing term, right? They may well be hosted in the US, right? And when they have open weights, they may well be, for all intents and purposes, truly open-source models hosted in the US. But it's not new. It's just going to accelerate this. And that's why you see the stress. It's just accelerating. And that 50% — instead of being niche or for tech-forward folks or venture-backed folks — you know, in a year it could be everybody, and that is material, just being in the zone and even materially cheaper. It's just going to get more and more attention.

I totally agree with that, actually, Jason. I was curious to see what you say, is that there wasn't... you, Carrie, you kind of led with the 'what do these new models mean?' I think Jason's cut is exactly right. It's exactly what you'd expect. It turns out there's five main Chinese LLM companies and a bunch of followers. It turns out that aggressively funded companies with smart engineers are just going to keep cranking through and building new models. And yeah, they're not state-of-the-art compared to the frontier models, but they're, you know, six months behind depending on how you measure it. So yeah, actually no new news about that. But Jason's right, quite a lot of fun news about how parts of the US responded to that. We had the small-p political response. So that's one dimension. You know, the policy adviser for OpenAI, formerly from the Trump administration, making some comments on Twitter leading to a wonderful firestorm that we'll absolutely talk about. That's one thread. And then another thread is just talking about what these models start to reveal about the economics of a model company. I mean, Jason hinted at it. We lump all these models in together, right? But let's take two: some of the DeepSeek models you can run on your PC or your Mac or on a laptop, right? Conversely, Kimi K3 is, I think, a 2.8 trillion parameter model. It's a huge honking thing, and you know, you need myriads of GPUs just to run it. So they're not, quote-unquote, the same thing. That's much more comparable in size, and therefore in terms of compute capacity, to US frontier models. So there's a lot we can learn about. I think we'll kind of talk about the politics first, and then maybe oddly enough talk about the inference implications and kind of as that goes into the opportunity for fireworks. So, lots of downstream implications, but zooming out, nothing amazingly surprising in the news that after three years of competent execution along a pretty defined trend, we now have three years and three months of competent execution around a pretty defined trend.

It's Dean Ball, and he is currently, I think it's out of policy or communications director for OpenAI. He just started there two weeks ago. Before that he was at the Trump part of the administration, kind of on AI policy, and before that a bunch of Hoover Institute type stuff. And he, you know, he set off a firestorm with the tweet, and then he did a little bit of 'Oh, I can't really post because I'm now at OpenAI, everyone was mean to me because I posted a bunch of stuff,' which was frankly a little naive comment. Because the two comments about the tweet: one is you're in a senior role at OpenAI; one, there was a hysterical tone to it, right? He used the word 'AI communism' and it was very kind of overexaggerated. And then secondly, you know, when you start even hinting about — I mean, we saw this when Sarah Frier did it — when you start hinting about significant regulatory changes that will massively benefit you, you got to expect that everyone's going to say, 'Dude, of course you're going to say that. You're, you know, that's your side.' And if you make the expensive closed-source product that sells for, you know, 10, 20 bucks, and the Chinese are shipping something for two bucks, and you say, 'Well, totally independently, just speaking as a common citizen, I think they should ban this shit,' you got to expect that a whole bunch of people are going to say, 'Dude, you're not talking as a common citizen. You're talking as the provider of the company who will jack up our rates the minute the stuff gets banned.' So, you know, it was a little naive not to expect that level of blowback.

Well, we'll see at some level. First of all, I think that guy at OpenAI had been there like two weeks, right? Two weeks. Whether this was — whether he used that as a reason to go on this or — as they say in the meme, Jason, two weeks so far. Yeah. So far.

Listen, I can't — I'm not a total expert. It's difficult for me to imagine the federal government's ever going to use a China-built model at this point in the US, right? It's difficult, and anything adjacent to that, it's difficult to imagine. Just simply — I mean, there's always in our whole history at tech, the ability of Chinese technology to penetrate many US buyers has been limited, right? It has certainly been limited in telecom and other spaces. So I think the quest — stepping back for a minute — the real question is how limited is it going to be, right? How limited are we going to... because it's going to be limited. The availability of China-built models to penetrate the US is going to be limited. The question is just how much. You know, Jesse Zhang had a Twitter article today or yesterday. I think today — that was good — and it was pretty good. I think people might have missed it because it's real data, which is what I like. But he said, 'Here's one of our most regulated companies. We have highly regulated folks. Just our token use here has gone up what looks to be about 2.5x since January.' Yeah. Okay. And the reasons are really interesting. I mean, I've lived this myself. The reasons are having supervisor models track the agents so the agents don't make mistakes, running multiple agents in parallel so they don't make mistakes. The more regulated you are, the less forgiving you are of an error in an agent. And so it's like four times the agentic use just to have multiple agents regulating agents. If it's already grown that much in the first half of the year, you know, the quest for equivalent models at a cheaper price is just going to keep going up. It's just going to keep going up. And so, but we've always had cheaper, pretty good solutions from other vendors. It's not new.

I didn't realize Bill had said that, though. There's something very pleasing about that, which I'll mention in just a second, right? Because one of the fun things about this policy dispute is it brings out the hater in everybody, right? And you know, Dean Ball said what he said, and then two people who can be controversial came down strongly on the other side, and I support them both. The first was David Sacks, the former AI czar, who basically said this is rubbish, stop. And then the second one was Emil, whom you mentioned before, Emil Michael — I'm never sure the pronunciation of his last name — who was a guy at the Defense Department who got totally sideways with Anthropic. I mean, what I like about that guy is that man knows how to hate, and one of his biggest hates for the last decade and a half has, of course, been Bill Gurley from his time at Uber. So I really find it... So if Bill and Emil are on the same side saying don't ban these models, then you got to know that there's got to be some truth in that. It's got to make you think, because that's an interesting lineup. But yeah, I think there — and I actually just saw literally as I came on, and look, this is the Trump administration, so things change every day. But a Politico piece today basically saying some version of 'we're not going to ban these things on any significant basis,' which as Jason points out is very different than saying the White House decision support system will be run on Kimi, right? Even if it's Kimi hosted in California, I think we can take it for granted it won't be. Right? But look, conversely, if you're a decacorn and you're a startup doing inference on customer support queries for a very boring consumer product, there is no reason why you should pay marquee prices when something 10x cheaper is available, and it would be horribly bad policy to ban that. Right.

Well, look, the one thing I will say, just to add — and Bill Gurley, another rich, grouchy billionaire — grouchy Bill Gurley's probably got 30 IQ points on me, okay? And he's seen it all, right? And even at his grouchiest point, I learn something from it, right? I always learn from it. So, having said that, I don't think you're going to convince me there aren't some data export risks with China-based models. You're just not going to convince me based on what I've done with all our agents in building. And if you're not going to convince me, I don't think you're going to convince 99% of the world that there isn't some security leakage issue. It's already scary how much of our data we put into these closed-source models in the US. It is scary. Here's Elon saying 'scam Altman' every day to create distrust, right? We cannot understand what these models do. They are connected to the internet. We cannot even — even if we have Fable read it and have it read it itself. I don't think you're going to convince most of us there isn't data export risk. And so I think that's going to lead to tighter constriction than this 'you know, leave everything open so we can compete and my portfolio companies benefit' argument. I think every CIO is being told right now, 'Oh, don't worry, if you host it on-prem, you remove any security risks and the back door then is removed that could potentially be there.' Why would you not be alleviated by that reassurance that on-prem would solve that solution? Why would you not be reassured? Rory's more of a historian here than me. You can mock our regulatory bodies, but they're here to answer those questions for us. Is it safe to drink that cup of coffee? The American Heart Association, I think, just said six cups are safe now, right? This week. Now I know. Now I'm cool, right? I was a little worried about my caffeine consumption when we did... No, no, seriously. I mean, I'm not sure they're right. Who has said my data is not being exported through the most complicated, borderline self-aware software of our lifetimes? Who can say that? Especially — and I admit this can be triggering — there is a history of data export risk with Chinese products. These are companies that are arguably run by the PLA. I'm not — I'm just saying my lifetime of experience says I'm not confident there isn't. And just the internet telling a CIO, I don't think is good enough. And if I were a CIO, it wouldn't be good enough to me, unless — as long as if I thought my job was on the line. I don't want to take this risk, CIO of some Fortune 500, Global 2000 company, unless everyone... I don't know, man.

I don't think it's triggering to say that there are IP risks in this. I mean, at the risk of being kind of level-headed here, the data is very clear that technically important US companies, Boeing for example, suffer continual cyber attacks, many of which are attributable to sovereign state actors, including China. It's a thing. So we're not being sensationalist or alarmist. It would be naive not to put it on the table. You're right, both comments. Second, companies — I'm thinking about... The problem proving a negative is, you know, if you have an... remember these are open — I occasionally say open-source incorrectly — they are open-weight, which means you can see the weights and you can run it yourself, but you don't have technically the full definition of open-source in the context of an LLM, which means seeing the underlying training data, which you don't, right? But you have the open weights. The question is, if the model is being run in a trusted US inference company, based — Fireworks, some of those guys, right? And you could get into a long technical question is, look, what can it really do? Could it initiate tool use on — I mean, it probably speaks — could it initiate tool use on the customer side whereby the model sends a command back to the customer to exfiltrate their data? You can imagine being able to use these models fairly comfortably and being fairly certain that you have blocked access and this can't happen. So I mean, I think you could, in logic, you could satisfy a technologist that the risk is not there. Whether you can satisfy a politician, whether you can satisfy someone who's just afraid of what they don't know, is, Jason, to your point, another question, right? I mean, and I think you're right, you have seen things like Huawei has effectively been prevented from selling to any cellular networks in Europe and the US because of this. You as yet — yeah, unprovable fear. So it's not crazy that there will be some level of, I think on the government side, some restrictions. I think a blanket ban would be massive overkill, to be very clear, right? But I think the interesting question it raises is this. Well, two questions. First of all, it's also worth pointing out that while we're talking about banning Chinese open-weight models, the Chinese administration are talking about preventing those companies from selling those models to the US. Like, just like we don't let them buy Nvidia, they're not going to let us buy their open-source models. So which is kind of totally zany. We think they're trying to sell it to us and we don't want to buy it, and they think they shouldn't be selling it to us because it's so powerful. So that's kind of just weird in of itself. But I think the really interesting question here, and it gets to Thinking Machines, is: is the open-source LLM — open-weight, low-cost LLM business a good business? And if it's a good business, why can't some red-blooded American company step up and give OpenAI and Anthropic a run for the money? And Jason, it's the point you made. If this is — I mean, where's Grok? Where's Gemini? Thinking Machines had an announcement last week. They announced a model. They didn't position it as, you know, state-of-the-art frontier, but they kind of — I think they made a comment on something that you can build upon. Inkling, I think it was called. So one of the — and you know, at one point Meta looked like they were going to go down this route, right? Is there a business? How can you make money? It's an interesting question. Can you make money as a maybe not completely open-weight but a low-cost US provider of these models and be competitive with those guys? Because you know, the open-weight models in China are getting 50, 70 billion valuations. Like, it's not Anthropic, but I wouldn't turn down a $50 billion outcome if someone could make a convincing case to me that a US company could do this. So I think that's one of the interesting questions here.

I've asked so many people why we don't have leading open models in the US. No one's actually given me an answer. We're still waiting for models from Reflection, which I think is kind of one of the hopes that we have. I completely — I was offered Kimi today, by the way, Rory, at 20 billion. This was — it fell into my inbox. 'I have an SPV for you. We do Kimi at 20 billion. We're oversubscribed, but we'll make room for 5 million for Harry.' Yeah, that's because we say such nice things about them. Thank you for your check. But no, look, we're going to gloss by it and I don't have an answer, but it's a huge freaking question, right? There's this new category called LLM intelligence. Two companies existing as premium products. Their combined market cap is $2 trillion. The combined revenue at this point is probably 100 billion plus or minus. There are four or five other companies in the US that are capable and have proven their ability to build something roughly comparable. None of them are taking advantage of this. And there's five Chinese companies that have proven their ability to build something roughly comparable and they're, you know, cranking night and day to take advantage of it. Where are you, Google? Where are you, Reflection, as you say? Where are you, Thinking Machines? Where are you, Llama? I mean, the fact that there's four or five potential competitors — it's just fascinating. Or are you asking them to dance? I'm asking them to ship, you know, a story, and I'm kind of throwing it in here as a wedge.

Because the market's in flux. Everyone's figured out they need this. OpenRouter, like a lot of folks, was early and benefited from it and deserves it. Right. This is a repeat founding team that saw that there would be value to having a fairly heterogeneous mix of models that when we started this pod probably made no sense at some level. It probably seemed too nerdy and too niche and too cool-cat developer. Who's going to need like — yeah, sure, there's a little, it's cool, but you know, guys like Rory and me, we're going to stick to the big guns, right? And everything broke well for them, but it's still a niche product that more and more people are going to build variants of themselves. And is this the plumbing they will pick? Will you know, if you're on a lot of platforms, if you're on adjacent platforms, if you're using Databricks gateway, they'll have their own harness, they'll figure this out for you. There will — I don't know whether Ramp's competitor even makes sense, like I — but my point is it's something that's going to become embedded in so many vendors that if I could sell for a lofty multiple of my last round, I might check out at five or six billion. Like it just — because the next — you've achieved a certain amount of victory in a market that's going through radical change and becoming part of everything. I might take the offer.

Jason is answering the question, why is it a good time to sell? And you know, frankly, as you yourself, Jason, have said, and I say it too, the private market liquidity window opens so rarely that it's always a good idea to pay attention when it does. So I think it's easy to understand that side of it. I actually think the interesting side of the discussion is the other side. Why would someone want to buy, right? And I think when I saw that article, I was like, yeah, that makes sense. If you think about the last conversation we had, you know, what's in the zeitgeist right now, it's this whole idea of can I get escape model dependency, manage my costs, have a whole load of alternatives easily available to me, right? If I'm someone who makes my money as a hyperscaler hosting, especially someone that doesn't have their own just one in-house model they're pushing like Gemini. If I'm maybe Amazon in particular who's made a business of saying I'm going to support all the models. If you had maybe even Microsoft now that the divorce is coming true from OpenAI, you know, maybe this would be a great product to own if I was a cloud hyperscaler. So I will admit I had that moment of — because you often see, and it's step out, you often see this in this kind of market. You have these businesses where intellectually over a 10-year period, you can say, 'Hey, margins are going to be tough in that business. It's going to be compressed and maybe it won't at scale in the end be an amazingly valuable business because it won't be able to extract margin.' But on the other hand, when growth is so quick and people's urgency to adopt technology is so fast as it is right now, if you have a crucial piece of the plumbing at just the right time, when two or three people need that piece of plumbing, you can find yourself in a very interesting position in terms of M&A, right? Because it may well be that — and this is where the finance guys miss it — it may well be that the NPV of the company on a standalone basis is, you know, a couple of billion, not huge, but the value to it right now to a hyperscaler if they could shift 10% market share in the enterprise to them over the next half a decade by saying, 'Dude, we are the cloud, we are the model-agnostic people, and we'll make it easy,' could be interesting. So I remember thinking, I wish I was in that — I mean, which is always how you know what a venture guy really thinks. It's like, damn, I bet you they could at a good offer right now. It's just interesting. So yeah, I'm with you. I think it's an interesting time to sell, interesting time to buy. I mean, Rory, if you were on that board, would you sell at five to six billion?

You know, it's always hard to — I mean, my first comment is whenever you get an offer, I always do the same thing. I say to the founders: one, the window's open. It doesn't open often. We should take it seriously. Two, I'm going to support you whatever you want to do. Three, if there's concerns that you have that you've been sitting on and not telling me, now would be a good time to share so we can make an informed decision, right? And then go away and think about it, and I have a whole process of how you talk to them about it, right? I don't think you should pressure people into selling. I think your job is to give them whatever experience you have to bring to bear, and then they'll make the decision. Because you know, in the end — and it's so funny because founders agonize about this when they think about control and they think, 'Oh my god, these people are going to make a sale' — even if as the VCs we have board control, 70% ownership, and a drag-along for the founders — Jason knows this — if the founders who are core to the business don't want to sell, it's not going to happen. So the first thing I tell the founders is it's your decision, which I think is very empowering because it takes it away from 'people are going to make you' — you know, that's the beauty of being private, unlike being public where you don't have that degrees of freedom. We could come back to that later. So yeah, I would say to them, take this seriously. Do you think you can be worth 3x this in three, four years? Do you think that's worth it? But yeah, I would definitely say take a day out of your life and think about this long and hard.

The way Harry phrased his question was very VC-centric. 'Would you sell at six billion?' — what, I can't do the accent — 'Would you sell at 6 billion?' Okay, I — that is — I hate the term triggering, that triggered me. That triggered me because this is a very VC-centric way to think of it. I've got an asset. What am I sitting at in my last round? 1.8 billion. Okay, there's going to be dilution. There's time value of money. There's IRR impact. Is the six billion worth it for me? I think for a founder, when you start to get into nosebleed offers in absolute terms, it has to be 10x to go for it. It's not worth it for 3x. It is not. Okay. Let's say I own 15% of OpenRouter. Okay. For the money. Okay. It's not — how much am I going to take home? Okay. At 4 billion, I'm going to take home 600 million. Okay. I got — now, one of the founders I think is super rich, right? But put that aside. I got 4 million in the bank. 400,000. 40,000. I'm going to walk away with 600, 700 million. 3x for you — it doesn't — like, maybe later in life it's now 10x. And building something — and this is another trite VC — is building something truly generational. You kind of know as a founder when you're on that path. Okay. And so the VC — 'what, should I do it for 2.8x on my last round?' You know, it's just — it's the right way as a finance guy to think about it. Don't get me wrong, but it's a terrible way for a founder to think about it because there's way too much risk for not enough money. Like going from — again, going from 40,000 in the bank to 400 million versus 800 million — it's irrelevant if there's risk. This is — and sometimes there is a handful of work. There is a handful of work in those next three years. It is a handful of sweat and a handful of market change and a handful of people that quit and move on and a handful of competitors that they're looking pretty good and they may pass you. 3x not good enough, man. Got to be 10x.

You know, when you face that sell decision, you don't not sell because you think you can make twice as much in a year, right? Because you just never know. And it like, you know, and so in the end, even though I didn't think I'd agree with him, in the end I did. It's like what I think he's saying is let's leave aside the 'what do you want to do with your life' questions from a return perspective. You know, don't think incrementally. If you have a chance to sell a company for six billion and make 600 million, and you think you can run it another three years and get, you know, 1.2 billion — that might, risk-adjusted, that mightn't be worth it if your current net worth is $40,000. And that's actually good financial advice. I mean, it doesn't — in other words, if you turn down a big-ass offer, you better be sure it can be way bigger. You better have high certainty and high bigness. I think that's a fair comment.

Yeah. If it's 10x — this is my net advice — if you know in your heart and soul you are building a company 10x bigger than this, right or wrong, like I don't know, then go then effing say no and go for it. Here's a few more shares. In fact, friends, let me reload you. But they only best at 10x. It wasn't what I thought where I thought you were going, because I actually thought you were going to say, yeah, something that you also — yeah. Don't sell also just because financially you feel you should. I mean, because the other thing you're not taking into account is that depends on the person. Some people just love running that company and frankly don't want to sell. It's their life's work. And I also think you have to respect — I mean, my point, you also have to respect that. I mean, again, this gets back to the — there's no one answer. The founder decides. I've known people who are like, 'This is my first hit. I'm going to take it,' who read exactly that logic. I am not —

I'm going to make 60 million bucks. I don't have one million buck and maybe I can make 120 in four years, but I'm taking the 60. And I've also known other people who, to a rounding error, have said this is what I want to do for the rest of my life. Why would I take that money? I've got, I've made three or four million in the secondary. I got my house. I'm done.

Sometimes you're, I mean, as to your job by this team, but sometimes if you're early, you can gain a lot of traction in something that becomes somewhat commoditized. It's just, it's just the way it goes. And if you, and the perfect outcome is to sell the moment it becomes commoditized, but before everyone fully realizes it. That's when they'll give you the money, but that's before the value decrease rather than increase. And my gut is it might be, it might be now. That crossover moment. And I think Rory kind of made a version of that point. It might be now. It spaces commodify. It doesn't kill everybody, but it might maim you, but it also makes it attractive. It also makes you attractive to acquirers for a window, and then that window closes. The commodification window closes. And probably why Cursor wasn't dumb to sell at 60 billion. I think we can agree that's true. I don't love the commoditization description. I think it's an overloaded term, but yes, I...

That's exactly it. Where... Yeah. Which segues to the next topic, which is Fireworks. Yeah. And inference in general. Yeah. I mean, take it away, Rory. I'm going to butcher whatever context that you want to take it on.

No, no, you do. No, you do first, because I'm just... Are you sure? I'll lay the framework and then you can just destroy it. Steamroll away.

Yes. First of all, I agree. Yeah, inference is a huge... It goes back, ironically, to the prior comment on open-weight models. This kind of standalone inference is a big business, right? And you know, obviously, you know, inference is both something that's done within the frontier model companies where they do their own inference, and people like Microsoft and Google provide the capex and provide the compute for that. But people like Fireworks and Base10 and Together, they all make their money offering a variety of these open-weight models to third-party developers and enterprises that want to use open-source models to do AI, right? And it, as I said, the two trends go together. They're exploding because the open-source trend is exploding. So if you're Base10, if you're Together, if you're Fireworks, if you're Together, this is your moment, right? So yeah, this is because this is how you access those. Because I can tell you one thing. We can, going back to the discussion about open-weight models from China. It's one thing to decide to use an open-weight model on Fireworks in the US. What you're not going to do is be using the API back to China, even if they'd let you. Right? So this is a one-to-one linkage between the open-source, the open-weight trend. Right? These are the companies that are benefiting massively from that trend, and it's not the only kind of route for inference. There are, you know, inference for US-based models, etc., etc. But the vast bulk of it is, oh my god, I'm sourcing Qwen, I'm sourcing Kimi, I'm hosting Qwen, Kimi, I want to use it as Cursor, I want to get someone to provide me some inference. These guys exist, and you know, they have lots of customer skew at the high end. I believe, you know, companies like Cursor are probably big customers of all these guys. At least they were until they were acquired by, um, acquired, and probably still are, right? So yeah, I mean, it's a great... candidly, I think, you know, it actually goes back to the point I made earlier. Some businesses where, and I think, you know, you can look at and say, oh my gosh, the cost... you know, you have margin compression in your future because you're buying your compute from the neoclouds, and you know, you're offering this product, and are you going to be scrunched? And margins, you know, were probably slow for a while, but now the beauty of it is demand is massive. So whatever compute you have today, whatever compute you've already signed up for, and these guys sign up for compute from the neoclouds in general, are starting to build their own. Whatever compute you own now, you can charge way more, which means what looked like a lowish gross margin business has now probably become a very attractive business. So, not only are they probably going to grow 5x to a billion, but they're probably growing 5x to a billion with expanding gross margins.

Yeah. And just to be clear for people, what that means is... Yeah. And that's exactly where I thought they'd be, and good on them, right? 30%. In other words, what they're saying, and this is going to be an... and I agree with that sentence. It also means that the challenges I hinted at are there in the future, right? Because what they're saying is, if I'm buying data center compute and then effectively selling data center compute with hosted LLM, at some point I'm going to want to own my own data center assets to have more control of my destiny, which means vertically integrating downwards, which also means a ton more capex. So these are going to become way more capex-intensive businesses. There is a risk of commodification here, even though, even with massive complexity and massive capex.

And just to dimension that, because I agree, Jason. I mean, look, I'm an app investor. It hasn't... you know, you look back and you... what's going on here? You've got companies like, you know, Fireworks doing a billion dollars. There's very few apps companies doing that. And, you know, zooming out a million miles, my mental model is I divide the AI world up into three buckets. It's the making AI, the infrastructure layer, right? And you're right, the spend there is $800, $900 billion a year. Then there's the two foundation model companies themselves, and they're doing plus or minus $100 billion a year, right? And then taking those guys out, rounding up every other apps company, right? You struggle to make 40 or 50 bill. You struggle. You start with Cursor at four because I think coding is an app. You know, by the time you're chucking in Harvey, you're adding two, 300 million, right? It's amazing. I mean, just the difference in spend. And, you know, at some point, the people spending a trillion dollars a year are going to want some apps to pay for all this, right? But right now the volume has been front-end loaded on the infrastructure side, and at some point the revenue has to match it. But right now, infra has been the place to be. Like, there's probably more money being spent on training data for the foundation models, you know, the Mercor, Surge, and that, than pretty much any app company outside of Cursor. In fact, probably the sum of all the apps companies outside of Cursor, right? You know, are probably less than the amount that Anthropic and OpenAI are spending on training data, which is just amazing.

That I can guarantee when you look at Mercor, 2 billion in ARR, two billion for Mercor, Surge, another billion, you get to four or five billion, and you know... Surge is three. Yeah. Handshake's one. I mean... I mean, yeah, maybe if you start throwing in on the other side the consumer products like Higgsfield, you get to roughly the same place, but it's astonishing the scale of the investment versus the scale of the apps at this point means that all the action's on the infra side for now.

I've been building this agentic recruiting app just to recruit from the SaaStr community. It's been fun. I've learned a lot building it, right? Hopefully, it can ship in the next week or two. But to really get it great, it needed labeling to make its... Now, I'm going to put model in quotes, right? It uses Sonnet and Opus, but so there's different definitions of model, and it was good, but man, once I started labeling all of this, it got exponentially better, right? Built my own, built our own little labeling tool. And so you need your own micro model, whether it is some sort of reasoning layer that you build on top of Claude or ChatGPT or Kimi or Gemini, like it's still your own model, even if it's not technically a model, right? Because you have your own reasoning layer with your own rules, your own weights, your own... But you want more. If you have the resources, you want to go further than that, right? You want your big M model as soon as you're at a certain amount of scale, and it's not cheap all in, right? You are going to want to have your own model, right? Like a Harvey, Cursor. So some version of this, I think, and the folks that are going to want to... at any application level, folks that are going to want to use just the generic models is just going to decline to prototypes, right? Prototypes and proofing.

Yeah. Or absolute state-of-the-art kind of small parts of the overall task. But agreed. Parts. Yeah. Little parts, right? Like, I mean, again, you're going to want to use the expensive tool for the expensive parts, right? And you're going to want to use the cheap tool for most of the parts, from the customized tool to your usage.

But man, the outputs are just literally an order of magnitude better once you do it. So everyone wants your own model. And so I do think whether that always benefits Fireworks or not, it doesn't matter. As long as they pick up some of the bigger end that scales, it is... The generic models are great, but it is amazing how much better you can do than them for any specific workflow. You can do epically better.

Personally, I totally get it. Like having a subject matter expert go in and answer 20 questions about a disease, about a history. I mean, it's a lot of professors and teachers that they have there, right? That model, right? The amount of power you can get in a domain by having a subject matter expert answer just 20 or 30 questions, right? Five minutes, 10 minutes. The amount of power you can add versus the generic LLMs, which are a sea of mediocrity combined into one giant LLM. Okay, every mediocre history professor, every mediocre doctor that doesn't even know what caused your runny nose is in the LLM. But if you get the best people training on the best answers, it's a step function. I'm less smart on this seeming low end of the model, right? This commodity thing that people made fun of Mercor, but I ain't making fun of it anymore. I tell you that much. And these models are a sea of mediocre all combined in a giant soup that gets better. These domain experts are so powerful in tuning your model to get the better output. So powerful.

I think the answer, though, is really a derivative of the big question, which is, you know, because your statement, your companies are going to want their own model, is probably true, right? And the real question is not that. The real question is, will that be additive to the rough trajectory of the foundation models as it's established today? In other words, coming, you know, at or close to 100 billion combined revenue, growing nicely, or does it start to take away significantly? Because, you know, to answer your specific question, if the foundation models continue to grow, and we just saw an article in The Information that, you know, for all the training data companies, the vast majority of their revenue comes from the foundation models, to which your correct response is, no shit, of course it does, right? If that continues to grow and you have an additive market in enterprise of all these companies, you know, JP Morgan building the JP Morgan model on top, then, you know, it's net expansive, and net expansive is by definition good and reduces customer concentration. And I think that's what people like are forecasting, right? If, on the other hand, you know, which is hard to contemplate today, if these enterprise models, if these open-weight models really impacted the growth rate of Anthropic and OpenAI, then obviously when your 80% customer slows down, it would have a significant impact on your growth rate, right? But, you know, if it's any consolation, Harry, if that happens, worrying about your Mercor valuation will be the least thing people are worried about, because you'll see an implosion of much bigger market cap entities, right? And that, frankly, is the billion-dollar question. You know, can these two foundation models maintain their growth trajectory, which is starting to become profitable, at least in the case of Anthropic, in the face of all this open-weight competition, in the face of this pushback on costs and basically kind of push for ROI? If they can maintain this trajectory for even another one or two years, then everything's fine and everyone's fine. And right now the data says they are. If you start to see slowdown on those two ARR growth rates, then, you know, all bets are off, because the pressure, because the amount of commitments they've made assuming that 10x growth rate continues will mean that even if it slips to a 2 or 3x growth rate, there's going to be a mad scramble.

Yes, Harry, it will impact it. It might impact at 1% or 50%. What you're really saying, what the question you're really trying to ask is, does it reduce that growth rate to sub-100% within one or two years? Right? And the answer to that question is, I genuinely don't know. And if I did, I'd be trading that stock. Because if, let me be clear, if you know the answer to that question, that one question, you know the answer to the entire direction of the US stock market for the next two years, because all the hyperscaler RPO, all of it is a function of the commitments they've gotten from the foundation model companies. And yeah, you can say if the open-weight models explode, there will be demand for inference. And yeah, you will have this kind of transition from, oh, I sold it to OpenAI, but I should have sold it to, I don't know, Cursor or Base10 or someone else, and the capex will get repurposed, but it will be a big-ass dislocation, and I just genuinely don't know. I mean, it's the million-dollar question.

I think the tough, the really tough part, I mean, it's Captain Obvious, right? Is, can they afford for it not to? And what I mean is, look at what's happened with Fable this week. Okay, Fable went from, you can't use it, it's not secure. Then the government let you use it. Then, hey, we're going to turn it off except for variable usage on July 15th. Now it can be 50% of your whole usage for the month. Why did they change when they don't even have enough capacity to serve it? Competition, right? Competition, right? So if, listen, if they price Fable at Sonnet rates, I think they'll own the market. Yes. I'm oversimplifying because you don't need Fable for anything, but literally, so the question is, can they afford to compete? And this is the stressor, right? Of course they could have 17 variants of the model at 17 price points. Like, that's not the issue. The issue is because they have to pay to train these damn models and other reasons. They have just this high cost base and they're subsidizing it with venture capital, right? Whether we call this venture capital or not, private capital. And so, but listen, you just cut the price of Fable 5 by half tomorrow, you don't need these Kimis, Geminis, but can they afford to? And over what schedule? And the fact that you can use Fable for half your credits is pretty telling, right? They're pushing it as far as they can, but that's the limit today. They can afford to compete 50%.

And we don't have time for it because I do think we should spend at least half the show on stuff other than AI model companies. But I think Jason's insight is correct about price. And if this was a software product with no gross cost of goods sold, that's what they do. I mean, Microsoft, I mean, and this is one of the big... I've seen a bunch of articles on this. Again, it's as you'd say, Jason, Captain Obvious, but it's worth emphasizing. In the great software wars of the last couple of decades, someone like Microsoft was able to just use price ruthlessly because there was zero cost of goods sold, and they just bundled the browser in with the operating system, bundled all Office in together, didn't cost them anything, and it just wiped everyone else out. But as you're pointing out here, there are real, even at the margin, even after you fully paid for your training costs, there are real physical costs to serve these models. And you got to cover your nut. You got to cover the marginal cost of the model, of the inference, which gets you to, I don't know, two, three bucks kind of blended average token. Then you got to recover the cost of the training, and you got to recover it pretty damn quick because it only lasts, you know, 12, 24 months before it's obsolete. And then on top of that, you want to make extraordinary profits because you're being valued at 20 times revenues. And if you're valued at 20 times revenues, you better be like Microsoft with 40% operating margins. So when you look at all that, you're right. You know, it's kind of back to the thing I said. You can squint at that and say, 'Ooh, there's lots of things that could go wrong here.' When you look at those fundamentals, as yet, the thing that's saving you right now, if you're growing 10x year on year and you have any kind of positive and improving gross margin, it just covers all the nut, right? The minute that growth rate stops, Jason, if the only way you can keep that growth rate up is by lowering your kind of price per token, then your gross margins start to deteriorate instead of continue to improve. That in itself would be a different ballgame. So you are right, price could solve it, but it would be a painful way to solve it.

Yeah, you have to start building your own chips and building your own everything, all the stuff you're trying to do to solve this problem. But I think it's just a... I mean, not to be... I think it's just a pricing problem, right? I mean, there's a bunch of issues underneath, but... And I would argue they've already bundled it. Like the consumer apps of Claude especially, but also ChatGPT, they bundled everything. I can get $10,000 worth of tokens for 200 bucks and I can just do just about anything in it, right? It's just outside of the consumer, it's not massively bundled and subsidized, right? Just because you're still... Yeah. You remember the old days in software, Jason, when you'd have to say, 'I promise I'm only using this for personal use.' You remember that? And licensing, right? Well, if you're telling those nice Claude people that you're only using your personal subscription for personal use, they're going to find you, dude. They're going to find you. Well, yeah. It's just... Fable is very good. Yes. That's why they're going to find a way to charge for it. Very good.

You know, context is so funny. I don't want to get distracted. We're like, how much did Ben... Ben sold it for 500-some-odd million to Netflix, right? 87. Our jaws drop, and like, and we're arguing whether we should sell a portfolio company for 6 billion. Well, is it really worth our time, gentlemen? It's really only a 4x to the last round. You know, and on the last fund, it's not even a returner. I don't even know. I don't even know if I may not even be retained as a GP at the firm if this is as good as I can do. Oh my god, he sold the company for 500 million. Rory, do you know what? I find that triggering. Look, good on him. I mean, look, I mean, you know, no surprise. It turns out you can make more money with capitalism, techno-capitalism, than acting. I mean, you know, turns out Bill Gates is richer than, you know, the most famous actor in the world, right? It's... Yeah. No surprise. All right, Rory, I'm going to listen to you then. Move away from this AI pure-play discussion, and we're going to move to some Irish twins. The Stripe and Advent deal to take PayPal private. Does that sound okay? Passes the... Good. I mean, you got to talk about it. It's... We got to talk about it. So, this is a big deal. Was it inevitable Stripe would acquire PayPal? There were rumors of it a couple of months ago. This is obviously taking those rumors one step further with the offer. Rory, how did you think about it?

I think that price clears all... I mean, I think it's super interesting in a lot of different ways. One is just a diff... I mean, they both process kind of 1.9, 1.8 trillion a year, right? And as yet, Stripe, and we'll talk about revenues and profits in a second, Stripe is valued at like 150 billion. And I think, what was the offer for PayPal? I looked at it this morning, but didn't... Is it 50 bill, 35? Hang on. It's about... I thought it was 58 or 60. 50-something billion, right? And so, yeah, I mean, it's like Stripe taking advantage of PayPal trading at sub-10 times profits and deciding to go for it here, right? I mean, in one sense, it's a ballsy move because you're taking on a lot of operational complexity. On the other hand, it's a chance to really, you know, transform and double your footprint. Because as I say, the revenue, not the revenue, the payment processing is roughly the same. Revenue is tricky because Stripe reports revenue net, which is around six billion, plus or minus. PayPal reports gross, and I think it was... and I checked it, but with my cold, I'm a bit feeble-minded today. It was about, you know, around 30 billion. So it was trading about 1.7 times revenues. So if you look at that five versus 30, I'm like, 'Oh, it's 5x. PayPal's 5x bigger.' But it turns out on a like-for-like basis, PayPal is still bigger, but it's about one and a half times the size. It's still a company buying something one and a half times its size for what looks like, you know, a third less, because they're doing a joint deal with Advent, a PE provider. So for, you know, a lot less of its market cap. If they pull it off, they will look back and go, 'Wow, that was an amazing deal,' right? It also gives them, and you know, and their economics will be amazing. It's a little like the Dell transaction. You know, obviously the scary thing is it takes your perfectly wonderful company that's nice and running smoothly and, you know, is a desirable place to work and all the positives that we all know about Stripe, you know, smartest guys ever, killing it, nice place to work, good reputation, and they're going to have to do a lot of hard-nosed stuff to turn PayPal around. And there'll be a lot more pushing and shoving in the future because, you know, you're probably going to be looking at that place and saying, 'We're going to get rid of a lot of people. We're going to rationalize a lot of stuff.' So it's a different muscle, but I give them credit for it. It's a big ballsy play to double your market cap.

Yeah. The part that I struggle with a little bit, you know, listen, obviously there's at least a decent synergy here, right? And in a PowerPoint slide, there's a ton of synergy. Plus you get Venmo, you get a lot of stuff, but two methods. The thing that is always a head-scratcher to me is blending something that's growing 7%. Because no matter what you say or do, unless you can radically shove those products through your channel, your blended growth rate goes down. What's Stripe growing today? I don't know, 30, 40%? It's between 20 and 30. So it's not that much bigger, Jason. That's why. But seven, but seven. So, okay, hold on. Help me, Rory. You're better at math than me. But if I take 30 and seven, that's 37, and divide by two, I'm only growing like 18% now. I've fallen below the Mendoza line of 20% growth at scale. There's no such thing as a Mendoza line for growth at 5 billion and above because you can get out, right? I mean, I think the real point is... but to take... but I found it stressful in M&A observations. Not quite at the scale, mind you, but it is stressful when it meaningfully decelerates you, right? It will meaningfully decelerate them in the short term. Even if I'm not sure how the accounting works, right? Even maybe they only have to recognize half of it because of this Advent thing, but they're going to have to recognize some of this revenue, right? As a joint venture, right? So it's going to accelerate their growth. It's not stress-free for... plus you have the operational need. Plus, I mean, even all the layoffs they're going to do, that alone may not reaccelerate growth. We've certainly seen this at a handful of portfolio companies, right? That's just the stressor for me. I've learned over the years that when you have one messy codebase and another codebase and you're like, 'How the hell are you going to combine these companies in different motions?' You figure, as crazy as it sounds, you actually figure that part out. And the answer is you don't fix it. You fix it over five years, or you have an LLM lift. But the real answer is you don't fix a lot of these things that seem like you can't rationalize them between the organizations. Everyone's got 11 products spaghetti together. Even tech leaders have it, right? It's just the nature of M&A. My guess is this is one where you have, frankly, one well-run company for the last decade and a half in Stripe, and you have another company that, you know, ever since the PayPal mafia walked out, has been just a revolving door of executives and is a real mess, and they've dissipated their opportunity. So yes, I mean, the interesting thing is normally this is the kind of deal you do after you go public because you have the market cap and you just price the deal. And, you know, I was thinking, my first glance was, ooh, it's probably a lot harder to do this as a private company because you can't issue 50 billion of stock, right? So you have to look at debt, you have to do Advent on the other hand. And again, I wanted to read the detail, I didn't get to it fully before this meeting. On the... maybe they're using Advent to almost keep it slightly off balance sheet for a period of time while they rationalize it, right? So I don't know. It would be easier to consummate this deal and just be done as a public company, but obviously Stripe has chosen not to go public. So, at least yet. And so, but it may well be that even though that makes it less easy to do, it may also have pushed into this kind of contained strategy with Advent, right? Yeah.

Let me just step back. Rory's got even more experience, the two of us, but it's just a dance. The board rejected it, right? And the fact that the board rejected it means to me that they're going to accept it. You reject it because no no investment bank will tell you you're allowed to make your highest offer up front. It's like not a it's like you probably breach your fiduciary duty if you make your you have to offer like uh whatever. You have to have another five or 10% to put into the deal. So it's a dance. They they they're going to accept it. They're just uh it's a bunch of mercenaries in a brand new CEO who's probably going to make nine figures for uh 10 or 12 months of work. They're go by rejecting it. It means they're going to accept it.