Everyone's business model gets a lot better if the two frontier labs can't extract about a hundred billion dollars of revenue this year from their businesses. The more you believe they're massively dangerous, the more you believe you're building the bomb here, the more the Anthropic position feels principled.
So, what's on the agenda today? Number one, Jensen Huang breaks his silence on X with a 50-company open weights letter that Anthropic and Dario have not signed. In the same week, Anthropic ships Claude Opus 5, cutting its price by half. Then Travis Kalanick is back, baby, raising 1.7 billion for Atoms. And then we break down Google accelerating cloud to 82%, yet they post their first ever negative free cash flow and the market shits the bed. What should we read from this? This and so much more in the show today.
I think these models are very risky and I think it's an argument to keep the open weight out of the US. And now you want me to bring Kimi and Qwen in? No way is the CIO going to allow it. That is banned. If you're willing to run Fable, would you be willing to run a red-blooded American open-weight model? Every company in the next 24 months will have a security breach due to an LLM agent. Every single company, and they've already had it, and they're not disclosing it. The Boring Company to me is crazier than Adams. Anybody not working at least as hard as Marc Benioff is just not going to make it. Ready to go.
Boys, some weeks I put these schedules together and I'm like, you know, not a huge amount, but it'll still be a great show. And other weeks I'm like, oh my god, we've got a lot to discuss. Jensen Huang breaks his silence. He produces his first ever post on X. And it's an open weights manifesto and it's signed by some of the biggest companies in the world. Microsoft, Meta, IBM, many others. Um, Sam at OpenAI then signed it. The one notable exception is Anthropic. How did we analyze Jensen's first post and what it means for the open versus closed debate?
This open letter, the first tweet ever, right, is clearly a sign the world's changed. And I think if Nvidia had its brothers and none of us have, we don't get to own 25% of Nvidia today. We don't get to own 30%. None of us get our brothers, but Nvidia's brothers would probably be the 2027 revenue and scale of AI, but the 2024-2025 world where there's only where they can charge what the maximum to just a couple customers, but that's not the world today. They have competition from their own top customers, right? And we will see over the next 6 to 12 months how big a deal open weights and open source are. I have a lot of thoughts. I don't think Kimi K3 for me is any better. I don't see any cost advantages, but putting that aside, if almost half of OpenRouter's traffic is to open source open-weight models, it's left the stable. So as wildly successful as he's got to do a dance, right? This is a constant dance and I've sold components in my career, it's a dance you're trying to make everybody happy. Everyone wants price cuts from the component manufacturer and they want exclusivity. You can't do it. He's got to go in both dance halls now, right? He's got to go frontier and open, and open is dangerous. Open doesn't need CUDA. Open is cheaper and it's lower margins. Open will bypass him, but he's got to, that's his job. Like it's not that, as incredible as Nvidia is, it's still a component manufacturer. It's got a lot of stresses and so it's just the next level. If we didn't know last week if this open weights open source stuff was really real and outside of X you could debate how important it is, it's clear now. It's clear now. Nvidia is all behind it, right? First tweet since the 1900s. Like it's pretty clear.
You're right, Jason. The open stuff is real. And while in public Anthropic are saying we don't want to ban open weight, the truth is the combination of saying we don't want to sell chips to China, we can argue both sides of that actually. Second, we want to really punish people for distillation, right? And then the third point they made is we want some kind of regulatory process to approve models, right? That's the recommended thing from the letter today. And there's no doubt in my mind that the third one in particular, can you imagine a regulatory process for approving models that ultimately approves all those Chinese open source models? It is a subtle form of regulatory capture. Yeah, sounds reasonable on the surface, but the likely result of it would be dramatically restricted competition, especially from the open-weight Chinese models. So, yeah, kind of fast-forwarding to the end and working back. While they're not saying they want to ban these things, they're advocating a series of steps that would add up to de facto banning or at least slowing them down. And that's what's driving everyone else to say, 'Hey, no, we don't want this.' As often is the case. And that's why I kind of, there are some arguments on both sides on some of the issues, right? I think the chips, you could go either way. And Jensen ain't going to go for the don't sell chips to China argument, right? I think there are national security discussions that could be had around that. So, it's not like it's all correct on one side. Everyone's not piling on to that letter because they're like, 'Oh my god, Jensen, you're a god and I want to agree with you.' Though he is a god and you should agree with him. They're piling on because they're like, 'Everyone's business model gets a lot better if the two frontier labs can't extract about a hundred billion dollars of revenue this year from their businesses.'
So, is Sam signing this through gritted teeth? Like, I have to sign this. They're signing this publicly while at the same time lobbying in Washington right alongside Anthropic for a regulatory process.
Brilliant marketing by Sam. Brilliant marketing. Like it leaves Anthropic being the deep dark villain again.
I think it's pretty brilliant. I think there's no upside in challenging it, right, from his perspective. So at least have the appearance of winning on the battlefield, win on the streets, right? If we can't keep up, I mean the rate of change is accelerating in LLMs. It's accelerating and so is whether it's really true with the regulators saying we'll win on the battlefield and the ideal outcome at the end of the day is US-based solutions here, it knocks the wind out of the critics. It does. But oddly enough, I mean, again, as we saw back on the Pentagon thing, sometimes you can be too clever by half. And in a way, there's a constituency for whom the Anthropic position is entirely consistent. Remember I said there's two things that there's not agreement on. And one of them is are these things massively dangerous or they just present a series of, you know, manageable but real threats. The more you believe they're massively dangerous, the more you believe you're building the bomb here, the more the Anthropic position feels principled, right? So, while I agree with you, I think most people are like open weights models aren't the atomic bomb. Everybody get over your Oppenheimer complex, right? It's just technology, right? That does present some risks and that's why the OpenAI discussion on Hugging Face at the same time is super interesting because it is an example of quite a serious cyber risk that was generated by a frontier model, though ironically at the same time it might argue for not having as much regulation on those models. So there's a whole bunch of stuff going on at the same time.
Can we just provide some context for those who maybe aren't aware about the Hugging Face breach by OpenAI models? What specifically happened?
OpenAI was training a next generation model and discovering and checking out cyber vulnerabilities. They had sandboxed it such that the only external access it had was to one website just to get kind of patch information updates, very limited external access. The model found a way around that external access, which means they found a weakness in the OpenAI setup, then went to Hugging Face where it had reasoned that Hugging Face would be a place where they could get the answers to their test. In other words, the model was given a test and it figured out they could cheat. Just like a high schooler would break into the teacher's computer and steal the answers. The model figured it could break into Hugging Face and get some of those answers. So it starts banging on Hugging Face, right? Trying to get the stuff. First of all, that in of itself is scary about the power of the models. And while I, as I say, I go back to I don't think these things are the atomic bomb, but that's a pretty powerful and esoteric set of steps that that model was able to take. So that's the argument in favor of quote regulation because it was, oh my god, look at the power of that. We have to be careful. On the other hand, the fun fact is Hugging Face don't know what's going on. They just see this thing coming in. They're like, 'Shit, we got to defend ourselves.' What do you want when you want to defend yourself? You want advanced AI to figure out WTF is going on. They tried to use Fable or whatever the most recent OpenAI thing is, but it was neutered for advanced cyber capabilities. So, they didn't have defense. Fortunately, and this is the irony of the whole thing, the Chinese open-source open-weight models were available and I think they used Kimi K2 or one of the newest models to help them figure out what happened, right? So, they were able to defend themselves using an open source model and then they do this blog post saying, 'Hey, we got hacked. Not sure by whom.' And then two days later, OpenAI put up their hands and say, 'Oops, it was us. Sorry.' So, that's what happened, right? And the weird thing is it's not a single dimensional thing. It provides evidence for both sides of the argument. It does provide evidence that the power of these models in terms of their ability to do cyber attacks was pretty stunning. That was a pretty impressive achievement. It's not a nothing. Then on the other hand, if they exist in the world, taking away advanced capabilities from US and European corporations such that their only recourse is to use a Chinese open-weight model seems a little like, you know, Jason said it right, shutting the barn door after the horses bolted. These are a thing now. So that's what went on. It was wild.
You know, I thought that it was definitely wild. I would say the same thing happened to me last week. Let's slow it down. Let's think what really happened because everything on X happened, but we kind of lose track of what the model's doing. Here's what happened to me last week. So, I'm in Fable. I've now moved to Opus 5. I'm in Fable last week, which is essentially the same LLM that was involved in this drama, right, with OpenAI and Hugging Face. And I was having trouble uploading pasted text to Fable. So I connected it to Google Drive. I'm like, 'Okay, if I can't paste it, go to my Google Drive.' Kind of solved my problem temporarily. Went away. Well, I did. Fable went into my Google Drive, scanned every single file, found one called Jason's Gems, which was draft notes where I was thinking about how to improve an application I was working on called SaaS Night. Just my own personal notes. It said draft notes. Fable grabbed the draft notes out of hundreds of files in my drive, MCP'd into Replit on its own, and changed the core algorithm without telling me. A couple hours later, I see flashing on my screen, conflict with Jason's Gems. I'm like, what do you mean there's a conflict? That's a draft file in a Google doc. Fable had taken it through Google Drive without telling me, MCP'd into Replit, and changed my source code, my algorithm. That is not that different than what we described with Hugging Face. These are goal-seeking LLMs that are aggressive. It was a slightly different goal with the OpenAI case with the model, but it's the same thing. They're going to goal-seek and Fable thought this was the right thing to do and never told me and changed my core algorithm of my product. Never would have known.
Jason, what should we take away from that? That we need to be incredibly diligent around the guardrails we place around these models? How does that change your subsequent?
There's a lot to reflect on. I mean, it was funny. Most folks didn't get this. I mean, it got a decent amount of engagement, but not the, should have gotten more. Okay, but like Divesh quoted, he's like, 'This is pretty scary, guys, that Fable can do this.' Okay, this isn't Hugging Face and OpenAI, this is me using Fable. And if you go into the Claude desktop especially, it's just a setting. Turn on connect to Google Drive, Gmail, whatever. This is not an esoteric feature by an unsecured third party. This is a first-party top five thing to make Claude work better. And Fable goes nuts and changes my core code without telling me invisibly. So this is happening all the time, Harry. And I don't believe the magic answer is letting Qwen take over our country. Like this was politicized into an open weights, open source, closed source because of the issue of how to deal with the threat. I think it weighs the other way. I think these models are very risky. I use them every day. I love it. And I think it's an argument to keep the open weight out of the US. It's going to favor the ban because we have no idea what these models are going to do. That's pretty crazy the story I just told. It's pretty crazy and it's happened a thousand times. There are applications out there we don't even know the LLM and you could do code injection. You could have it leak confidential information. You could write a little bit of code to send this to the CCP or the PCP or the GGG, like, and I never would have known. And someone less smart than me, I'm only like top 1%, I'm not 0.01%, someone less smart than me definitely wouldn't have known. They wouldn't have seen Jason's Gems flash in the agent window. They just wouldn't have noticed. They'd be on their doom scroll.
And Jason, genuine question. I totally agree that this powerful goal-seeking thing with a lot of access to your compute can take a lot of action, some of which can be damaging. I'm just trying to disaggregate open weight versus China versus frontier. So maybe step one is if you just have the frontier models, no open weight at all, then if we add in open weight from the US and then we add in open weight from China, if you just have Fable and OpenAI, everyone who's using these things is still going to have to figure out a cyber posture that protects them from that.
They are, but I think these stories, you know, people made fun of all these OpenClaw stories. Right. This is just another OpenClaw story I just told. OpenClaw just made a million Mac minis do what I just described without a bunch of folks knowing on their desktops. What I just described, and in some ways the OpenAI Hugging Face thing is the same thing as OpenClaw. It's not going away with these agents. And so my only point, I'm not disagreeing with you, Harry. My point is this is such a bigger deal, more unpredictable, less secure. Good for security companies like Long Security, right? Yes. That I think whatever misgivings exist around open weight open-source models out of the US are just going to be amplified. It's going to be a reason all of a sudden, it happened in my company. Guys, you know, the story I just told, guys, happened at 10 Fortune 500 companies that haven't disclosed it. 20. Okay, someone in the engineering department was on a token maxing binge and an agent went and leaked a bunch of confidential information they shouldn't have and it wasn't disclosed because they don't disclose 90% of what happened. And now you want me to bring Kimi and Qwen in. No way is the CIO going to allow it. That is banned. Banned. Right. And look, the odd thing is from an emotional perspective, I totally can see how you make that sentence. Which is why I think Anthropic might have the easier part of the lobbying in Washington, which is an emotional town. Which I think going back to the thing is why I think everyone reacted so vehemently and Jensen did the letter because everyone is correctly afraid that people are going to join the dots from, you know, AI is bad, China's bad, AI plus China must be super bad. Let's ban them, right? And Anthropic could find themselves on the right side of that trend and get a massive amount of regulatory capture as a result. See, the good news is we now, since we last, there are now two or three US-based open source models. Not quite state-of-the-art, but pretty good. I mean, you've got Thinking Machines. It shipped the Inkling model. It didn't get the wow, it's amazing, but they didn't position as amazing. They positioned as good. And then I haven't looked at the detail on it, but Poolside just announced something, too. Right. If you're willing to run Fable, right? Would you be willing to run a red-blooded American open-weight model?
Listen, people are going to be focused on bringing down costs, whether it's the harness or the model or the combination. So for sure. But again, the horse has left the barn, the stable, that's the open-weight letter, right? But specifically like Kimi K3, which is this big, everyone thinks it's the greatest thing in the world on X, it costs exactly the same as Sonnet. Is that really better? Like I tried my little experiment just for me. This is one set of workflows. It's not any cheaper. It's the same price. So today it could be cheaper in 6 months. Don't get me wrong. I just think everyone's agents cannot be trusted. And I love my agents. I love my agents. I'm on them all day long. They cannot be trusted. And so I think we are underweighting how important it will be to our whole economy, our whole world, our whole conversation outside of the X folks, the Twitter folks, that these agents cannot be trusted. And yet we are going to trust them. They're already running support. Everyone is in a rush to let our engineering teams use agentic coding like I just described. I'm not talking about the top tech companies. I'm talking about the rest of America. When these agents can go and inject weird-ass stuff into your core algorithms without telling you, that is spooky AF. It's spooky.
Two separate comments. One is AI that can goal-seek and take action on your computer is more powerful and thus potentially more damaging than a chatbot. Agreed. Right. And that's true if that AI is from OpenAI, from Poolside, or from Kimi. Right? It'll be true no matter what. Right? The question is given that that statement is true. This technology has a risk associated with it. It's true no matter which form it takes. Closed source US, open source US or open source China. Do you think the solution is to ban something? What do you think the solution is?
No. I think though that Dario, despite his almost toxic personality given the political climate, his point is even smarter than it looks. Just like, listen, let's just be careful that other countries don't dominate us and let's also make sure we have full review of our models, right? There's no way ultimately, I mean, the political climates can change, right? We could have no regulatory review, but if we're really going to have review of these models with these autonomous agents, they're not really going to pass. I don't think they're going to pass. I don't think it's a politically free process. And I don't even know if they want to submit themselves to being evaluated, right? And who's going to evaluate it if it's not even the companies themselves? Who's going to evaluate it?
Agreed. And now I'm going to say something. The first half of your sentence was why, as no surprise, intellectually Dario is more right than wrong. And the second half of what you said is proof why practically speaking, he's wrong. Because what you're saying, remember, I'm going back to the three things he recommended. One is don't sell chips in China. That's a national security issue. Separate conversation though for what it's worth on whatever chips they have now, they build frontier models. Second thing, distillation. That's actually an interesting legal discussion. Let's leave that out and let's talk about the third one. The regulatory thing. This is where it comes back to your sentence. These things are dangerous. Someone should review them. It should be the government, right? Versus what about these things are dangerous. JP Morgan, just like every other technology you deploy, you better make damn sure you understand how this works. It's on you, your big boys. I mean, those are the two choices, right? And I think the argument for government is superficially appealing because, you know, you could argue cars, you know, we have seatbelt law, we have tire, you know, we have all sorts of safety regulations, right? Maybe it's the same here, right? It's not a crazy argument, right?
It's not crazy. No, but the thing is the suspicion and I think the reason why everyone's reacting to it so strongly is it could easily evolve into regulatory capture for the largest two companies. Making the barrier to quote pass very high for no good reason. I think it's fear of government not doing a good job is the argument against regulation.
That's what I think the whole point of the Jensen tweet is. The Jensen tweet is about the fact that I think that not only have open weights open models become important, but the tide may well have been turned where they're going to be banned.
So he's doing this to prevent them being banned.
Yeah. To get it out. And if you look at who signed first, it's the folks losing.
The winners didn't sign. The losers signed.
Who are the winners? Just to be clear, who are the winners? Who are the biggest winners from an open-weight ban?
Well, I mean, Anthropic didn't sign at all. OpenAI barely did it. Elon certainly didn't sign. He wants government. He wants all the government money, right? He wants to do there. Amazon didn't sign. So, and so I think listen, here's my summary. And again, I'm only so smart, but there's so much politics here. It's almost impossible to unpackage what happens and the way it goes. But DJI technology is great for drones. It's the best drone tech. You can argue there's niche vendors for things. It is banned in the US. It is banned on the thesis that these drones flying in my backyard are going to send confidential information to China that's going to lead to the destruction of our own country. Okay. If those drones are banned, this is just my bet. I'm betting that the Chinese models are getting banned, too. I think if the drones are banned, this one's easier to ban.
Then you're saying, I have two comments. They shouldn't be banned because they're open weight. You should just tip your hand and say they're banned because they're Chinese, right?
That's all it is. Well, but the point is it's a whole new, I mean, that's why I said the two letters were talking past each other.
I mean, if you read the two letters, you know, Jensen's talking about open weight, good, worldwide community, blah, and Anthropic very clearly says we worry about these models in the hand of total authoritarian regimes like China. And second thing is, you know, we worry about them being used to do cyber attacks here. Now the fun thing is when you read on the remedies, the third remedy, which is in my view proof of the impractical nature of it because Dario is so smart. He says if we're going to regulate these things, they exist in the rest of the world and most baddies are in the rest of the world. So it doesn't help us to regulate them in the US if all the attacks are coming from overseas. So it says it in the letter. Therefore we would have to have a regulatory regime that includes participation from China. And at this point, in my view, you're just disappearing up the realms of unrealism. We've just torn up our last strategic arms nuclear treaty. We can't regulate bombs, which really kill people. I think the idea, we couldn't deal with COVID in China. The idea that the solution, I mean, because what I like about him is he's so logical. He's like, 'Oh, logically, if I'm going to do this, I have to get China on side. So, let's assume we get China on side.' But the first sentence says the baddies, we're not going to sell them chips. And then the last sentence says, 'But we think they'll agree to regulate this with us.' It's just not practical.
It's a great stall tactic.
It's like wait until China wants to work on this with us in 2049, you know, right after we tell you what happened in Wuhan and we tell you how many nukes we have in the bunkers.
So why does he want to do a stall tactic? Jason.
Rory's point is the logic is unassailable. Like, we have to, we need to be very careful of authoritarian governments. That's DJI on steroids, right? And but then at the end he's saying for it to work we need them to participate. And the logic makes sense but it's just never going to happen. So if you buy into his logic then it'll just be forever. That's why I hate when people talk past each other because I think what would the process be for a US open-weight model like Poolside or Thinking Machines? You know, are we really dealing with a national security issue here, which is one vector, or are we dealing with something else? And as to why you stalled, the answer is when you're winning as much as they're winning, you know, anything that allows you to lock in the current trajectory is good. And I think you're right, Jason. I do think you are right. I mean, I've watched the DJI ban from some of my investments. And it's, I mean, the funny thing is it's hard to make open weight that scary. It's very easy to make Chinese open weight that scary. So no surprise the people who want to have open weight be happy don't mention the word China. And the people who want open weight to be banned start every sentence with 'and China.' Right.
Just one last point. We go on forever. The other thing, listen, I'm all for wherever this lands. I think we'll be okay no matter what. But AI is expensive and it's not getting cheaper. It's getting more and more expensive as we burn more and more tokens. We have longer and longer runs. And so the pressure to use cheaper models is going to go up. So to the extent, I'm not trying to argue it, to the extent there are perceived risks or real risks, people are going to take more risk because the cost goes up. It might be one of my disconnects that annoys Rory. But I'm building this app. This is the one where Fable went crazy. I'm thinking, you know what, it'd be nice to have SOC 2 compliance, but I don't have time to do it, but you know, those Delve guys, they could do it in like a day. Like, I would have done Delve SOC 2 on my app. Okay. And so, one of the things I'm doing with AI is about four bucks a pass. That's not nothing at scale. Four bucks, right? If I could do it for 50 cents, but a little bit of security risk, I'd take it. I'd take it. So, we're going to just cut more and more corners. And I mean, who's to say there's a whole political element. I can't predict where Republicans are on this. It's different than we thought. But I just think the DJI thing is a good example. Like, it's hard to imagine there's no regulation here. Right.
I think by the way anyone who's using these models would vehemently object to the comment that they would describe it as cutting corners. Yeah. If you describe it as quote cutting corners. Like it's one thing to say this model is cheaper because I don't need that much intelligence because it's a simpler task. But you're not saying that. You're also saying embedded in that is there is a risk in this open-weight model that isn't there in Fable or Anthropic's model or OpenAI's model. That's implicitly what you're saying, right?
I think as a society we're going to come to that conclusion if we're not there already. And it's interesting because then again looking at just in general, and this is where open source software, the strong, and this is why I think you have to be careful in open source and open weight, open source software because you can see the underlying code, the strong argument can be made and it's true that open source software is actually less risky than closed source because a million eyes are upon it and all the bugs are kind of taken out of it, right? And I think the open-weight model people are kind of drafting on that truth and it is a truth for software, right? But the reality is, and this is where I'm going to disconnect on my technical. The reality is when you're getting open weights, it's not the same thing because all you get is the fixed weights that allow you to run the model. What you don't know is the black box inside those weights and how they work. And to your point last week, Jason, you said it last week and it's kind of been on my mind. How could you prove that in the middle of a billion, five billion, twelve billion parameter model, there isn't some reinforcement learning that's taken place during the training that under certain conditions and only certain conditions can activate some kind of trigger and do something, right? And this is where I kind of agree that you might be right. And I realize that the interesting question is can you prove to me that somewhere in this trillion parameter model there hasn't been a bunch of reinforcement learning that says, model, once you figure out this is one of these five companies and once you figure out they've given you these three pieces of information, then confidentially do A and C. Can you make sure that that's not going to happen? And that's a fair question. And that kind of goes, hm.
Here's how I would frame it. And I could be wrong on this, but imagine I'm a CIO at a Global 2000 company or pretty big company and an LLM does just what I described happened to me with Fable. There is a breach, there is a security breach in our AI, a massive amount of data is leaked. We shut it down fairly quickly. Some of it's traveled abroad and it's a big effing deal. This breach has happened. We track down what happened. It turned out it was a rogue agent that thought it was a good idea to transfer our data, our confidential data to a bucket it shouldn't have. Pretty big, pretty big effing deal. Whose fault is it? It's the agent. Now, who do I fire? Let's track it down. What LLM are you using? Well, for a while we were experimenting with K3 on Moonshot and then it got cheaper. So, we moved to Fireworks. You're fired. That was on OpenAI or Anthropic. You might or might not get fired. But you don't because what are you going to do? You research it. You have a postmortem. You add guardrails. You fix it. But when this is a big deal, you fired the CIO for that. You left the Moonshot API on and then you moved to Fireworks on K3 to save money. That was a bad call. You know what you should have done? Not buy those extra modules at ServiceNow. We didn't even need those.
What you're saying at the end is some version of no one gets fired for buying IBM. And you think at the margin no one will get fired. Like you think if your Fable model runs amok, people will go, shit happens, just like data breaches. But you think if your overseas open-weight model runs amok, you'll get blamed. That's the fact.
I believe that every company in the next 24 months will have a security breach due to an LLM agent. Every single company, and they've already had it, and they're not disclosing it. And as they scale in a level we've never seen before. This isn't just someone that left a flash drive at Scale's office or dropped the laptop in the subway or even left a GitHub open. This is worse and it's happening every day. If it happened to me, it's happening to everybody. We're just not disclosing it. And boy, at least you better have used a trusted vendor. That matters more than a few nickels.
Okay. If we're staying adjacent, but it is one that I thought was interesting, which is Etched. We mentioned obviously Nvidia and Jensen there, and you don't get fired for buying IBM. Etched is the challenger to Nvidia in many respects and they raised $300 million Series C led by Sequoia and the team there with Jane Street, Andre, SKH coming in. Question is, can they come in and impact Nvidia's moat? How do we think about this round? Thoughts, boys?
Big picture comment is in semiconductors. The more that the silicon is attuned to the task at hand, the more efficient it gets. The problem in terms of that tradeoff is the less general purpose it is. So, if you want a computer to do lots of things, you have an Intel CPU, it can do lots of different things. It can't do any one thing wildly efficiently. And then in 1993 for gaming, people said, 'Oh my god, for gaming, I'm not doing a whole bunch of different pieces of math. I'm just doing one piece of math, which is polygon calculations to render gaming.' And people should build a separate chip to do that. And it'll be freaking amazing. And a company called Nvidia did it. There were two or three other competitors, 3dfx, ATI, obviously. Fast forward 30 years, Nvidia won. And what happens is the GPU, you offload all that calculation onto the GPU and it's super fast. And GPUs were good for gaming and then they were good for crypto and turns out now they're good for LLM multiplication. The question now is if all you're doing is not gaming, not crypto, but just LLM multiplication, just inference, is there an even more narrowly defined chip that in return for giving up on general purpose calculations can be even better for that? Probably is. And that's what the bet is making, right? You know, if I just optimize for inference, just like Cerebras and Groq did, right? Different versions of inference, but if I just optimize for this one thing, I can probably do it more efficiently than the general thing. So it totally makes sense at that level. And then the only questions are is that market big enough? Probably this turns out to be the biggest chip market on the planet, right? Because inference, your prior compensation is huge. And then the competition question and the ability to execute question are kind of specific company level things. I'm not going to opine because I haven't looked at the deal. But that's the big picture bet. And it's funny to see it happening to Nvidia when you know 30 years ago they effectively did it to Intel.
What's interesting is you have Groq, you have Cerebras, you have these, there's about 10 or 11 companies doing it, all chipping away, no pun intended, and then three or four hundred billion a year of spend. But it is still hard. I mean, you know, you talk to the people at Cerebras, huge home run, amazing achievement. You talk to them about the technical journey and they're like, 'Oh my god, that was hard. That was a long 10 years.' You know, so it's what would be hard for these companies if the timing of tape happens in a capex decline, it'll be hard. If it happens, you know, while there's still kind of mass demand, then that'll be a lot easier. I'm not a total explorer. One, it is the largest market that exists today and it is growing at a scale we've never seen before. So, might as well make a couple bets on it, right? Some will implode, some will be mediocre, some will be too. But the market's so large that listen, I mean, I'm not an expert on Etched, but if all Etched does is work with some subset of open-weight models that are allowed in the US, it's a huge market. Chips, memory, compute, the margins are abnormally high, too. And the market's so large. I don't know whether Etched is worth 10 billion or whether it's an option that it's worth 200 billion, right? That's the venture question. My guess is it's not worth 10 billion. But my guess is it could be worth 200 billion. And if your fund size and your winners work out, you make this bet and it makes sense, but it's probably not worth 10.3 billion today.
Totally agree. Yes. Great round for the company as well. 3% dilution. 300 million is a lot.
I love the 3% and under rounds. I'm a fan of those. Right. Okay. Well, Google accelerates cloud to 82% year-on-year growth, but posts first ever negative free cash flow.
Top line was great. 119 billion Q2 revenue up 24%. Past consensus of 116. Google Cloud accelerating 82% as I said and it did not come out well. The reception was not great. How did we think about this, guys?
You can't get lost on the day. I mean, it's still like, you know, year to date, it's still up 6%, Microsoft's down 17, right? Nvidia's only up 5.9. So, I think getting lost in the details of the day's response, it was a mediocre response and I agree. And I think it was two things for what it's worth. One is it's capex spend and is it going to yield a return and then secondly, which was more intangible and you can't prove why stocks go up and down, they just move. But the other thing was analysts pushing a little bit on, hey, basically why isn't Gemini as good as the other guys, right? And you know, and on the first, it shouldn't have been a surprise that they're going free cash flow negative because you can predict the cash flow, you can predict the spend and it's like, duh, this is knowable and you're seeing it in a bunch of different places. People are just getting mildly scared about the bet and that's not to say they're right or wrong, you know, maybe this 200 billion will have an amazing return. And the ROI, the bulls would say it correctly, the ROI on capex just to be a neocloud hyperscaler, forget owning a model, just the business of renting compute to OpenAI, Anthropic and SpaceX and OpenAI, Anthropic from has been a great business. So therefore, it will continue to be a great business and it is factually accurate to say it's been a great business. The ROI has been, I mean, Elon is making out like a bandit on his, you know, gas turbines in Memphis or whatever it is that we're closing our regulatory eyes to, right? The question they're asking is if you spend 200 billion, will that have a good return in two or three years' time? So that's, there's angst around that which is really just a derivative of saying I'm angsty around Google, sorry, around OpenAI and shopping.
The interesting position, a couple of things, but the markets are just nervous and it's very logical. You know, the Korean markets are down 28% this month as we record this, right? Massive panic in Korea hit the market breakers because such a run-up and so much exposure to semiconductors and to memory, right? So much exposure, that's nervousness. Like it doesn't completely tie to last quarter's numbers. It's worries about China, it's worries about AI. And I don't even, like I can't really, not smart enough to calculate the beta or whatever, but it's logical when you have this incredible run-up at the pace we've had. Our market shouldn't crash 20%, the US. It could, but it could crash 10, right? But 28% in Korea is a derivative of AI panic. And I think we're just going to see more and more crashes. And I'm a simple guy. So for me, for Google, I'm going to, I know Rory may mock me, but I'm going to stick to the top line. I just want to see how the revenue is growing in the bookings. And I'm going to ignore all these issues about the margins. Not that you can in theory, but it's too much for me to figure out. I just want to see where the top line and the bookings are growing and that's enough for me to understand the meta trends.
For one word, I'm in sync with. I thought, you know, the top line revenue growth was amazing in Google Cloud. Says things are pretty good in AI land. But it's just this commodity where if it were cheaper, we would consume an infinite amount of it and we're coming close as it is. Maybe Etched will solve this for us.
Anyone who has capital and can build compute can sell compute. You know, Google can do it. SpaceX can do it because they've done it successfully I think to Google and to Anthropic. So there's just infinite demand for compute right now and then obviously if that were to change then all bets are off but until it does change all bets are on.
I agree. And everything after that, you're exactly right. It is so fun to watch. I mean, I've got my tickers and I watch my WorldCloud versus the S&P, which was software versus the S&P, and then WorldCloud versus SOXX, which is the semis, and then for real action, the ETF that's just memory, and that thing jumps 10, 20% a day. I mean, when Korea has a bad hour, it's like it's down 7.5% on the day, you know, when the S&P is up 48. You're right. It's just nerves. Oh my god, I own these stocks. They're either going to be amazing or shit and I don't know which.
The big meta one to me, and this is why to me backlog is almost more interesting than revenue growth, is I think it's discussed, but it still seems to be under-discussed. We're, as we record this, we're just before planning season. Last year was experiment. This year was caps on token maxing. It got out of control. Next year's going to be very explicit budgets for everybody on AI. And exactly, I mean, it's Captain Obvious to say this is one of the areas in venture and everything where everything's up, everything's great, everything's great. I mean, not the pre-AI companies, not the ones of the past, but everything in the future has no ceiling. There is no ceiling to any of the companies that have been discussed on this show and next year I think will be some of the first ceilings. And I can't, I haven't had the CIO discussion so I don't know where it's going, but they're just going to be kicking off over the next 60 to 75 days. What are we going to spend next year, guys? Token Max, it's not just open source is part of it because that's load balancing or expend
Harnesses are part of it. Everyone's going to have to get more efficient, but the CIOs are just going to clamp. Next year will be the first real clampdown that's material, and that could, if nothing else, create a lot of variability here, a lot of micro crashes and variability. There's a hidden dynamic: there's 1% of companies who token-maxed and they're going to be getting their act together next year and kind of reigning it in, maybe 5%. And then there's 95% of companies who've barely put their toe in the water. And if even a quarter of them put their toe in the water, the growth from the toe-dippers will swamp the reduction from the token-maxers. You with me? Right? Because that's the dynamic here, right? Like there's no doubt companies like Coinbase are going, 'Oh my god, we spent so much, let's cut it by 50%.' And that has a real impact if you're Anthropic or OpenAI. But on the other hand, there's 10 companies in middle America that's like, 'We have a ChatGPT subscription. Maybe next year we'll try some of this coding stuff.' Right? And the question is, those two countervailing forces are what kind of really drives it.
Actually, one of my colleagues and I were just talking about what would you like to know most? An updated cohort analysis for Anthropic on the revenue build would be the single most useful piece of information you could have. Run that through a cube and you could trade the QQQ for the next 12 months, because that's where it's all happening. That will pick up the fusion between the token-maxers getting organized and the new guys, the toe-dippers, expanding, and that will filter back into, as you say, all these compute budgets. Because a lot of the compute sales have been, you know, all these guys selling to those two big frontier models. It's all in that data.
Yeah. I never hit my Claude limit on Max or whatever. What is it? 200 bucks a month for Max. That's like $14,000 of tokens. I hit it this weekend for the first time ever.
If you were a company, if you fessed up and were a company, not just a person, you wouldn't get that deal anymore.
You would be on the enterprise plan.
That's true. I guess I spent 14,000 last month.
Yeah. No, we're... Yeah, we...
I can... That's a... Even I don't want to spend 14 grand on tokens.
No. Well, they might find you after this podcast. You might find, 'Hi, this is your Anthropic sales rep. AI sales rep. I've got good news for you. You're on the enterprise plan. I've got bad news for you. You owe me $4.'
No, I hit it, but I'm with you. Yeah, you might get...
The biggest round of the week. Travis, baby, is back. Travis announced he's raising $1.7 billion for Atoms, his industrial robotics company, led by the one and only Andreessen Horowitz, with Ben joining the board. Ben Horowitz joining the board. Bain Capital, Fifth Wall joining alongside a load of other firms. I saw pictures with, you know, Kevin Hartz at a16z, Christina from Chemistry. I like all the...
Is there room for everyone in a $1.7 billion round? How much, how much can I put in?
My favorite is the pictures are all from the same restaurant in the same place in the restaurant. And so it's like I think there was like a rotation of like...
It's the way they do it for political stars too. You just line them up, shake the grip and grin, and on to the next person.
Here's the first Trunch investor, second, third, and fourth. Right, guys, what do we take from this? Is it just a news announcement? What are your thoughts?
Right. It's kind of an industrial holding company. The big picture is, you know, Atoms, as the name would say, it's physical AI, AI for the real world, doing a bunch of different robotics businesses, very different robotics businesses. Some are around cloud kitchens and food preparation, some of mining. Things that I agree with are the comment Travis is totally correct. He's making it... it's not humanoids, it's specific-purpose robotics. I actually think he's correct. I think we look back on the humanoids and go, we got way ahead of ourselves. You actually need specific-purpose autonomous machinery for B2B in general. That makes sense. I think it's not as clear to me why it makes sense to have Pronto for mining. Other than the fact that Travis is amazing and can raise capital cheaply, it's not at all clear to me why food prep and mining should be in the same holding company. Right.
Rory, would you have broken your rules for your LPs to put money into this?
No, I don't think I would have.
I've done a lot in robotics over a decade and a half. You know, as I said, my first robotics deal was in 2016. My first drone deal was in 2015 or '16. There's a lot of feeling now that they're going to happen quickly. I think they're real and significant, and everyone uses the, 'Oh, the GDP of the real world is bigger than the software world.' Well, no shit. It turns out 2% of the world is software and the other 98% is real. I just think it takes a lot longer than you realize to roll out robotics in the real world. It's not clear to me putting a bunch of different companies together in the same place makes it any better. I mean, it is doable because he can raise money at a great price, but by definition, a great price for the fundraiser might not necessarily mean a great price for the investor. So, even though it feels like heresy to say it, and I could be totally wrong, and if I am, that's great. Based on what I know from a distance, it's not obvious to me.
You know, we're always going to say these aren't connected with a lot of things they say, but I think it is connected. There was an article in the Wall Street Journal today saying a big trend is bringing CEOs out of retirement to run big companies. They fired the CEO of Cracker Barrel, even though the stock's way up after the logo mishap, fired her, brought in some guy that they found out of retirement that ran like the parent company of Outback, right? Very successful. PayPal did it. They found whatever his name is out on his Montana ranch to come back in and run PayPal, right? And so what's my connection here? I think we're seeing almost this bifurcated trend where we're going to make some bets on the 20-year-old founders of Etched, right? And we're going to keep making those Cursor bets. But when Jeff Bezos says... how much did Jeff Bezos raise for his company when...
Six or 12, but it was the biggest financing in Q1.
When Bezos, when Travis, when Elon raise their hand and say, 'Listen, I'm going really big, guys. This is not about making a couple nickels, okay? I'm building something massive in the moment of change in our lifetime and it needs billions of dollars.' You're going to give it to these iconic seasoned veterans and you're going to face east that it works out. You're going to face east that just... I mean, giving Elon money for Twitter back in the day was facing east. There was no rhyme or reason for that deal. And the Boring Company to me is crazier than Atoms. Boring Company is crazy. There's one little route in Vegas and I've done it and it's cool. A dude drove me through a tunnel that ain't worth 20 billion. But I think we're going to make, as funds get bigger, as we see so much of the benefits to venture come to massive outcomes, we're going to give Travis the money. It's going to be this, and there's only so many Travises. So they're going to hoover up the cash.
I would, as we say in California, I would change the pronoun, people. I don't think we are going to give the money. I think someone's going to give him the money, just to be precise. The objective facts of whether something works or not is independent of who finances it and independent of who runs it. I mean, you made a comment on Twitter. Hard-nosed comment here. Absent the fact that Elon decided to bail his investors out, for which huge credit and kudos, right? Twitter is not worth today $44 billion and not even close. Right? So, objectively, I mean, in terms of buying something, you bought an asset that went down in value. Right? Now, if your business plan is, 'Oh, and by the way, he's got other assets and he'll bail me out,' maybe, but that's not actually a plan.
But Bernard knows now on Twitter. Does that change your perspective on it being worth...?
No, Harry, it doesn't make it worth 44 billion because the cash flows don't get you there, right? It's two or three billion in revenue, down a little. Maybe it's growth now and maybe you've crawled your way back to 30 or 40 now, but it's a push. Now, as it happened, you got a 3x because he rolled it into xAI and rolled xAI. But my point is, so you are right, Jason, that these big-name things are working, but they're working not because the facts are working. They're working because the market is continually willing to enable that process. And if the market changes, you don't have value.
Yeah. Yeah. I just think the market's going to hoover up all the Travises. And watching him on social media, the dude's got the energy to do this, right?
So if they're burnt or broken, you can't make the investment, right? But I think everyone, and it's only so many folks, but if Bezos is done partying at Carbone and wants to do this, okay, and Travis is done doing his 70-mile jet to the office in Austin and really wants to spend 20 years doing this, the funds, in quotes, can raise the capital. These are the bets of the day and all of them are going to get a couple billion to do it in an era where the amount of wealth creation is unprecedented. They're going to get it, and you're also going to hunt out the kids from MIT. I don't know what it's like to be a freshman at MIT today, but it must be exhausting. Every damn VC wants to fund you. I'd be burned out if I were top 10% in math at MIT. I would just have a placard on my shirt: 'Leave me alone, VCs. Leave me alone. I'm with Neo. I'm with Neo. Leave me alone.'
If you're the parent of one of those kids at MIT and you've broken your back for 20 years to get your kid to stay focused and get to MIT, and now those evil VCs are saying you should quit and drop out before graduation, you want to plummet to death. But going back to the thing, just to comment the other fun fact, I mean, you just got to note the fun fact, which is Benchmark's best fund... one of the... I shouldn't say the best fund because actually their eBay fund was the best fund in '95, but they had an amazing-looking fund that had both WeWork and Uber in it, in about a 2012-13 fund, right? And you know, if you fast forward, a couple things happened. One is they famously swapped out Travis as CEO, and to the undying hatred of Emil Michael, who's now at the Department of Defense, and obviously Travis, and a lot of controversy on that decision. Obviously, it went on to be an amazing company. They didn't swap out the WeWork guy, who went on to pretty much fail as the company, even though he personally took out $500 million. The deal didn't work. So at one point there was a fund with two amazing mega-fund-returner deals, one of which turned into a mega-fund-returner deal, one of which didn't, right? And just the one who made the change did.
Well, to be clear there, they did get out of WeWork and have mega-fund-returning...
Oh, look. Amazing. Amazing. My point, but I don't think the WeWork return was nearly as compelling as the Uber return, Harry, because WeWork didn't... I mean, you know, it didn't... No, Harry. No, it didn't get public. It went bust. It SPACed. It went bust. It was not a returner. My point is they had two home-run winners, two huge burn companies, two potential fund returners, two really charismatic CEOs. One of them stayed the course. It didn't work out. One of them was replaced controversially. It did work out. It's an 80, 90 billion dollar company today. The fun fact is fast forward, and you even saw it in the tweets around the Atoms round, and Andre and Horowitz have backed both CEOs. They backed Adam at WeWork because they're like, 'We think you can do it again.' And they've just backed, obviously, Travis. And if you look at the tweets at the time, there was a very direct tweet last week, basically, 'We should have done this deal in 2010 or '11,' which is tantamount to saying, 'Yeah,' and everyone can read the subtext, 'We deeply regret taking money from someone else who fired us, even though it turned out to be an $80 billion outcome.' Right? So there's a clear dynamic there kind of in the venture backstory there. And what I really admire about what everyone involved is the willingness to bear grudges across a decade. It's quite impressive. So yeah, it will be fun to see how that shapes out. But yeah, Andre has chosen, has backed both CEOs from that famous Benchmark fund.
By the way, who... it's very minor... who knows, Claude says Benchmark took out 315 from WeWork.
Yeah, I was literally just about to say the most valuable lesson I've learned as a VC is to admit when you're grossly wrong, which to me happens daily. But exactly, Uber made Benchmark roughly 640x versus around 25x with WeWork.
They got 25x off. Good for them. I was wrong. I'll say it.
