Invest Like the Best with Patrick O'Shaughnessy
· Gavin Baker, Patrick O'Shaughnessy
Cerebras has an overwhelming ratio of on-chip compute and memory, relative to shoreline I/O. Well, they're really smart people. They did something really hard. They're trying to see if they can put an optical wafer right on top of that, and then that solves that problem. I'm sure they're looking at hybrid bonding of DRAM to get around the — much-discussed on X — alleged limitations that are not true. A Cerebras machine can theoretically run any size model. There are sizes of models where they're much better than other sizes. So Cerebras — what I think is interesting is they did something different that's hard to do, really hard to do — wafer-scale computing. I do think there's a role for these. I would just encourage them all to make a different trade-off, try and do something hard. Everybody's going to get funded after the Cerebras IPO. It's not going to be a problem. But it took Cerebras three generations of chips to get it right. Andrew Feldman, the CEO — you can just see how hard it was, what he did and that whole team did, to get where they are today. They need to have the grit to do that, the resilience. This first chip is a failure. It happens. Can you come back and make a second chip? But one last thing on this topic. This is going to be amazing for the useful lives of GPUs and may single-handedly save private credit. [00:00:00] S0: Say more about that. What do you mean by private credit?
· Gavin Baker
[00:00:00] S1: Well, private credit is in pain from these SaaS loans, and however much they're marked down, they probably need to be marked down more. If the public companies are struggling to adapt, how's a debt-laden company going to adapt and invest in what is a very different margin structure business? There's a lot of private credit in GPUs too, and they were underwriting that to, I think, 3 or 4 years. The disaggregation of inference means that these GPUs are going to have 10 or 15-year lives. The AI skeptics are like, "Oh, these companies are all cooking their books. The useful life of GPUs is only a year or two. The useful life of CPUs is only 4 years because of rapid technological change." No. What rapid technological change has done — with the disaggregation of prefill and inference — is mean that you can put a Cerebras system or Groq LPUs that Nvidia acquired, effectively in front of a Hopper or even an Ampere, use that Hopper and Ampere for prefill, and extend the useful life of that GPU until it melts. They do melt, so they have a time, but maybe you don't have to run them as fast. This is going to be really good for the whole private credit industry. It's going to help finance the AI buildout, because if you can start to finance GPUs at more like 5% or 6% instead of — I think CoreWeave's lowest financing was low 7s — that actually mathematically changes the cost to finance this buildout. We've had this technological innovation that's going to lower the cost of financing, extend the useful life of compute on Earth. One last thing that's interesting about that — my friend Jaimin from Coatue just did a podcast, and Coatue had a deck where they talked about, "Hey, sellers of shortage are doing so much better than the buyers of shortage" — buyers of shortage being the hyperscalers. But if you own a giant installed base of what is currently in shortage, that's also a very good place to be. We're hearing CPUs are way more important than they were in an agentic world. They do all these things around orchestration, tool calls. The biggest CPU fleets in the world sit at the hyperscalers. Some of these hyperscalers may catch up a little bit to the sellers of shortage.
· Patrick O'Shaughnessy
[00:00:00] S0: I want to talk about this idea of different and hard, applied outside of the infrastructure piece of this. Now you're starting to interact with new founders, existing CEOs and founders that have to adjust to this new world. What are you seeing from the most AI-native founders that aren't building chips or infrastructure or models, but just people using this technology to build other stuff? How do they feel the most different to you, if you've observed differences?
· Gavin Baker
[00:00:00] S1: This isn't just for chip design. To me, it's always been a fundamental question for venture. There are different ideas that are obvious to everyone on planet Earth as soon as they hear it. If that's where you are in venture — if it's not hard to do, if it becomes obvious to the world before you have built scale — scale is the ultimate advantage — you're in trouble. The great thing Amazon had was, I think it was obvious to a lot of people, but it wasn't obvious to the retail CEOs. Amazon, they were very smart. Any e-commerce company that VCs invested in, they would destroy. They'd be like, "Oh, that's so cute. We're going to take our margins in that to negative ten thousand percent." That's like the guys at Wayfair, they did something hard. Amazon tried to kill them and they failed. Those were tough, operationally really competent CEOs. For me in venture, I always look — is this going to be obvious to the world before this company could build scale? Or is this both not obvious, different, and really hard to do? A lot of founders are really struggling with this in AI. People are becoming worried today that in Jensen's five-layer cake of AI, the profits are accruing to energy, accruing to data centers, accruing to chips, accruing to models — not really accruing to the applications. Cursor and Cognition got to scale. They focused on coding. 18 months ago, OpenAI was doing everything under the sun. The people focusing on coding were Cursor, Cognition, Anthropic, and it was really right to focus on code. Amjad Masad, the founder of Replit, tweeted something I thought was so smart — something like, "Bitter lesson-adjacent is the fact that coding might be the shortest path to ASI and useful AI. Because if you're really good at coding, you can write yourself code to do anything. " So it was really smart of those companies to focus intensely on coding. They all probably got to a scale where they have a place. I think Cognition is doing something really, really different. But a lot of founders are really struggling. They're trying to get confidence that in "niche-ier" areas…
· Patrick O'Shaughnessy, Gavin Baker
[00:00:00] S0: They won't get steamrolled. [00:00:00] S1: They can get to them and get a data moat before the model companies get to that niche, or that it's a small enough niche that the model companies won't do it themselves, but it can still produce a venture outcome. [00:00:00] S0: Is this related to what you would call the token path? I know you've used that phrase with me before. [00:00:00] S1: Yeah, I think it comes from a guy at Altimeter, Jamin Ball. He just said, "If you're a software company or an AI company of any kind, you have to be in the token path." So Databricks is in the token path. Comparable companies are in the token path. If you're not in the token path and you're not in some really niche thing, life may be hard. Even for these vertical niches, if you talk to the people at the model companies, they're skeptical of some of these, because all of the data that's being generated in these niches comes from humans. You're betting that you're able to use that proprietary data in this narrow vertical to train a model that's lower cost than the frontier labs can ever get to. Maybe that's a good bet, but you have to be very, very careful. On the other hand, if the returns to these frontier tokens relative to other tokens come down, there's going to be an explosion in value creation at the application layer. Another really important point is that I have a belief that whenever he wants, Jensen can probably get pretty close to the frontier. [00:00:00] S0: With his own model?
· Gavin Baker
[00:00:00] S1: With his own model. I don't think he wants to do that, but that is what OpenAI and Anthropic are trying to do to him, unsuccessfully. He's a very logical thinker. This is the logical countermove. You will see that open-source frontier — which today consists of Chinese models with stolen American tokens. Somebody told me that DeepSeek — the original one was only 150,000 reasoning traces. There are many ways to launder this if you're a Chinese company. You can hit all these different APIs, you can make it hard. The American labs are working really hard on anti-distillation technology. Chinese open source — they're doing really impressive things in a very resource-constrained way, but there's a lot of distillation. This is why, in addition to there not being enough compute to serve Mythos, they did not want it to be distilled. They wanted to use Mythos, distill it themselves, use it to RL their next model, whatever it is. And then what eventually OpenAI and anyone on the frontier will do — there's going to be some very interesting game theory because it's a new kind of prisoner's dilemma. The old prisoner's dilemma was around, "Hey, you're in a prisoner's dilemma where you have to spend." The new prisoner's dilemma is going to be, if you are at the frontier, do you release that model via API or not? If everyone at the frontier agrees not to do that, then Chinese open source — if one person defects, they're going to have the best model. They're going to have a lot of revenue and cash flow. And then of course, resources equal intelligence, so they'll start to pull ahead, and then that will lead to everybody else releasing it. So it's a new game theory. It's kind of the same game theory that you have with Taiwan Semi, Samsung, and Intel. The reality is, if a company like Nvidia or AMD were to ever really, really use one of these other foundries, that foundry would get better really quickly. So Jensen is going to keep open source a certain timeframe behind the frontier. That's going to be a very interesting thing to watch. And then by the way, open source gets monetized. There's this misnomer that open source is free. Open-source tokens cost energy to produce. You need to make a margin on GPUs, and the open-source model companies almost always get a revenue share.