bak

The AI revenue numbers look incredible right now. Chamath Palihapitiya's warning is that they're sitting on an ROI that, by his own math, barely exists. "The problem with enterprise revenue is at some point the person that's spending it has to see an ROI. I asked Fable 5, Anthropic's new model. I first asked it, what is the lift of the S&P 500 earnings per share growth since 2024 from AI? And they answered, oh, it's 50%. So then I looked through it and I said, well, no, you're including the money that Nvidia makes from selling chips to Amazon." "So I asked a different question, which is, then what was the EPS growth of the S&P 493? And the answer was 9%. And I said, okay, well that's different. And I said, unpack that. And the overwhelming majority of that was from pricing power sitting on top of inflation. And then the other 3% was from buybacks." "And so the answer, as far as all publicly available data, was that the actual ROI was somewhere between zero and 2%." "Enterprise is probably a little bit more brittle because there are fewer buyers and they're more demanding. Consumer on the other hand all of a sudden becomes an incredible safe harbor because you have tens of millions of buyers... it inoculates you from the vicissitudes of an ROI discussion." "At some point, you'd have to be an idiot not to ask, well, who is paying you this? And can they sustain paying it to you? ...you're spending a million dollars a year on tokens and that million dollar a year is doubling and tripling and quadrupling. At some point you're going to have to show an ROI that's above the risk-free rate of return, otherwise you're going to have some angry investors on your hands." Full Episode w/@chamath : @theallinpod

Jensen Huang refuses to anthropomorphize AI agents, and he explains why. "It's electrons, not atoms. It's not biological, has no consciousness. It's not awake." His analogy is deliberately domestic: "It's a tool. It's like my vacuum cleaner that's roaming around the house, cleaning up the house, doing something that I used to do." Then the dishwasher, and where he lands with it is on how we get used to things: wonder first, then routine. "A hundred years ago, when the first dishwasher came along, and now it's doing dishes by itself, it must have been magical to watch it, and we call it a dishwasher, which is a little bit like a human." And on the habit of humanizing it directly: "I think right now we tend to imbue too much human properties to it. It's nothing close to that. It's software. It's computers." His objection runs deeper than practicality. To him the human framing is simply false, not a matter of what's easier to work with: "it's nothing close to that." A human-sounding name tells you nothing about what the thing is. The dishwasher got a name that's almost human too, and watching it was still a kind of wonder, and wonder doesn't argue with how a thing works, it just settles into routine. (That last step is drawn from his words, not stated by him: Huang himself lands the anecdote on "we'll get used to it.") Only then does the operational layer arrive, the reason the distinction matters to him at all: "We know exactly how it's working, because obviously we created the harnesses around it. We obviously know how it works because it's getting better all the time. If we don't understand how something works, how do we make it better every time? And if we don't understand how something works, how do we improve it? How do we fix it?" For Huang, understanding and improvability underwrite each other. "Created the harnesses" gives you "we understand it, so we improve it," and "it's getting better all the time" runs the other way, "it improves, so we understand it." Both arrows sit right there in the quote. Which is probably why he won't move the line, the line between what you can improve and what you can only name: "I think that we ought to keep it there." Full conversation is on @LangChain 's YouTube channel

David Sacks says China's open-source push was always a tactic. You go open until you catch the frontier, then you close. Exactly what Sam Altman did at OpenAI. "Last week I explained why it would be harmful to the US to ban open models. So if you're China and you want to harm the US, maybe you would want to. It does kind of make sense, because our companies are benefiting a lot from all this R&D that they're doing." "For example, the number one model in China, as I understand it, is ByteDance's model, which is already closed. That's kind of like their ChatGPT equivalent and it's always been closed. Then you've got Alibaba's Qwen, which was open and now I think is going closed, and Z.ai, which has GLM 5.2, catching up to what was then commercially available as the American frontier at certain tasks. They're going closed too, after having been open." "So this is, I think, the tactic. You stay open until you catch the frontier or you get close to it, and then there's a really compelling incentive to go close, because you want to capture all the value for yourself. Which by the way is exactly what Sam Altman did famously at OpenAI." Jason: "Not only did they go from a nonprofit to a for-profit, they went from open models to closed models. So it's exactly paralleling what Sam realized 3 years ago." Reuters reported this month that China's Ministry of Commerce is already discussing exactly this — restricting foreign access to top models from Alibaba, ByteDance and Z.ai, the lab behind GLM 5.2. Sacks, on why open source works when you're behind: "If you want to catch up, you go open. Because you're not going to make any meaningful revenue on closed anyway, because you're not close enough to the frontier. So why would anyone buy your product? But if you go open, you get the developer community on your side." Jason: "And you get utilization, more people use it, which in AI gives you reinforcement learning."

Chamath Palihapitiya says his company's token costs are doubling every 45 days, while productivity gains are only around 5%. "I sat down with my CTO today and I said, how are we doing on token spend? And he said the most incredible thing. He said, right now our token costs are doubling every 45 days. And I was like, ugh. And he said, yeah. And I said, well, what is the downstream productivity? And he said, maybe 5% max." "And I said, okay, so my costs are doubling every 45 days, my upside is essentially flat. And he said, basically. And I said, well, explain why that is. And he said, honestly, what we're finding out is that you need to use a lot more tokens to get to this next iteration of improvement, because we've effectively already asymptoted." "And I said, so what should we do? And he said, honestly, we have to figure this out. So we're going to take a step back and try to figure out what to do. I don't know how many other companies will actually go through this reckoning now, but the point is everybody in the next three or four years will for sure go through it." "So I suspect that if you can get out now, you should get out now, before all of that starts to seep into the water table, because I think that's probably what allows you to get out at a huge price and raise a huge amount of money." @theallinpod

David Sacks says China's open-source push was always a tactic. You go open until you catch the frontier, then you close. Exactly what Sam Altman did at OpenAI. "Last week I explained why it would be harmful to the US to ban open models. So if you're China and you want to harm the US, maybe you would want to. It does kind of make sense, because our companies are benefiting a lot from all this R&D that they're doing." "For example, the number one model in China, as I understand it, is ByteDance's model, which is already closed. That's kind of like their ChatGPT equivalent and it's always been closed. Then you've got Alibaba's Qwen, which was open and now I think is going closed, and Z.ai, which has GLM 5.2, catching up to what was then commercially available as the American frontier at certain tasks. They're going closed too, after having been open." "So this is, I think, the tactic. You stay open until you catch the frontier or you get close to it, and then there's a really compelling incentive to go close, because you want to capture all the value for yourself. Which by the way is exactly what Sam Altman did famously at OpenAI." Jason: "Not only did they go from a nonprofit to a for-profit, they went from open models to closed models. So it's exactly paralleling what Sam realized 3 years ago." Reuters reported this month that China's Ministry of Commerce is already discussing exactly this — restricting foreign access to top models from Alibaba, ByteDance and Z.ai, the lab behind GLM 5.2. Sacks, on why open source works when you're behind: "If you want to catch up, you go open. Because you're not going to make any meaningful revenue on closed anyway, because you're not close enough to the frontier. So why would anyone buy your product? But if you go open, you get the developer community on your side." Jason: "And you get utilization, more people use it, which in AI gives you reinforcement learning."

🚨 POD UP! Fifth Bestie Brad Gerstner is BACK! @altcap fills in for @friedberg: -- OpenAI vs Anthropic IPOs -- The Open Source Decision in July 2026 -- Meta's New Model, Zuck's Price War -- China to Crack Down on Open Source? -- Trump Accounts Launch (0:00) Bestie intros: Brad Gerstner fills in for Friedberg! (2:58) OpenAI vs Anthropic IPOs: Why it matters who goes first, what they learned from SpaceX, unlimited TAM of intelligence (27:39) The open source decision, Meta's new model, Zuck's price war, AI duopoly (54:29) CCP considering putting export controls on Chinese models, is open source ending in China? (1:03:09) Trump Accounts launch, getting young Americans bought back into capitalism

Thoroughly enjoyed doing this podcast with @AmirF15336 who is an impressive young man. Recorded in late February of this year. Time flies!

NVIDIA CEO Jensen Huang: "The more AI we use, somehow the more people we have to hire." It's a claim that cuts against the mainstream view: the more AI his company uses, the more people it has to hire. The work shifted rather than disappeared. "Coding is like typing, and so they're gonna do less typing." The engineers who used to write code now build agents: "now we have a lot of software engineers, building agents. They used to code software, but now they're building agents." The task moves up a level, from writing the instructions to designing the system that follows them. As Huang puts it himself: "They're gonna be more systems engineers." And that new level turns out to be bigger than the old one. Building an agent isn't a single task. It's the model, the tools, the memory, the guardrails, and the tests that prove the whole thing works. "They're creating evals. They're creating benchmarks. They're creating guardrails." Every capability you hand to a machine has to be specified, wired together, and checked by a person before anyone trusts it in production. Huang thinks his engineers prefer it this way. "Every one of my software engineers prefer to be building agents than to be writing Python code." The routine gets automated. The part that needs judgment expands to fill the space it leaves behind. That piece of the logic comes not from Huang but from his interviewer, LangChain co-founder Harrison Chase (@hwchase17). In the exchange about evals, he notes that judging an agent's quality is best done by the people who already know the domain from the inside: "quantifying whether it's good or not is oftentimes best done by subject matter experts who already live inside the enterprise and can easily give feedback." That work can't be handed to the same automation it's meant to judge; it has to come from someone inside the company who knows what "good" looks like. Huang agrees, saying "That's right," and takes it further: every professional, from a doctor to an engineer, is now building themselves an agent, handing it the routine, and moving up to the level where judgment matters. Nvidia is its own proof. It closed its most recent fiscal year with about 42,000 employees, up from 36,000 the year before, a roughly 17 percent jump in a single year. Automation removes tasks, not work. Specifying those tasks, building the systems, and verifying the output are still human, and there's more of it than before. Cheap, fast intelligence also widens the horizon of what's worth attempting at all, and on that wider horizon, as Chase frames it, the question isn't "what can we automate from before" but "what couldn't we do before that we can do now." Both movements point the same way: toward people. Which is why the company deploying AI most aggressively is also hiring at one of the fastest rates in its industry. Full conversation is on @LangChain's YouTube channel

Full episode now on X: My conversation with @GavinSBaker, founder and CIO of Atreides Management and one of the sharpest technology investors of the last 25 years, is out now. I sat down with Gavin to talk about how a kid from Texas who planned to live out of the back of a pickup truck, climb full-time, and write the next great American novel instead became one of the most respected deep-tech investors of his generation, and why he believes everything important he learned came from getting humbled early. Gavin spent 18 years at Fidelity, rising to manage the $17 billion OTC fund. He met Jensen Huang at 23 when NVIDIA was a tiny stock, showed up to a near-empty 6 PM meeting with @elonmusk when Tesla was a $1.5B company, and has read essentially every NVIDIA and Tesla transcript since. In 2019 he left to found Atreides, where he invests across public and private markets with the same first-principles lens — watts, wafers, and the physics of compute. What makes Gavin different is that he treats investing as a game of self-knowledge as much as analysis. We spent most of the conversation not on stock picks, but on temperament, failure, and what actually separates great investors from everyone else. We spoke about: - A free-range Texas childhood with an unlimited book budget and asking his parents if dinner was a "one-book, two-book, or three-book dinner" - Writing a bearish semiconductor note at 23, weeks before the dot-com crash, off a simple inventory analysis - Blowing up on large-cap pharma at 25 — going from department star to the very bottom - "You either panic early or double down late, and almost no one does both" — and why knowing which one you are is everything - Why being early is the same thing as being wrong - Reading two books during every drawdown, and finding a number you can improve when everything else is going wrong … and much more. This was filmed in February 2026

A rational conversation on where AI is actually going with @benedictevans For 20+ years, Benedict has been one of the clearest, most reliable thinkers on where technology is heading, and how it'll impact our lives. He was @a16z's resident "thinker" for 5+ years, and has spent the last six as an independent analyst tracking the most important tech trends. As you’d expect, he’s spending all of his time on AI. In his words, "AI is eating the world." We discuss: 🔸 Where value will actually accrue in the AI stack 🔸 Why AI labs are suddenly buying consulting firms 🔸 The rise in anti-AI sentiment, and where it leads 🔸 Why distribution is becoming the ultimate moat 🔸 Why the right question about your job isn’t “What percent can AI do?” but “Is this a task or a job?” 🔸 Why things will probably be okay Listen now 👇 youtu.be/BD3vLtWhT5A