New @ThePeelPod with Navin Chaddha Navin's an 18x Midas Lister who runs Mayfield, a 56-year old firm betting $3B on AI. We get into what’s actually going on with $1B+ inception rounds, the math behind the $6T AI software market, why AI is 10x overcapitalized today, what he learned working with Satya Nadella in the 90's, and lessons from founding the last company to IPO before the Dot Com Crash. Full episode here + links below 0:00 Lumilens: Zero to $3B revenue in 14 months 0:50 Connecting GPU's is AI's next bottleneck 5:07 Investing $3B in AI and semiconductors 9:39 Where a $1B round actually gets spent 12:51 The six-layer AI stack 15:00 Why AI is overcapitalized by 10x 17:47 FOMO is for sheep 21:13 Real revenue vs vibe revenue 22:45 Backing vertical models 24:22 The best firms have one North Star 27:18 Are semiconductors still cyclical? 29:10 Why inference will dwarf training 31:19 What happens after every infra build-out 36:27 Real vs fake AI adoption 38:21 What a correction does to AI stocks 40:46 How FOMO pulls VC's into hot categories 44:13 What Navin looks for in founders 50:23 Everyone hating a category can be a buy signal 54:12 The cloud argument everyone got wrong 57:20 The $6T of white-collar work AI will take 1:01:45 How AI startups beat incumbents 1:06:54 Startups die of indigestion 1:11:49 Mayfield’s investing formula: people-first 1:19:18 What cricket taught Navin about building companies 1:23:17 Dropping out of Stanford to start VXtreme 1:29:22 Lessons from the last IPO before the Dot Com Crash 1:31:25 Joining Mayfield instead of starting a 4th company 1:33:29 Unfinished business (backing a $1T company) 1:35:54 Could you tell Satya would run Microsoft? 1:38:58 Investors he respects, founders he missed

I asked @alfromnexhealth why most future value in AI will accrue to the action layer: "First, it was the chip makers, like Nvidia. Second, it’s OpenAI and Anthropic. The companies that do the R&D. The research to build the actual LLMs. The third wave, which hasn’t truly come yet, but more and more of the ecosystem is realizing it, is the companies and the infrastructure that takes the LLMs and enables them to actually action stuff in the real world. This is the layer that sits between the physical world and the software world. Where these LLMs can then actually go into the physical world and action that data. A very basic example: if you’re a healthcare practice today, you can use the OpenAI API to build a chatbot product on your website. A patient can chat with it about their condition. And it’ll recommend the right provider, and a time slot in the office to book an appointment. That chatbot or LLM needs to connect to the physical world. "What day is the doctor in the office? What room in the office is available?". All in real time. Because the patient is talking to that chatbot in real time. To then surface, “Hey, here’s what’s available,” and take your information and write it back as well. Otherwise the patient will show up and it won’t be on their calendar. So if you just track the hype cycle: it's chip makers, the companies that build the LLMs, and then the action layer that sits between the LLMs and the physical world."

Claude is Walter Cronkite for the stock market. @mjmauboussin talks about how a breakdown in diversity leads to bubbles and crashes. Gavin thinks Claude might be causing that right now. ”Everyone I know in the public equity investment business, whether retail or institutional, everything immediately, every piece of news gets fed into Claude—Claude, Claude code, sometimes a Claude agent. Claude is probabilistic. There's probably not that much variation in the way it's interpreting this news. People talk about the fragmentation of media and how it used to be like Walter Cronkite was the only voice of truth, and now we don't have that anymore. It's like Claude is Walter Cronkite for the stock market, and everybody just believes whatever it says.”

"The state of payment collection in healthcare is insane today. Your doctor usually doesn't actually know who paid their bill. At most offices, employees are manually matching up payments in the PoS terminal to appointments in the system to guess who paid which amount."