Nvidia smashes earnings (again), Google's Woke AI disaster, Groq's LPU breakthrough & more

The episode opens by mocking its own intro. Jason introduces Chamath Palihapitiya (Social Capital), David Friedberg (the "Sultan of Science"), and David Sacks (Craft Ventures, dialing in from an undisclosed hotel room); Chamath calls the intro "low energy and dragged out" while Sacks explains he is sick. The banter detours into a Superman II gag about General Zod before Jason resets, unsure whether this is episode 167 or 168 and concluding "Who cares?"

The first real topic is Nvidia's third straight earnings blowout. The hosts cite a 15% move on Thursday that added roughly $247-250 billion of market cap in a single day — framed as a record, topping Meta's earlier jump. Quarterly revenue is about $22.1 billion, up sharply year over year, gross margin near 76%, with guidance around $24 billion next quarter. The driver is a data-center ramp that began only about four quarters earlier as hyperscalers build generative-AI infrastructure.

Friedberg and Sacks explain the mechanism: GPUs were built for the vector math behind 3D graphics, the same math underlying AI, so the LLM explosion made GPUs the chip cloud providers need — Nvidia was, in Sacks's words, in the perfect place at the perfect time. A $2.7 billion buyback under a $25 billion plan supports the stock alongside expanding margins.

Chamath frames the core tension: sustained over-earning invites competitors, and only a monopoly (he cites Google, now chipped at by Perplexity) lets you over-earn for decades. Nvidia is over-earning massively, so the question is who competes those profits away in training versus inference. He invokes the Bezos line.

Your margin is my opportunity.

Friedberg asks whether infrastructure spend will exceed value realized at the application layer. His analogy is the early internet: buyers first paid for pricey Oracle servers before learning that cheaper, higher-failure-rate hardware plus smart software was better for large-scale distributed data centers, so Oracle never took that market; the same undercutting later hit high-end networking vendors.