AI Radar

Noteworthy AI conversations and writing, summarized from publisher notes.

Gavin Baker on Nvidia vs Google, Scaling Laws, and the Economics of AI

· Invest Like the Best

Gavin Baker (Atreides Management) joins Patrick O'Shaughnessy for a sweeping discussion of the AI infrastructure landscape. He argues that Gemini 3 confirmed pre-training scaling laws remain intact—an empirical phenomenon humanity can measure precisely but not yet theoretically explain. He credits the post-training breakthroughs (reinforcement learning with verified rewards and test-time compute) with bridging an 18-month gap during the delayed Blackwell rollout, calling reasoning the development that "saved AI" when no new chips were available. Baker frames the central strategic contest as Nvidia versus Google/TPU, arguing Google has been the lowest-cost token producer and has used that position to "suck the economic oxygen" from competitors. He expects that dynamic to shift once GB300 and Rubin arrive, since Blackwell-trained models should be materially better and will reset the cost curve. He sizes up the four leading labs (OpenAI, Gemini, Anthropic, xAI), predicting xAI will ship the first Blackwell-trained model thanks to Elon Musk's data-center build speed, and noting Anthropic's cash efficiency and recent $5B Nvidia deal. He also explains why Meta, Microsoft, and Amazon stumbled at frontier model development despite massive resources, and why Blackwell widening the U.S. frontier gap may turn China's rare-earth leverage into a strategic miscalculation. The conversation ranges into data centers in space (cheaper solar, free cooling, faster inter-satellite links via laser), power as a natural governor that favors the most compute-efficient designs, and a warning that application SaaS firms are repeating brick-and-mortar retailers' e-commerce mistake by clinging to 80% gross margins in a world where AI natives run at 35–40%. Baker closes with an unexpected personal origin story—from aspiring ski bum and wildlife photographer to investor after stumbling into a DLJ internship that introduced him to the "greatest game of skill and chance."

AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.