Is AI a Bubble? Gavin Baker on Data Centers, GPUs, and the AI Economy
A16z's David George interviews Gavin Baker, Managing Partner and CIO of Atreides Management, on whether the current AI investment cycle constitutes a bubble. Baker argues it does not, contrasting today's build-out with the 2000 telecom bubble. While 97% of fiber laid at that peak was "dark," he notes there are no "dark GPUs" today and that the largest spenders have seen roughly a 10-point increase in return on invested capital since ramping up AI capex. He also highlights that the hyperscalers collectively generate about $300 billion in annual free cash flow with $500 billion in cash, providing ample buffer against the $40–50 billion cost of lighting up each gigawatt of capacity. On competitive dynamics, Baker frames the chip race as primarily a contest between Nvidia and Google's TPU, with Broadcom and AMD functioning as an effective partnership serving Meta and others. He predicts that several high-profile ASIC programs will be canceled over the next three years, especially if Google begins selling TPUs externally. He also argues that round-tripping deals between Nvidia and OpenAI are rational competitive responses rather than bubble signals, and dismisses concerns about GPT-5 ending scaling laws, noting it was a smaller, more economical model rather than a frontier push. On software and economics, Baker contends that AI-native businesses will have structurally lower gross margins than traditional SaaS, similar to how cloud margins sit below on-premise software, and urges incumbents to embrace margin compression as a sign of real adoption. He sees outcome-based pricing becoming more common, particularly in customer service, and views the human form factor debate in robotics as effectively resolved in favor of humanoids. Consumer AI, he suggests, will reshape search and commerce through personalized assistants and affiliate-driven flows rather than traditional ad-led models.
Original sources
Podcast The a16z Show
Episode:Is AI a Bubble? | Gavin Baker on Data Centers, GPUs, and the AI Economy ↗
AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.