Gavin Baker on AI Market Jitters
Gavin Baker, portfolio manager at Atreides Management, argues that the sharp July sell-off in AI stocks is disconnected from underlying fundamentals, which he says are actually accelerating. From conversations across Silicon Valley, he reports that every quantitative metric he can find—GPU rental pricing, DRAM spot prices, token growth—is improving, with some companies paying 50–60% more for compute when contracts roll over. He attributes the market drawdown to a confluence of narrative shocks (Meta renting compute, open-source model releases like GLM 4.5 and Kimi K3, China's DUV news, and rising credit spreads) rather than deteriorating demand. Open-source models, in his view, are simply shifting margin dollars from frontier labs to the AI infrastructure layer, since a token still requires the same compute regardless of who serves it. Baker sees credit conditions—rising real yields, widening spreads, and NVIDIA CDS blowouts—as the only genuinely worrying signal, but argues the build-out can ultimately be funded by operating cash flow if hyperscalers monetize at rates anywhere near current Blackwell levels. He calculates that hyperscale operating cash flow could reach $2 trillion versus modeled $1.3–1.4 trillion, potentially removing $700 billion of credit demand. He also discusses NVIDIA's emerging "credit wrapper with revenue share" model for GPU buyers, the strategic importance of long-term supply agreements with memory companies, and the bullish implications of SpaceX's rumored 8-gigawatt compute build-out. His biggest concern is regulation rather than demand, and he calls for the industry to better communicate the local economic benefits of data centers.
Original sources
Podcast Invest Like the Best
Episode:Gavin Baker - AI Market Jitters - [Invest Like the Best, EP.485] ↗
AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.