In this episode, Patrick O'Shaughnessy and Gavin Baker discuss the recent sharp sell-off in AI stocks, driven by market jitters over credit conditions, open-source model competition, and geopolitical news. Despite the sell-off, Baker argues that fundamental demand for AI compute is accelerating, with GPU prices rising and operating cash flows at hyperscalers improving, suggesting the market may be overly pessimistic.
Summarized by Podsumo
Gavin Baker notes that despite AI stocks dropping 40-60% in July, every quantitative metric—GPU pricing, token growth, and DRAM spot prices—is accelerating, with no signs of deceleration.
The market's concern over credit financing for AI build-out is countered by the argument that repricing of contracted compute will boost operating cash flows, potentially eliminating the need for debt.
Open-source models like those from Fireworks and together are shifting tokens from high-margin frontier models to lower-margin ones, which actually increases total compute demand and benefits infrastructure players like NVIDIA.
NVIDIA is trading at its lowest forward PE in 10 years, reflecting market skepticism that it is over-earning, but Baker highlights NVIDIA's strategic moves like credit wrappers and equity stakes to bridge financing gaps.
Regulatory risks, particularly data center moratoriums in New York, are flagged as the biggest threat to AI growth, but the industry is failing to tell a positive story about job creation and community benefits.
"I haven't been able to find one negative quantitative metric about AI. In fact, every metric is accelerating."
"If you believe the quantitative demand signals from Silicon Valley private companies, the installed base of compute will reprice higher, operating cash flows will accelerate, and you can fund most of the build-out internally."
"The only time the semis have been cheaper than now were Liberation Day and DeepSeek—and those were B bottoms."