Why model-layer margins should be negatively correlated with CoreWeave, Oracle and Nvidia
GavinSBaker · x · 2026-09-21
Investor Gavin S. Baker pushes back on the "cars boosted steel and oil" analogy for AI compute: token production inputs aren't all commodities.
His core argument: CoreWeave, Oracle and Nvidia valuations should logically be negatively correlated with model-layer margins. Token production cost is effectively those companies' revenue; the higher the margin OpenAI/Anthropic add on top, the fewer tokens get produced and the less revenue flows to infra. He adds that an OpenAI/Anthropic monopsony would be bad for everyone in AI infrastructure.
Related event: Investor and Gary Marcus Clash Over LLM Economics and Neoclouds(2 posts)→
More from Venture
- Anyone's agents actually making money? A dev's reality check on x402 agent payments — thranduilsson · 2026-09-21
- Compute lending panic overblown: $10B+ in data center deals priced in two months — Jsevillamol · 2026-09-21
- $4.2M ARR AI acquisition target turned out to be one GPT-4o call and 600 lines of glue code — alex_verem · 2026-09-21
- Tesla Robotaxis now cost slightly more than Uber in Austin, but scale could flip pricing — JOBhakdi · 2026-09-21
- Commentary: AI buildout may be disinflationary, but the Fed uses it as cover for Main Street neglect — McDonaghMatthew · 2026-09-21
- 'Nobody paid me a single dollar' became a book title within a week — and someone bought it — itsOmSarraf_ · 2026-09-21