Don Wilson on the GPU Market's Trillion Dollar Question
Source
omny.fm
Date
Why it matters
Compute futures and GPU market structure shape how inference capacity gets priced and hedged. Hearing a market maker's view helps you reason about future compute cost and availability.
Key takeaways · AI-distilled
Don Wilson argues GPU futures would let a neocloud deploy GPUs, sell a strip of futures against them, then sell on-demand compute or short contracts to less creditworthy buyers, lowering its cost of capital.
Wilson says GPU futures could list November 9 if the CFTC takes no action after its extended comment period, with CME contracts first reaching three years out; the Silicon Data index tracks neocloud on-demand prices.
Wilson says hourly H100 on-demand prices fell for months, then turned up around Thanksgiving when a stronger Anthropic model lifted inferenceRunning a trained model to get answers — the phase where AI is actually used, as opposed to trained.Full definition → demand; B200, B300 and H100 prices recently surged again as the Silicon Data tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → index fell.
Wilson says DRW is at the thought stage on token futures, which would let traders price the margin between renting GPUs and selling open-model tokens, a spread he calls the crunch spread after oil's crack spread.
Terms in this piece · Glossary
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
