📊 CME and Silicon Data launch compute futures - October 5.
AI capex spending in 2026 is projected to reach $765B (according to Goldman Sachs), surpassing investments in oil and gas ($681B) for the first time. By 2031, it is expected to nearly double to $1.6T. Morgan Stanley estimates AI diffusion across the economy at $40T in opportunities - all tied to compute as a fundamental resource, which currently lacks any hedging instruments.
https://chamath.substack.com/p/compute-financialization
CME addresses this through two contracts on NYMEX: Silicon Data H100 Rental Index Futures and B200 Rental Index Futures. Each tracks the hourly rental cost of the corresponding Nvidia chip - like an oil future, but instead of a barrel, it's a month of GPU rental. They cover three risks at once: rental prices fluctuate with demand, old hardware rapidly loses value with the release of new chip generations, and data center construction takes 2-3 years without the ability to lock in future capacity costs.
However, compute suffers from the same ailment that killed futures for onions and uranium - concentration and non-fungibility of the commodity. Nvidia currently holds almost all chip supply, although there are thousands of buyers, including 60+ neocloud providers with revenues of $25B+ by 2025. And two H100s from different data centers are not the same: Silicon Data ran the same load on 3500 GPUs across 11 clouds and found performance variability up to 34.5% on the same chip model. It seems the contracts will need standardized "grades" of GPUs, as has long been the case with oil and electricity.
#nvidia@rvnikita_blog #cme@rvnikita_blog #compute@rvnikita_blog #ai@rvnikita_blog