CME Group is turning raw computing power into a tradable commodity. The world’s largest derivatives exchange announced a partnership with Silicon Data to launch two Compute futures contracts that will track the hourly rental costs of Nvidia’s H100 and Blackwell B200 GPUs, giving AI companies a way to hedge against the wild price swings in the GPU rental market.
The contracts are set to debut on October 5, 2026, pending approval from the Commodity Futures Trading Commission. If greenlit, they would represent the first publicly tradable reference prices for AI compute costs.
What these contracts actually do
Each contract will represent one month of GPU rental and trade on the New York Mercantile Exchange. The underlying benchmarks come from Silicon Data’s indexes, which track what it actually costs per hour to rent an H100 or a B200 from cloud and data-center providers.
Two separate contracts cover the two most commercially significant Nvidia architectures. The H100 has been the workhorse GPU for AI training since its release, while the B200 represents Nvidia’s newer Blackwell generation, which has been drawing intense demand from hyperscalers racing to build out inference infrastructure.
