Two contracts, built with the compute-pricing firm Silicon Data, would let buyers take positions on what Nvidia's H100 and the newer Blackwell B200 cost to rent, priced off indexes that follow what GPUs rent for by the hour — giving investors a way to trade the cost of computing capacity, and AI developers and data-center operators a way to hedge it.

CME Group is preparing to list the first futures contracts pegged to what it costs to operate the chips that run artificial intelligence, a step that turns computing power into something that can be bought and sold as an asset. The two contracts are due to launch on October 5, subject to sign-off from regulators, under a tie-up with the compute-pricing firm Silicon Data.

The contracts let buyers and sellers take positions on what it costs to rent Nvidia's H100 and the newer Blackwell B200 graphics processors, priced off Silicon Data indexes that follow what GPUs rent for by the hour. Each contract represents a month's rental of a single H100.

Carmen Li, who heads Silicon Data, said in a statement that for years two companies buying the same GPU capacity could wind up paying wildly different amounts, with neither able to work out which one had got the better deal — and that the contracts would hand them a standard to measure against. Compute futures, she said, would give the market its first openly quoted, tradable price for the resource behind every AI system.

Quoted H100 prices diverge widely. Nvidia has cited a one-year rental rate for the chip that climbed from roughly $1.70 an hour in October 2025 to $2.35 an hour in March 2026. Silicon Data's own figures, separately, put the median cost of renting an H100 from a major cloud provider at about $9.34 an hour during the second half of 2024, down to around $6.26 an hour twelve months later. The two gauge different arrangements over different stretches of time and do not form a single price line.

The debut arrives while Wall Street hunts for fresh routes to bankroll the construction of AI infrastructure and to take a stake in it. One is Nvidia's push with big asset managers, which might steer as much as $500 billion into it — a separate effort that so far rests on memorandums of understanding rather than signed contracts, with each project still requiring a final agreement.

Contracts on compute would stack a further tier onto that financial machinery, and a distinct one: exposure to what computing capacity costs, rather than to the data centers themselves, the chips inside them or the companies doing the building.

Oil and electricity have long traded on that logic — a price companies and investors can buy, sell and shield themselves against moves in. What would be new is the thing being priced: the capacity to run AI.

Cover image: “Chicago Board of Trade Building, Chicago, Illinois (11004312754)” by Ken Lund from Reno, Nevada, USA, Wikimedia Commons, CC BY-SA 2.0, resized, re-encoded.