U.S. takes first step toward oversight of the AI compute market

The U.S. derivatives regulator CFTC on Wednesday opened a public consultation on financial contracts tied to AI computing power. These compute derivatives could one day help companies hedge against rising or falling GPU capacity prices. There’s no new rule yet, and the regulator has not approved any trading product with this move.

The Commodity Futures Trading Commission is asking market participants for information on the nascent trade in compute. The consultation covers the size and liquidity of the spot market, reliable pricing data, potential manipulation, surveillance, customer protection, and perpetual futures.

Responses can be submitted for sixty days after the request appears in the U.S. Federal Register. A fixed calendar date will only be known after that publication.

The CFTC frames this as a first step to determine what rules are needed. That distinction matters. The consultation does not create a new legal asset class. If listed, compute contracts fall under the existing framework for U.S. derivatives exchanges; the question is how to apply that framework to this new underlying market.

Compute refers to available processing capacity, such as GPU servers in a data center. Developers rent that capacity to train AI models or generate large volumes of outputs. Prices can shift due to chip shortages, electricity costs, data center availability, and demand swings.

A derivative derives its value from such an underlying price. In theory, an AI company could lock in a future price for compute. If rental rates spike later, gains on the contract could offset part of the higher costs. A cloud provider, by contrast, might hedge against falling prices.

This mirrors how airlines hedge fuel or how farmers use futures for crops. Compute, however, is less standardized. An hour on one GPU isn’t automatically equivalent to an hour on another chip type, in a different data center, with different network or software terms.

This didn’t come out of the blue. CME Group announced on May 12 that it aimed to launch a compute futures market with Silicon Data. On August 11, the exchange set a planned start date of October 5 and unveiled H100 and B200 contracts, all subject to regulatory review. That plan isn’t the new development today.

What’s new is that on August 19 the CFTC formally opened a request for information to assess the market and its risks. The official CFTC page was public by 18:37 CET; Reuters published its story at 19:39.

An exchange must show that any settlement index is reliable and hard to manipulate; the CFTC reviews that under its existing approval or self-certification process. A contract keyed to the rental price of a single specific GPU on one platform is easier to distort than a broad index. It must also be clear what happens if capacity gets scarce or a data center goes down.

A compute future is not the same as Nvidia stock. Its value doesn’t track a chipmaker’s profits or share price. Nor is it purely an electricity futures contract, even though power prices do affect data center costs.

The intended underlying is access to compute. That makes the market appealing to AI firms, cloud providers, and data center operators—but complex for investors. Contract design, counterparty risk, and potential illiquidity can drive large losses.

Scarcity isn’t hypothetical. In July 2025, OpenAI said it intended to use more than a million GPUs that year. That older example shows the scale of demand, but it isn’t a current capacity reading for 2026. When supply and demand whipsaw, markets need transparent pricing and ways to hedge risk.

The consultation is American, but many European developers buy or rent capacity on global clouds settled in U.S. dollars. A liquid U.S. market could therefore influence their contract prices. That doesn’t mean Dutch consumers will soon buy GPU futures in a regular investing app.

Next come market feedback, potential rule designs, and reviews of specific contracts. Wednesday’s move chiefly spotlights compute as a new economic commodity. Trading it is not a done deal.