The Commodity Futures Trading Commission requested public comment on August 19 about how U.S. exchanges should list and oversee derivatives tied to artificial intelligence computing capacity.
Summary
- The CFTC requested public comments on compute derivatives, market oversight, manipulation risks and customer safeguards.
- The deadline will fall 60 days after publication in the Federal Register, which remains pending.
- CME plans two October 5 futures launches tied to Nvidia H100 and B200 rental benchmarks.
- The agency is examining cash market liquidity, reference prices, manipulation controls and perpetual compute futures.
- Compute futures could help AI developers hedge rental costs, but no contract received approval yet.
The 19 page request, identified as RIN 3038-AF77, examines market liquidity, benchmark reliability, manipulation risks, customer protection and perpetual compute futures.
Comments will remain open for 60 days after the document appears in the Federal Register. As of August 20, the document had not been published there, meaning the CFTC had not established the final submission deadline.
The request does not propose a rule, approve a contract or authorize an exchange to begin trading compute derivatives.
CFTC treats computing capacity as an emerging commodity
Compute refers to the processing power used to train and operate AI models. Companies often obtain that capacity by renting access to graphics processing units through cloud providers or specialist data center operators.
The CFTC said the underlying commodity would typically be access to rented computing capacity, such as the hourly rental price of an Nvidia B200 processor. A contract could also reference another unit, including access to a specified quantity of AI inference tokens.
Chairman Michael S. Selig said the U.S. “cannot win the AI race without a robust derivatives market for compute.” He described the request as an initial step toward creating “clear rules of the road,” rather than a final regulatory position.
Compute futures could allow AI developers, cloud companies and data center operators to hedge changing rental costs. Traders could also use the contracts to speculate on future demand for AI infrastructure.
Private pricing creates manipulation questions
A central concern is whether the underlying compute market is liquid and transparent enough to support reliable futures settlement.
The CFTC said much of the economic value in compute markets appears to flow through private bilateral agreements. Prices can also vary across providers, regions, processor models and contract structures.
The agency asked whether a cash settled contract should use prices that regulators cannot fully observe, verify or monitor. It also requested information about transaction volumes, market concentration and the proportion of trades completed at publicly disclosed prices.
Compute providers could potentially influence both available capacity and benchmark inputs. The CFTC therefore wants feedback on surveillance arrangements, information sharing and safeguards against distorted settlement prices.
Perpetual compute futures remain under consideration
The request asks whether perpetual compute futures would offer risk management functions unavailable through contracts with fixed expiration dates.
Perpetual contracts have no standard expiry. They generally use recurring funding payments or another mechanism to keep their prices aligned with the referenced market. The CFTC asked whether this structure would introduce additional risks requiring new protections.
The inquiry follows the agency’s wider examination of perpetual derivatives. The CFTC has already allowed certain cryptocurrency perpetual contracts, although CME Group is challenging that policy in federal court.
The legal dispute does not directly concern compute. However, it could affect how exchanges and regulators classify perpetual contracts across new commodity markets.
CME targets two October compute futures launches
CME Group and Silicon Data plan to launch two cash settled compute futures contracts on October 5, subject to regulatory review, according to an August 11 announcement.
One contract would represent one month of Nvidia H100 rental costs. The other would track rental costs for Nvidia’s newer Blackwell B200 processor. Silicon Data would supply the pricing benchmarks.
As previously reported, CME targeted an October launch using daily GPU rental benchmarks. The planned date does not guarantee that regulatory review will finish by then.
The CFTC is also requesting input on retail disclosures, anti money laundering controls, position limits and the geopolitical sensitivity of advanced computing capacity. It specifically encouraged empirical and data based submissions.
Once the Federal Register publishes the request, participants can submit comments through Regulations.gov or directly to the CFTC. The agency may use the responses to inform future rules, guidance or its review of exchange filings.
