Compute derivatives are financial contracts whose value references the price of access to computing capacity — in practice, the hourly rental rate of AI accelerators such as Nvidia's H100 and B200, though other underliers have been proposed, including a stated volume of large-language-model inference tokens. The class moved from proposal to imminent listing during 2026, and with it the question of whether compute exhibits the properties a regulated derivatives market assumes.
The proposed contracts
CME Group and Silicon Data, a GPU market-intelligence firm backed by the trading firm DRW, announced on May 12, 2026 that they would launch a compute futures market, and on August 11, 2026 set the launch for October 5, 2026, pending regulatory review. Two contracts were named: Silicon Data H100 Rental Index Futures and Silicon Data B200 Rental Index Futures, each tracking an index of hourly GPU rental costs published by Silicon Data, each representing a month's rent for one accelerator, cash-settled and USD-denominated, listed under the rules of NYMEX (Source: cmegroup.com).
The CFTC has extended its review of the planned contracts, which became public on September 22, 2026: according to a CFTC letter to the exchange, the agency is seeking comments from the broader industry, and the contracts will not be approved in time for the early-October launch CME had sought (Source: theinformation.com). Only the abstract of the reporting was retrieved.
The stated rationale is price transparency. Silicon Data chief executive Carmen Li described the pre-existing condition as one in which "two companies buying the exact same GPU capacity could pay wildly different prices with no way to know who got the better deal," and the contracts as giving the market "a public, tradable reference price for the resource every AI system runs on." CME Group chairman and chief executive Terry Duffy called compute "the new oil of the 21st century" (Source: silicondata.com).
The hedging case runs in both directions. Buyers with known future capacity needs can lock a rate; capacity operators can hedge revenue. Silicon Data reported in July 2026 that its B200 six-month forward sat near $5.25/hr against a spot rate around $5.62/hr, so the term structure at that observation paid for commitment (Source: silicondata.com). Its indices cover the H100, A100, and B200.
The regulatory question
Exchange listing runs through the Commodity Futures Trading Commission. On August 19, 2026 the Commission issued a Request for Comment on the Listing of Compute Derivatives Contracts (RIN 3038-AF77), published August 21 at 91 FR 54259, seeking input on what a designated contract market should consider before listing such a product.
Its preliminary assessment is skeptical on the threshold question. The Commission wrote that compute markets are fragmented, that price formation "primarily occurs in opaque bilateral transactions," and that compute "may not yet exhibit certain of the characteristics of commodities that typically underlie a commodity derivatives market, including fungibility, standardization, and sufficient liquidity." It observed that the price of a given unit of compute is "likely … derived from a bundle of factors (e.g., provider, region, contract structures)," so a functioning market would require standardizing those variables, both in the settlement index and in any physically deliverable specification (CFTC, Request for Comment on the Listing of Compute Derivatives Contracts).
Three problems the request identifies are specific to this underlier rather than generic novelty:
Reflexivity of the reference price. Published compute price series are built in whole or in part from posted rates that the capacity providers themselves administer, with the remaining transactions executed on venues a small number of participants operate or dominate. The Commission asked what would prevent a provider from moving a settlement index by adjusting a posted rate, directing capacity onto or away from an indexed venue, or executing or declining to execute during the observation window — and asked directly "whether the parties best positioned to influence the reference price are the same parties that supply capacity."
Unobservable inputs. Because the Commission's preliminary understanding is that non-public bilateral agreements carry the majority of economic value, it asked whether it would be appropriate to permit trading in a contract settling to a price "computed from data that the Commission may not be able to observe, verify, or surveil, in whole or in part," and whether any audits or independent verifications of that transaction data exist.
Feedback into the cash market. The request asks whether the existence of a listed contract settling to a published index would itself change provider incentives around publishing posted rates, structuring bilateral reservations, disclosing utilization and committed-capacity data, or allocating capacity among purchasers.
Alongside these it raises deliverable-supply estimation for position limits under Core Principle 5, surveillance feasibility and information-sharing arrangements with compute venues under Core Principle 4, anti-money-laundering and know-your-customer questions for intermediaries, disclosure obligations arising from a contract settling "against a geopolitically sensitive commodity" including for retail participants, and whether perpetual compute futures — contracts without a fixed expiry — offer risk-management features fixed-date contracts cannot, and what safeguards they would need.
Antecedents and scale
The Commission situates compute against commodities with comparable difficulties rather than treating it as a wholly new case. It cites electric energy, which "must be used or consumed on an as-needed basis" and cannot be bought in commercial quantities ahead of time and stored, and the 19th-century development of standardized exchange contracts recounted in Merrill Lynch v. Curran, 456 U.S. 353 (1982), where fungibility of both the commodity and the contract was the precondition for an active market.
On scale, the request cites Federico M. Bandi's (Early) AI Compute Asset Pricing (July 1, 2026), which estimates that the 2025-Q4 installed compute stock implies a gross compute service flow of roughly $430 billion to $1.3 trillion per year, or about 1.4% to 4.0% of US GDP. The same paper frames the compute price as sitting "at the interface between infrastructure and model production," carrying information about the cost of deploying and operating data centers on the supply side and the value of training and inference workloads on the demand side — so that compute futures would allow financial markets "to aggregate and reveal information about the future of the AI economy."
Policy framing
The Commission's framing is competitive rather than prudential. It cites the recommendation in America's AI Action Plan to ensure access to large-scale compute for startups and academics by improving the financial market for compute, and Executive Order 14179. Chairman Michael S. Selig described the request as "the first step toward establishing clear rules of the road for American compute markets" (Source: cftc.gov).
A compute futures market would add a further layer to the financing structures built around the AI infrastructure buildout tracked in Private Credit & AI Infrastructure: rather than taking exposure to data centers, chips, or the firms building them, a participant could take exposure to the price of the capacity itself (Source: cnbc.com).
Open questions
- Whether a compute reference price can satisfy the Appendix C standard — reliable, acceptable, publicly available, timely, and computed from a cash market that is itself liquid and not readily susceptible to manipulation — given that the majority of transacted value is bilateral and undisclosed.
- Whether the index survives the heterogeneity the Commission names: chip type, region, provider, and contract structure.
- When listing proceeds, now that the CFTC has said the contracts will not be approved in time for the announced October 5, 2026 date (the comment period runs 60 days from August 21, 2026).
Relationships
- regulated-by: Commodity Futures Trading Commission (CFTC) — the agency with listing jurisdiction
- depends-on: Compute Governance — compute as a governable input
- related: Strategic Compute Reserve — a public-provision alternative to market pricing of compute access
- related: Private Credit & AI Infrastructure — adjacent financialization of the AI buildout
- related: Kalshi — event-contract venue under the same DCM framework
- related: Nvidia & TSMC — AI Compute Infrastructure — supplier of the accelerators the indices reference
- supports: America's AI Action Plan — cited by the Commission as policy basis
Sources
- CFTC, Request for Comment on the Listing of Compute Derivatives Contracts — CFTC, Request for Comment on the Listing of Compute Derivatives Contracts, RIN 3038-AF77, August 19, 2026.
- (Source: cftc.gov) — CFTC Release 9286-26, August 19, 2026.
- (Source: cmegroup.com) — CME Group, October 5 launch announcement, August 11, 2026.
- (Source: silicondata.com) — Silicon Data / CME Group partnership announcement, May 12, 2026.
- (Source: silicondata.com) — Silicon Data, "GPU Futures: How Compute Is Becoming a Tradable Commodity," July 17, 2026 (forward-curve figures).
- (Source: cnbc.com) — CNBC, August 11, 2026.