GPU Compute Futures and the CFTC: What Happens When Computing Power Becomes a Commodity

De Silva Law Offices, LLC | October 17, 2026

AI companies often rent their computing power by the hour, on Nvidia chips in another company’s data center. The price of that time has moved sharply since 2023. Apart from private reservations, buyers and sellers have had few ways to hedge it.

CME Group and Intercontinental Exchange (ICE) have each proposed futures contracts that settle on an index of GPU rental prices. CME’s New York Mercantile Exchange (NYMEX) asked the CFTC on August 11 to approve contracts on Nvidia’s H100 and B200 chips and planned to begin trading October 5. On September 21 the CFTC extended its review to November 9, and its product database lists both contracts as “Approval Pending (90).” ICE is waiting for the CFTC to close its request for comment on compute derivatives, which takes comments through October 20.

We recently examined the SEC’s approval of Cboe’s binary options on company performance figures, which made numbers a company reports about itself the basis for a securities option. Compute futures would do something similar under the commodities laws.

In our view, the jurisdictional question is the more settled one. The harder questions are whether the compute market can produce a settlement price reliable enough for futures trading, and what changes for the companies in that market once it does.

What NYMEX and ICE Have Proposed: How GPU Compute Futures Work

Both proposals are monthly, cash-settled futures on an index of hourly GPU rental prices. The main difference between them is the index.

Sources: NYMEX Submission No. 26-370; ICE Futures U.S. notice of September 29, 2026.

Silicon Data, a benchmark firm backed by the trading firm DRW, gathers rental and private transaction prices from neoclouds, hyperscalers, colocation markets and rental platforms. It then adjusts each price for rental type, geography, CPU platform and GPU variant (methodology). Ornn says its index is built only from executed transactions (ICE announcement).

Exposure to compute prices already trades in other forms. Kalshi has listed event contracts on GPU rental prices this year that settle against Ornn’s index (DCD). In July it began publishing forward curves derived from its own trading, which it offers as reference prices for compute swaps and over-the-counter deals (Markets Media). Polymarket’s U.S. exchange filed terms in August for contracts on Ornn’s GPU index and on an Ornn index of the per-token prices that four AI model developers charge (filing).

In May, Architect Financial Technologies acquired a U.S. designated contract market to launch an exchange for compute futures and options, subject to regulatory review (Markets Media).

Is Compute a Commodity Under the Commodity Exchange Act?

It very likely is, once futures on it trade. The Commodity Exchange Act (CEA) defines “commodity” with a list of farm products. The list is followed by “all other goods and articles” and “all services, rights, and interests ... in which contracts for future delivery are presently or in the future dealt in” (7 U.S.C. § 1a(9)). Renting time on a GPU is a service. The CFTC’s request describes the underlying as rented access to hardware the buyer does not own, and it calls compute “a scarce, capital-intensive commodity.”

Courts have read the definition broadly. In CFTC v. My Big Coin Pay, Inc., 334 F. Supp. 3d 492 (D. Mass. 2018), the court denied a motion to dismiss. It accepted the CFTC’s theory that a virtual currency with no futures market of its own was a commodity, because futures traded on bitcoin, another currency in the same class.

The definition excludes only two things by name: onions, whose futures Congress banned in 1958, and motion picture box office receipts, excluded in 2010. The second is a useful precedent for the review under way now. In 2010 the CFTC designated two exchanges that planned to list box office futures, on the condition that they seek approval for such contracts instead of self-certifying them (Commissioner Sommers).

The CFTC’s review focused on manipulation risk, the reliability of box office data and whether the contracts could be used to hedge. Over the objections of the major studios, it approved both in June 2010. Weeks later, Congress removed box office receipts from the definition in the Dodd-Frank Act.

New underlyings come within the statute by default. The CFTC decides whether a particular contract meets the Core Principles, and when a whole category has been removed, Congress has removed it. Current federal policy favors a market in compute derivatives. The White House’s July 2025 AI Action Plan called for improving the financial market for compute. Chairman Michael Selig has said that “America cannot win the AI race without a robust derivatives market for compute” (CFTC).

Core Principle 3: Can a Compute Index Support a Futures Market?

That question is likely to decide the NYMEX review. Under Regulation 40.3, the CFTC must approve a new contract unless its terms violate the CEA or the CFTC’s rules (17 C.F.R. § 40.3(b)). Core Principle 3 permits an exchange to list only contracts that are not readily susceptible to manipulation (7 U.S.C. § 7(d)(3)).

For cash-settled contracts, the CFTC’s guidance in Appendix C to Part 38 looks hardest at the settlement price. It should be reliable, publicly available and timely, and it should come from a cash market that is liquid and not itself easy to manipulate. The CFTC’s request explains why compute struggles with that standard. In its preliminary view, compute markets are fragmented, most prices are set in private bilateral deals, and dominant participants may have significant pricing power. Compute also “may not yet exhibit” the fungibility, standardization and liquidity of the commodities that typically underlie derivatives.

Several of the CFTC’s questions go to the index itself. It asks whether to allow trading in a contract that settles on data the CFTC “may not be able to observe, verify, or surveil.” It asks whether the firms best placed to influence an index are the same providers that supply the capacity and contribute the rates and transactions behind it. It also asks what it should do if no compute price series meets the Appendix C standard.

The public record does not yet show how the exchanges answer. NYMEX’s filing says its index is “supported by a robust level of trading in the underlying cash market.” The supporting data, however, is in an exhibit filed under a request for confidential treatment. In January, Ornn’s chief executive said its index draws on several partner spot markets, with a single GPU broker as the largest contributor (The Block).

Both exchanges use a familiar safeguard. A final price averaged over a month of daily index values is harder to move than a single day’s reading, and many power and freight contracts settle the same way. Polymarket’s compute contracts go further and bar the index source’s insiders, and the owners of any entity able to influence the settlement data, from trading. That may suit an event contract. In a futures market, though, many of the firms that post rates and report trades are the providers and marketplaces that need to hedge. In our view, exchanges should watch those accounts closely during the settlement month rather than bar them.

Compute also cannot be stored in quantity. An unused GPU-hour is gone, so no buy-and-hold trade ties spot and futures prices together, and neither proposal provides for delivery. In a footnote, the CFTC cites electricity, which has the same limitation, as a historical antecedent. Many electricity futures, however, settle on prices set in markets run by regulated grid operators. Compute has no public price of that kind.

Ocean freight is a closer match for compute. A day of a ship’s time, like an hour of GPU time, is lost if no one buys it. Freight rates are also set in private charters. Even so, freight futures have traded for years. They settle on Baltic Exchange rates built from the assessments of shipbroker panels. Shipowners and charterers, who have the most direct stake in the price, cannot serve on those panels (Baltic Exchange). Compute indexes draw largely on rates and trades supplied by the providers and venues themselves. For compute, the index provider’s data, methods and governance have to do the work that independent panels do for freight.

Basis Risk: What a GPU Compute Hedge Covers

A futures position hedges a company’s compute costs only to the extent those costs move with the index. An H100-hour from one provider is not the same product as one from another. Location, network interconnect, cluster size, uptime terms, software support and the provider’s credit all affect the price.

Silicon Data’s own figures show how wide the spread can be. In April and May 2026, its hyperscaler H100 index stayed between about $7.40 and $7.55 per GPU-hour. Its neocloud index ran from about $2.47 to $2.76 over the same period and was roughly five times as volatile (Silicon Data).

The NYMEX contracts settle on on-demand prices for U.S. capacity. In the CFTC’s preliminary understanding, non-public bilateral agreements carry most of the economic value in compute. Long-term reservations are one example. A provider selling on-demand capacity has direct exposure to the index. An AI lab that has reserved capacity for several years has far less until the reservation ends.

A company that hedges at scale can apply under NYMEX’s rules for an exemption from position limits “for bona fide commercial purposes.” Before it does, it should measure how its own costs have moved against the index. Where the correlation is weak, a futures position can add risk instead of reducing it.

False Reporting and Manipulation Risk for Compute Providers and Data Contributors

The CFTC’s anti-fraud and anti-manipulation authority is not limited to futures. Section 6(c)(1) of the CEA and CFTC Rule 180.1 also reach conduct in connection with “a contract of sale of any commodity in interstate commerce” (7 U.S.C. § 9(1)). If compute is a commodity, that language would likely reach the compute cash market. My Big Coin was itself a fraud case about sales of a virtual currency, not futures. How far this authority extends into ordinary GPU rental agreements has not been tested.

The provision most relevant to compute benchmarks concerns false reporting. Since the Dodd-Frank Act, the CEA has treated as manipulation a “false or misleading or inaccurate report” of market information that affects or tends to affect the price of a commodity. The report must be made knowingly or with reckless disregard (7 U.S.C. § 9(1)(A)). The statute even mentions price reporting services: a good-faith mistake in sending them information is not a violation (7 U.S.C. § 9(1)(C)). Section 9(a)(2) separately makes it a felony to knowingly deliver false or knowingly inaccurate market reports (7 U.S.C. § 13(a)(2)).

The CFTC has used Section 9(a)(2) against companies in a similar position. Beginning in late 2002, it charged energy companies whose traders had sent false or inaccurate natural gas trade data to index publishers such as Platts. Those indexes priced physical gas deals, and futures traders used them for price discovery (CFTC). By November 2003, the CFTC had imposed more than $130 million in civil penalties in energy trading cases (CFTC release). The CFTC’s LIBOR orders, beginning with Barclays in 2012, also found false reporting.

Compute indexes depend on the same kinds of inputs: posted rates, reported transactions and broker data, often from firms with an interest in the price. A provider that reports transactions that did not occur, to move a settlement index, could face liability under these provisions. So could a broker or marketplace that passes along data it knows to be inaccurate. The CFTC has also asked whether exchanges should have information-sharing arrangements with every venue and provider whose data enters a settlement price. That could bring those firms within the futures market’s surveillance.

Posted rates are a less certain case. GPU Economy publishes an index built on posted rates. It told the CFTC in its comment that one provider changing its rates in every region it lists could move its weekly index by up to 10.5 percent. A provider that posts a rate it will not honor, to move an index, is doing something much like spoofing. The CEA’s anti-spoofing provision would not reach it. That provision applies only to conduct on or subject to the rules of a registered entity, such as an exchange. Rule 180.1 may reach a false posted rate, since it covers reckless false statements and false reports. We are not aware of any case applying these provisions to one.

Compute marketplaces have a separate question to consider. A platform where customers buy capacity they intend to use is a market for a service. A platform that lists standardized forward contracts, lets multiple participants trade reserved capacity among themselves, or offers offsetting positions or cash settlement may be offering futures or swaps. That would bring DCM or SEF registration into play. Ordinary reservations, where the buyer takes the capacity it pays for, are likely to be treated as commercial service contracts, or as forward contracts, rather than swaps or futures (see 77 Fed. Reg. 48,208 (Aug. 13, 2012)). A right to settle the difference in cash, a price option that can be separated from the reservation, or capacity that is not dedicated to the buyer could change that answer.

Perpetual Compute Futures

The CFTC’s request gives perpetual compute futures a section of their own. A perpetual never expires, and periodic funding payments keep its price close to the index. The CFTC asks whether perpetuals would offer hedging features that existing products do not, and what unique risks they would pose.

The Block reported in July that Architect’s Bermuda exchange now lists perpetual-style compute futures. In the United States, the CFTC approved Kalshi’s bitcoin perpetual as a futures contract in May. CME has asked a federal court to hold that perpetuals are swaps, a dispute we examined in September.

A perpetual would put more weight on the index. A monthly contract depends on the index during its final month, and averaging dilutes any single day. A perpetual’s funding payments are typically calculated at regular intervals from the gap between the contract’s price and the index, so the index matters every day a position is open. The depth of the index and the concentration of its contributors matter more for that structure, especially if it is offered to retail traders. The request’s customer-protection questions, which cover all compute contracts, ask about retail protections and about disclosures for a “geopolitically sensitive” underlying.

Corporate Data and Computing Power: How Federal Derivatives Law Absorbs New Exposures

Cboe’s KPI options and the compute futures proposals each make a listed derivative out of a figure that matters mainly to the businesses that report or pay it. The legal routes differ. Securities law reached company KPIs through the Exchange Act’s treatment of options “based on the value” of a security. Commodities law would reach compute through the CEA’s definition of commodity, which covers services on which futures trade.

Cboe’s contracts settle on figures that companies report to the SEC, where false statements carry liability under the securities laws. Compute indexes have no mandatory disclosure regime behind them, although knowingly or recklessly false submissions can violate the CEA.

Prediction markets moved first in both cases. Kalshi, a CFTC-regulated exchange, listed contracts on company KPIs and on GPU rental prices before any traditional exchange product was approved. In our view, the pattern is likely to repeat as the existing indexes for memory and per-token prices mature.

What to Watch After the October 20 Comment Deadline

Comments on the CFTC’s request are due October 20, and the docket showed 48 received as of October 16 (Regulations.gov). Submissions from index providers, compute providers and exchanges will show how the industry proposes to handle verification and contributor conflicts.

We will be filing our own comment on October 19. It will ask the CFTC to set out its expectations for compute derivatives in guidance. Under our proposal, a compute contract would settle only on an index whose administrator can give the exchange, and through it the CFTC, the data behind each settlement value. The comment will also ask the CFTC to make clear that a posted rate a provider will not honor can be a false report. And it will argue that exchanges should monitor index contributors rather than bar them from trading.

NYMEX’s extended review runs through November 9. Under Regulation 40.3, the contracts will be deemed approved when that period ends unless the CFTC issues a notice of non-approval or NYMEX agrees in writing to a longer review. A non-approval would not end the matter, since NYMEX could revise and resubmit. It would, however, be presumptive evidence that NYMEX could not truthfully self-certify the same, or substantially the same, product (17 C.F.R. § 40.3(e)(2)).

ICE has tied its listing to the end of the comment period, and Architect’s U.S. exchange is pending review. The CFTC may follow the comment period with compute-specific guidance on index governance, data verification or information-sharing with contributors. That guidance would set the terms on which a new group of companies, many with no prior CFTC experience, enters a regulated market.

De Silva Law Offices, LLC advises exchanges, trading firms, intermediaries and companies entering CFTC-regulated markets for the first time, including compute providers, compute marketplaces and benchmark administrators. Our futures and derivatives practice covers product classification, exchange and intermediary registration, compliance and enforcement defense. Questions about this article may be directed to info@desilvalawoffices.com or 312-500-8424.

NB: This article is commentary on a matter of public record and is not legal advice. Reading it does not create an attorney-client relationship. R Tamara de Silva is the Managing Partner of De Silva Law Offices, LLC, Chicago, and writes on financial regulation, derivatives and prediction markets.