CFTC Considers the Viability of Perpetual Futures in GPU Computing

The U.S. Commodity Futures Trading Commission (CFTC) began soliciting public comments on the listing of compute derivative contracts on August 19, 2026. This announcement was published in the Federal Register on August 21, 2026 (91 FR 54259–54264, document number 2026-17163, RIN 3038-AF77). The majority of the document discusses consultations regarding standard market structures, including considerations of fungibility, clearing indicators, and price manipulation risks. Under Section II.4, titled "Perpetual Compute Futures," the CFTC is seeking two questions from the public regarding contract types that are not listed on any designated contract market in the U.S. The comment submission period is open until October 20, 2026, which coincides with the deadline for public comments on SEC regulations regarding cryptocurrency securities. While "perpetual futures" are a familiar financial product for those experienced in crypto derivatives trading, the key point to note this time is the underlying asset, particularly the importance of understanding it before an increase in submissions from industry stakeholders involved in GPU rentals.

"Compute" refers to processing power, a resource heavily consumed by large language models. The commission has clearly defined the underlying asset without ambiguity. According to the document, the underlying commodity for compute futures contracts is generally described as "access to computational power borrowed from hardware not owned by the purchaser," with the rental price per hour of an Nvidia B200 provided as a specific example. It also leaves room for other measurement units, indicating that "access to a certain amount of LLM inference tokens" is another option. The crucial aspect is that it is a "rental." This compute contract does not reference the chips themselves, data centers, or shares of operating companies. It refers to one hour of someone else's chip (typically priced in USD per GPU hour). This is similar to how a power contract does not reference turbines but states "1 MWh."

Such benchmarks have already been published. Silicon Data has calculated values from multiple emerging cloud providers, hyperscalers, colocation markets, and private rental platforms, standardizing machine specs, rental periods, and regional factors, and publishes the ++H100 Rental Price Index++ (ticker: SDH100RT) daily. As of August 24, 2026, at 07:50 UTC, the index indicated a price of $2.53 per GPU hour.

This market demand arises for the same reasons airlines hedge jet fuel prices. Compute is the largest single cost in the profit and loss statements of AI agents and cutting-edge AI services, and rental providers have developed capital investment plans amid uncertainties in future demand forecasts. The CFTC refers to the ++AI Action Plan++ announced in July 2025 as a framework for policy, which suggests that the compute market should be structured to be more accessible to startups and academia.

One hour of H100 GPU operation, unlike almost all existing listed commodities, "cannot be stored = has an expiration date." No one holds it as inventory, and it cannot be carried over to the next quarter. Unused time is simply a loss. The commission also cites precedents from the 2012 power market in a footnote, stating that "electricity is consumed when needed and is not characterized by being bought up in advance for delivery and accumulation." This serves as the basis for treating compute as a commodity. However, the commission also addresses opposing arguments, explicitly stating that "compute is unlikely to meet the typical characteristics of fungibility, standardization, and sufficient liquidity in the commodity derivatives market," warning that the market is fragmented, price formation occurs through largely private bilateral transactions, and there is a risk of price manipulation by dominant participants. Reading the list of questions as a checklist of "realized conditions" reveals the essential hurdles.

|-------------------|--------------------------------| | Tests that Listed Contracts Must Meet | CFTC's Current Assessment of Compute | | Cross-producer fungible units | Weak. Price varies by provider, region, and contract type | | Standardization of specifications and grades | Unestablished. Standardization needs to be built | | Publicly observable trading records | Thin. The preliminary view is that much of the substantial value lies in private bilateral contracts | | Estimable delivery supply | Open. Seeking estimation methods based on Core Principle 5 | | Difficult to manipulate clearing prices | Open. Many quoted prices are managed by providers |

Questions regarding price manipulation are among the sharpest points in the document. If the compute index partially uses prices suggested by capacity providers themselves, the capacity provider will have the most influence on the clearing price. The commission directly inquires about measures to prevent price adjustments and biases in trading volume during the observation window. This is not merely a rhetorical question.

Section II.4 contains only two sub-questions and about 70 words, but this is why the document attracts attention from industry stakeholders beyond AI infrastructure. The first question is, "What advantages or risk-hedging functions can perpetual compute futures add for market participants compared to traditional or fixed-term futures?" The second is, "Are there unique risks associated with being perpetual, and do additional safety measures need to be implemented?"

Perpetual futures contracts have no expiration or delivery dates, and adjustments are made through periodic funding rates to maintain values close to the reference index price between long and short positions (not through convergence at settlement). This funding rate serves the role of "expiration" in fixed-term contracts.

Imagine applying this to an underlying asset that essentially has no delivery date, such as "rental compute." Rented compute is constantly consumed, priced on an hourly basis, and is never "stored in a warehouse." Fixed-term contracts that settle in a specific calendar month are merely a convenient boundary from a physical and economic standpoint. Operators looking to hedge compute costs three years out would need to roll every month, incurring basis costs each time. With perpetual contracts, rolling is unnecessary, providing a strong structural advantage for what is essentially a "non-storable commodity."

However, risks are also involved. In thinly traded markets, funding burdens can significantly affect the economic viability of positions, and the index anchor is weaker when it relies on "prices posted by the provider" rather than physical assets. Additionally, concerns have been raised that contracts without expiration lack the enforcement of physical delivery, creating a disconnect from reality. These are the main points on which the committee is seeking written responses from the industry.

Michael S. Selig, who took office as the 16th chairman on December 22, 2025, stated in a concurrent CFTC release, "Just as the U.S. market established a 'gold standard' in commodity trading that supports industrial economies, we will create standards for new commodities that support the intelligence economy." This is positioned as the first step in establishing clear rules for the U.S. compute market. According to voting records, there was a majority in favor, with no opposing votes.

At this point, there are clear limitations. The "public comment solicitation" is a consultation, not a rule-making or approval process, nor does it permit product listings. The original text does not provide grounds for anyone to list perpetual compute contracts, and October 20 is merely the deadline for public comments.

This regulatory discussion did not arise suddenly. On August 11, CME Group and Silicon Data announced their intention to release two contracts (Silicon Data H100 rental index futures and B200 rental index futures) by October 5, contingent upon regulatory review. Both are expected to involve the rental of a single GPU for about a month under NYMEX rules, with cash settlement based on the Silicon Data index rather than physical delivery.

There are two important clarifications. First, the target date of October 5 is merely a "statement of intent" dependent on regulatory review, and the transition from announcement to actual operation is a typical hurdle for such new markets. Second, both contracts are "fixed-term monthly futures," which is the very structure the committee is requesting comparative opinions on from the industry regarding perpetual contracts.

The timeline is also significant. Date-stamped products precede perpetual ones. This is the opposite of the cryptocurrency industry, where perpetual types generated significant trading volumes first, followed by market interest in fixed-term contracts. Data center operators managing GPU resources are nearing the point of officially answering, "Which truly helps with our risk hedging?"

Confusing these aspects is one of the most common mistakes in this field, and many reports are already in disarray. The committee is currently simultaneously publishing six separate public comment cases, each with distinct content, scope, and deadlines.

|-------------------------------------------------|----------------| | CFTC Comment Files | Deadline | | Perpetual contracts referring to standard futures 24/7 trading expansion and physically deliverable or storable energy products | August 26, 2026 (Wednesday) | | SEC-CFTC Joint Portfolio Margin Rules | August 31, 2026 (Monday) | | Review of Mandatory Swap Clearing Standards | September 28, 2026 (Monday) | | Duplicate Regulations for Commodity Pool Operators and Trading Advisors | October 5, 2026 (Monday) | | Conflicts of Interest and Affiliation | October 5, 2026 (Monday) | | Listing of Compute Derivatives | October 20, 2026 (Tuesday) |

The most commonly confused item is the first line. While two of the file names in this table contain the word "perpetual," the public comment on 24/7 trading pertains to "general standard futures for year-round round-the-clock trading," with the perpetual aspect limited in scope to "physically deliverable or storable energy products." Compute is not physically deliverable or storable, hence it became a separate document. Additionally, October 20 coincides with two public comments related to cryptocurrency. The SEC's deadline for the securities offering rule (91 FR 54510) is also on the same day, meaning authors submitting to both will have the same calendar preparation.

Are compute futures crypto products?
No. They are settled based on the rental price of GPU resources and have no relation to tokens or blockchains. The overlap with the cryptocurrency industry lies in the "design aspect," as the perpetual structure that the CFTC is soliciting opinions on has originally developed and been validated in the crypto-derivative market.

Can U.S. traders purchase perpetual compute contracts?
Not at this time; there are no pending applications for such cases in any U.S. designated contract market. The committee is in the early stages of asking the industry, "What requirements are necessary?" The two previously announced products by CME and Silicon Data are also not perpetual; both are fixed-term monthly futures.

What kind of hedge does compute perpetual provide?
It allows for exposure to the hourly cost of renting AI capacity to be hedged indefinitely. Cloud providers can lock in revenue, while model developers can cap input costs, eliminating the need to roll positions monthly or incur basis costs.

Is October 20 the start of something?
It simply means the end of public comment acceptance. There is no statutory countdown that becomes obligatory after the request for comments is submitted, and there is no guarantee that rule proposals or guidance will emerge.

The essence does not lie in "the product design itself." Cash-settled Bitcoin perpetuals have already been cleared in U.S. designated markets, making their contract design a debated area. The change that occurred on August 19 was on the side of the underlying asset, as the official presentation of "perpetual" and "compute" by U.S. federal regulatory agencies under the same headline is the first clear sign that "perpetual design has applicability beyond crypto."

There are two important dates. October 5 is the target date for the H100 and B200 contracts by CME and Silicon Data (contingent on regulatory review). If it is delayed, it indicates difficulties in the review process beyond the prior announcement. October 20 is the deadline for comment acceptance and the publication of submitted content, marking the day when discussions around the "ideal compute perpetual model" will be revealed in full text rather than press releases.

In the meantime, what to watch for is "which groups submit opinions." If exchanges and data vendors dominate, it will be a challenge for "market structure." If hyperscalers, emerging clouds, and AI labs join in, it will represent "real market demand," and demand will be the key to pushing public comments towards "listing."