Artificial intelligence is a strategic sector of the American economy, and the computing capacity that trains and runs AI models has become one of its scarcest inputs. Private investment in AI data centers, chips, and networking equipment reached roughly 0.8 percent of U.S. GDP in the first quarter of 2026. AI is changing how Americans work, transact, and produce goods and services, and it bears directly on the nation’s economic standing.

The market for compute is still young, and it does not yet have a derivatives layer where participants can manage price risk. In the second half of 2024, renting one of NVIDIA’s H100 chips on an online marketplace cost between $2.50 and $3.19 an hour, and a year later the same hour cost between $1.92 and $2.00. A data center that bought those chips on credit had no exchange-traded instrument with which to hedge that decline, and neither did its lender. When prices rise, the developers that rent compute carry the same risk with no widely available hedging tools. Pricing in the compute market remains fragmented and largely private.

Compute cannot be stored, so exposure to its price never pauses. A GPU-hour (one hour of rented time on a graphics processing unit, the chip that powers most AI workloads) exists only in the hour it is used. Data centers produce it every hour of every day, and developers, businesses, and, increasingly, software agents consume it around the clock.

The Commodity Futures Trading Commission, the agency that oversees U.S. derivatives markets, has spent this year building a regulatory path for perpetual contracts and is now considering derivatives on compute. In May, the Commission approved the first true perpetual contract, a futures contract with no expiration date, and issued a policy statement on reviewing perpetual contracts under its existing rules. In August, the Commission asked for public comment on the listing of compute derivatives. Announcing the request, Chairman Selig said that “America cannot win the AI race without a robust derivatives market for compute.”

Today, HPC filed a comment letter explaining that compute can support a regulated derivatives market and that both dated futures and perpetual contracts belong in it. Compute fits the Commodity Exchange Act’s definition of a “commodity,” the threshold a product has to clear before futures on it can list on a U.S. exchange. The letter builds on our joint August comment with trade[XYZ] on energy perpetual contracts, which made the case for extending perpetual contract markets beyond digital assets to core commodity classes.

Compute already shows the conditions that successful futures markets share: a commercial need to hedge, speculators willing to take the other side, a large volume of transactions, and favorable public policy. GPU rental rates have moved sharply in both directions. The Commission’s request cites an estimate of $430 billion to $1.3 trillion in compute services sold each year. Providers and their lenders lose when rental rates fall, and developers and inference platforms (the companies that run AI models for customers) lose when they rise, so each side needs the other to hedge. Early markets on compute prices outside the United States show speculative interest, and federal policy favors building deep markets for compute at home.

The unusual features of the compute market are questions of contract design that the Commission has resolved in other markets. Compute varies by chip, location, and term; its supply is concentrated in a handful of companies; and its hardware turns over quickly. The Commission adapted its deliverable-supply analysis for electricity, which also cannot be stored. Henry Hub in Louisiana became the reference price for American natural gas in large part because the first natural gas futures contract, listed in 1990, used it as the delivery point. Exchange-listed contracts can help a young cash market settle on a reference price.

Perpetual contracts give compute market participants a simpler way to hold continuous price exposure. A data center earns rental revenue by the hour, and an inference platform pays for compute every day, so neither exposure has a fixed end date. Since perpetual contracts have no expiry, a hedger can hold one position for as long as its exposure lasts, with no roll to execute, pay for, or mistime, and a new market can concentrate its scarce liquidity in a single instrument. The contract’s reference index can also move from one chip generation to the next under rules disclosed in advance, much as the price feeds for trade[XYZ]’s crude oil perpetual contract markets on Hyperliquid shift from one futures month to the next over several days. Perpetual futures are a useful tool alongside dated futures, not a replacement for them. Dated futures better serve participants whose exposure runs for a defined period, such as an AI developer’s training run planned for a particular quarter.

Onchain infrastructure already handles part of the compute market, and its share will grow as software agents take on more of the work of buying and paying for compute. Stablecoin-funded lending against GPU collateral, tokenized inference credits, and inference networks that settle onchain are operating today. BlackRock describes compute as “a distinct, large, and increasingly investable economic resource” and anticipates software agents procuring and paying for it without a human in the loop. Software agents that buy compute around the clock will need markets that trade continuously and settle through programmable infrastructure, which is how onchain perpetual contract markets already operate.

Our comment asks the Commission to take six steps:

1.  Facilitate the introduction of derivatives on both GPU-hours and inference tokens, the two forms in which compute is sold (rented hardware time and the units of model output that developers buy).

2. Articulate principles for reliable compute reference indices, and let exchanges list contracts that settle to any index meeting that standard, whatever the maturity of the cash market. We support the standard that Douro Labs, the core contributor to the Pyth Network, proposed in its comment, under which an aggregated, multi-contributor price feed of the kind Pyth publishes would qualify.

3.  Permit contracts to settle to competing reference indices instead of designating a single benchmark.

4. Address manipulation and concentration concerns through heightened market monitoring, information sharing, and appropriate position limits during the market’s early phase.

5. Permit perpetual compute contracts to proceed through the Commission’s existing case-by-case review of new contracts under Regulation 40.3 and the 2026 policy statement.

6.  Foster the development of both dated futures and perpetual contract markets for compute.

The country that builds the deepest market for compute will be best positioned to lead in artificial intelligence, and that market should be built in the United States. We commend the Commission for its work this year to bring perpetual contracts onshore, and we encourage the Commission to extend the product class to new commodity types, including compute. HPC will continue working with the Commission and its staff to give Americans access to onchain markets under American oversight.

Read our full comment letter here.