AI Compute Perpetual Futures: Hyperliquid Policy Center Asks the CFTC for GPU-Hour Perps

The Hyperliquid Policy Center filed a comment letter with the CFTC on October 20 asking the agency to let regulated U.S. exchanges list AI compute perpetual futures and dated futures tied to GPU rental prices and the cost of running AI models, per crypto.news. The letter was signed by HPC CEO Jake Chervinsky and senior counsel Brad Bourque. It is a policy recommendation, not a product launch, and nothing has been approved.

For perp traders the letter matters for a specific reason. It is the first time a Hyperliquid-aligned group has laid out, in a regulatory filing, how a perp should be built for an underlying that cannot be stored, has no deep public spot market, and can reprice violently around a single model release. Every design choice in it, from funding caps to index rules, maps onto mechanics you already trade on HIP-3 markets.

What the AI Compute Perpetual Futures Letter Asks For

The CFTC opened its compute derivatives consultation on August 19, 2026, under press release 9286-26, asking for feedback on cash market liquidity, manipulation, customer protection and perpetual compute futures, according to crypto.news. Chairman Michael Selig framed the consultation around the AI race, arguing the U.S. needs a deep derivatives market for compute to compete. HPC's letter is its answer to that request.

The core legal argument is that rented computing capacity is a commodity under the Commodity Exchange Act. If the CFTC accepts that, compute falls under the same agency that already oversees energy and agricultural futures, and contracts on it can be listed by designated contract markets.

HPC proposes two families of underlyings. GPU-hour contracts would track the rental price of specified hardware. Inference contracts would reference access to a stated volume of AI model tokens. TokenPost summarizes the goal as giving data center operators, AI developers and lenders a way to hedge swings in compute costs, and notes the CFTC has not approved any of the contracts HPC is seeking.

Why a Perp Fits Compute Better Than Dated Futures

The letter does not argue for perps alone. It splits the use cases. Dated futures suit costs with a known endpoint, such as a training run planned for a specific quarter or a scheduled capacity purchase, and successive monthly contracts could build a forward curve for lenders, per crypto.news.

Perpetuals are pitched for the other side: ongoing rental revenue and running compute costs that never expire. AMBCrypto puts the argument plainly: compute cannot be stored and is consumed as it is produced, so a contract with no expiry date matches the exposure. A cloud provider renting out H100s every hour has a continuous short exposure to rental prices, and rolling quarterly futures forever adds basis risk and execution cost at every roll.

The price history HPC cites shows why the hedge is wanted. Silicon Data's series puts Nvidia H100 marketplace rental rates at about $2.50 to $3.19 per GPU-hour in the second half of 2024, versus $1.92 to $2.00 in the second half of 2025, per crypto.news. That is a decline of roughly 23% to 37% in a year, depending on which end of the range you compare, with no listed instrument to hedge it. HPC also cites Epoch AI's estimate that private investment in AI data centers, hardware and networking reached roughly 0.8% of U.S. GDP in Q1 2026.

Funding Caps and Index Design: The Orderflow Details

The most useful part of the letter for traders is the safeguard list. HPC proposes caps on open interest and on the margin multiple users can run, plus funding rate caps or funding calculated over longer windows, specifically to limit sudden costs around model releases or hardware shortages, per crypto.news.

That second point is a direct admission of how a thin compute perp would behave. A new frontier model drops, inference demand spikes, spot rental quotes jump, and longs pile in. On a standard hourly funding schedule, the perp would print an extreme premium and shorts (the hedgers HPC wants to attract) would pay a large funding bill for holding the hedge they were told to hold. Smoothing funding over longer periods trades responsiveness for stability, and it means the perp can drift further from the index before funding pulls it back.

The index is the other weak point. The letter notes that much of the economic value in compute moves through private reserved capacity and multiyear agreements whose prices are not public. A perp that settles off marketplace quotes is pricing the marginal, spot slice of the market, not the bulk of it. HPC's answer is independent price contributors, outlier filtering, position limits, and advance notice with gradual transitions when an index swaps older hardware for newer chips, according to crypto.news. TokenPost adds that HPC wants standards for compute price indexes and the option to settle against different indexes.

Hardware rotation is a problem crypto perps have never had. BTC is BTC. An H100 index that rolls into a newer chip is a different underlying after the switch, and any open position carries that transition risk. Expect funding and basis to behave oddly around announced index changes if these contracts ever list.

Where This Sits in the U.S. Perp Pipeline

The compute letter is the latest in a run of HPC filings. On August 24 it asked the SEC and CFTC to regulate stock-linked perps as security futures, and on August 26 it requested a U.S. path for energy perps on WTI, Brent and Henry Hub natural gas alongside trade[XYZ], per crypto.news. On October 19 it submitted five recommendations on the SEC's Regulation Crypto Assets.

The competition is not standing still. NYMEX and ICE Futures U.S. have proposed cash settled compute futures, though neither had started trading when HPC wrote its letter, according to the same report. HPC also points to compute price event contracts and perpetual offerings already available through a Bermuda licensed venue. AMBCrypto notes HPC cited trade[XYZ]'s crude oil perps on Hyperliquid as precedent for an always-on contract on a real-world commodity.

The pattern is consistent: products like oil perps trade on Hyperliquid first, and the policy center then argues to U.S. regulators that the same structure works under existing law. The CFTC consultation has no announced decision date in the reporting, so any compute perp on a U.S. venue is a future event, not a current market.

How to Read Compute Perps if They List

If compute perps arrive on HIP-3 or a regulated venue, the first weeks will look like every new real-world perp: thin books, wide spreads, funding driven by a small number of directional accounts, and an index that hedgers do not fully trust yet. The signals that separate real hedging demand from speculation are the same ones you use on oil or equity perps today.

Watch whether open interest grows steadily while funding stays near neutral, which points to two-sided hedging flow. A funding spike with flat open interest after a model release points to speculative longs chasing a headline. Track cumulative volume delta against the index move: if the perp rallies on thin buy delta while the index barely moves, the premium is being set by a few accounts, not by the compute market.

On Buildix, the screener covers Hyperliquid perps including HIP-3 markets, and each pair view shows CVD and open interest side by side, which is the place to compare how a new real-world contract behaves against oil, equity and pre-IPO perps that came before it. The letter is a design document for a market that does not exist yet, and the funding cap section alone tells you how its authors expect it to trade.