The Hyperliquid Policy Center has urged the US Commodity Futures Trading Commission (CFTC) to permit futures and perpetual contracts linked to GPU rental prices and the cost of running AI models.
In an 20 October letter, chief executive Jake Chervinsky and senior counsel Brad Bourque called on the regulator to examine compute derivatives under existing rules. The organisation said data-centre operators, lenders and AI developers needed tools to manage changing computing costs.
The independent research and advocacy group, which supports regulated US access to onchain markets including Hyperliquid, argued that rented computing capacity qualified as a commodity under the Commodity Exchange Act. It also backed contracts based on AI inference costs.
The submission was a policy recommendation, not an announcement that Hyperliquid would launch compute contracts or that regulators had approved them.
Perpetuals for continuing costs
The group said computing exposure existed continuously. Data centres generated rental income by the hour, while inference platforms paid for processing whenever customers used an AI model.
“A GPU-hour exists only in the hour it is used,” the policy centre wrote.
It said perpetual contracts, which have no expiry date, could reflect these recurring revenues and expenses through funding payments between buyers and sellers. A data-centre operator could hedge against falling rental income, while an AI platform could protect itself from higher computing bills.
Lenders financing hardware also face risks because weaker rental income can reduce borrowers’ cash flow and the value of collateral.
For expenses with a fixed timetable, the letter supported dated futures. Examples included an AI training run planned for a particular quarter or a scheduled capacity purchase. A series of monthly contracts could create the forward price curve needed by lenders and investors when assessing financing arrangements.
Different contracts for GPUs and inference
The proposed products would cover GPU-hour prices and AI-model inference. GPU contracts would follow the rental cost of specified hardware, while inference contracts would track access to a stated quantity of model tokens the units used to measure text processed and produced by an AI system.
The centre said inference platforms and application developers would benefit from the latter. GPU-rental contracts would offer only an imperfect hedge because inference bills depended on both computing prices and model efficiency.
Using Silicon Data’s historical series, it said Nvidia H100 marketplace rental rates were approximately $2.50–$3.19 per GPU-hour in the second half of 2024, compared with $1.92–$2.00 in the second half of 2025.
It also cited Epoch AI research estimating that private investment in US AI data centres, computing hardware and networking equipment reached roughly 0.8% of US GDP in the first quarter of 2026.
The letter said most economic activity still occurred through private agreements. Hourly rentals were publicly visible, but prices for reserved capacity and multiyear purchase commitments generally were not.
Safeguards and regulatory background
The proposed framework includes competing benchmarks, independent price contributors, filters for unusual data, position limits and information-sharing arrangements. It also recommends leverage and open-interest caps, transparent funding costs, and clear explanations of liquidation thresholds and operator discretion.
The centre warned that thin markets, concentrated supply, limited transaction data and changing hardware could create manipulation risks. It proposed advance notice and gradual index changes when newer chips replace older models.
For perpetuals, it suggested funding-rate caps or calculations averaged over longer periods to reduce sharp costs during model releases and hardware shortages.
The CFTC began its compute-derivatives consultation on 19 August, seeking views on cash-market liquidity, manipulation, customer protection and perpetual compute futures. Chairman Michael Selig described derivatives markets as part of America’s AI ambitions.
The letter said NYMEX and ICE Futures U.S. had proposed cash-settled compute futures, although neither had begun trading when it was submitted. It also referred to compute-price event contracts and perpetuals offered through a Bermuda-licensed venue.
The centre previously sought regulated energy perpetuals on 26 August, including contracts linked to West Texas Intermediate crude, Brent crude and Henry Hub natural gas. On 24 August, it asked the SEC and CFTC to treat stock-linked perpetuals as security futures under joint oversight. On 19 October, it submitted five SEC recommendations on Regulation Crypto Assets covering airdrops, secondary trading, developer protections, automated token buybacks and state securities requirements.
