Hyperliquid's policy arm asks the CFTC to let Americans trade perpetual futures on AI compute

The CFTC opened the consultation on 19 August, asking about the size and liquidity of compute markets, manipulation risk, customer protection and perpetual compute futures. "America cannot win the AI race without a robust derivatives market for compute," chairman Michael S. Selig said at the time, calling the request "the first step toward establishing clear rules of the road for American compute markets"[2]. Comments were open for 60 days after the notice appeared in the Federal Register on 21 August[3].

The letter's core argument is that compute behaves unlike oil or wheat. "A GPU-hour exists only in the hour it is used," it says, and it can't be stored[4]. A data centre earns rental income hour by hour and an inference platform pays for compute on every model call, so their price exposure has no end date[1].

Hedging that with monthly futures means rolling from contract to contract, paying fees and taking timing risk each time. A perpetual never expires and is kept in line with its reference price by periodic funding payments between buyers and sellers, so one position can be held as long as the exposure lasts. The letter also points out that the news that moves compute prices, such as model releases, outages and export-control announcements, doesn't keep exchange hours[1].

It doesn't dismiss normal futures. Dated contracts still suit costs with a fixed end, like a training run booked for a particular quarter, and a strip of monthly contracts gives lenders the forward price curve they need to value GPU-backed loans[4].

The group wants two kinds of contract. One would track the rental price of specified GPU hardware. The other would reference a stated volume of AI inference tokens, the unit developers are actually billed in. Because a token's price reflects both hardware costs and model efficiency, it argues, a GPU-hour contract would only imperfectly hedge an app developer's AI bill[1].

The letter makes seven recommendations. The main ones: don't hold back listings until the cash market is more transparent, let contracts settle to competing indices that meet a disclosed reliability standard rather than one official series, and let perpetuals go through the existing review process the CFTC already uses, including its 2026 policy statement on perpetual contracts. It backs a settlement-index standard proposed by Douro Labs, the core contributor to the Pyth Network, and argues that onchain venues make trades, funding payments and liquidations directly visible to regulators[1].

On risk, it names three problems specific to compute perps and a fix for each: caps or smoothing on funding rates when a model launch or chip shortage suddenly reprices compute, pre-announced multi-day transitions when an index moves to a new chip generation, and volatility-based margin plus multi-contributor indices so trading in the derivative can't move a thin underlying price[1].

The letter notes that mainstream exchanges are already moving. NYMEX has submitted two cash-settled monthly futures, each covering 730 GPU-hours and settling to a Silicon Data index, and ICE Futures U.S. has announced futures based on Ornn's compute indices, though none had started trading. Architect has announced margined compute perpetuals for institutions through a Bermuda-licensed venue and bought a US exchange to list compute derivatives, and Kalshi lists event contracts on GPU rental prices[1].

If compute gets a proper futures market, AI companies and the lenders financing data centres could lock in prices the way airlines hedge fuel. Hyperliquid's camp wants the 24/7, never-expiring version that crypto traders already use to be part of that from day one. The CFTC now has to weigh this and the other comments before deciding what compute contracts it will allow, and nothing changes until it does.