Commodity brokers gear up to trade AI compute futures as CFTC review continues

Wall Street has futures for oil, corn, cattle and orange juice. Soon it may have them for renting an Nvidia chip by the hour.

Commodity brokers are preparing to trade derivatives based on the cost of renting AI computing power. The products would turn GPU time into something traders can hedge, speculate on and price years ahead, assuming US regulators sign off.

Who is lining up, and what they would trade

The brokers named in the preparations include ADM Investor Services, BGC Group and Nanhua USA LLC. Their plan is to facilitate trading in compute futures once the products clear regulatory review.

The contracts themselves are cash-settled. That means nobody takes delivery of a pallet of graphics cards when a contract expires. Instead, money changes hands based on where a price index lands.

The underlying price is the hourly rental cost of Nvidia GPUs, primarily the H100 and B200 models.

CME Group is working with Silicon Data to bring cash-settled futures that track GPU rental prices. Each of those contracts represents 730 GPU-hours over a month.

That number is not random. It works out to one GPU running around the clock for a typical month, which makes the contract roughly equivalent to leasing a single chip full time.

CME is not alone. ICE is advancing contracts that reference the Ornn and NATIVX indices. Nodal Exchange has partnered with Compute Desk.

The regulatory timeline

The CFTC issued a formal request for public comment on August 19, 2026. The comment window is scheduled to close on October 20, 2026.

The review tied to CME Group’s product launch has been extended to November 9, 2026. So the earliest decision point sits a few weeks after public feedback wraps up.

Meanwhile, proprietary trading firms, including Jump Trading and DRW, are securing capital and staffing dedicated desks in anticipation of regulated compute futures.

Why anyone wants to hedge a GPU

AI developers and cloud providers face uncertainty with compute. GPU time is the primary input for training and operating AI models, and its price can swing as supply tightens or new hardware floods the market.

A standardized futures contract would let these businesses stabilize that cost. A startup planning a large training run next spring could lock in a rental rate now, rather than hoping prices cooperate later.

On the other side, cloud providers holding large GPU fleets could hedge against rental rates falling. That matters when the hardware sitting in a data center depreciates with every new chip release.

Boston Consulting Group has projected the AI compute market could grow from $360 billion in 2025 to approximately $2.3 trillion by 2030.

What this means for traders and the AI industry

CME, ICE and Nodal Exchange are each backing a different index. Whichever benchmark attracts the most liquidity early could become the reference price for GPU time.

That gives the index providers — Silicon Data, Ornn, NATIVX and Compute Desk — an unusual amount of influence. A rental price index that underpins billions in contracts becomes a piece of financial infrastructure.

For brokers like ADM Investor Services, BGC Group and Nanhua USA, the bet is about being first in line. If the CFTC approves the products, firms with clearing relationships and client lists already in place can start routing orders on day one.

There are real risks to watch. Index-based contracts are only as sound as the data feeding them, and GPU rental pricing is fragmented across providers, contract lengths and chip models. The focus on H100 and B200 models raises a practical question: how contracts will handle the arrival of newer hardware that could make today’s benchmark chips less relevant.

The dates to circle are October 20, 2026, when the CFTC comment period is set to close, and November 9, 2026, the extended deadline for the CME review.