CFTC Clears Path for AI Compute Futures; CME Eyes October 5 Launch

CFTC Clears Path for AI Compute Futures; CME Eyes October 5 Launch

August 22, 2026

WASHINGTON, August 22, 2026, 08:21 EDT

  • The CFTC wants input on how to regulate derivatives based on computing power.
  • CME set to launch H100 and B200 rental-index futures on October 5, pending review.
  • Comments have to be in by October 20. There’s no new contract yet.

The U.S. Commodity Futures Trading Commission is reviewing derivatives linked to computing power. Exchanges are working on contracts meant to hedge swings in AI infrastructure costs. The CFTC posted its formal request Friday and set an October 20 deadline for comments.

The shift is key as advanced chip access has become a big operating expense. But a unit of compute isn’t like a barrel of oil—prices jump around depending on chip model, geography, memory, software, and network hookups.

The benchmark is the real product here. A futures contract only shields the buyer if the index matches the actual capacity in use. If not, the hedge and the charge might not move together.

CFTC Chair Michael Selig said the US “cannot win the AI race without a robust derivatives market for compute.” He said this review is a first move to clarify the rules. The request asks about market size, manipulation, customer protections and perpetual futures. CFTC statement

Planned contractReferenceFormatTimingStatus
CME/Silicon Data H100Hourly Nvidia (NASDAQ:NVDA) H100 rental indexOne month of rent; NYMEXOctober 5, 2026Pending regulatory review
CME/Silicon Data B200Hourly Nvidia B200 rental indexOne month of rent; NYMEXOctober 5, 2026Pending regulatory review
ICE/NATIVXCOIL energy-normalized compute indexU.S. dollar, cash-settledLater in 2026Pending regulatory review
ICE/OrnnOrnn Compute Price Index across GPU typesNot disclosedNot disclosedPending regulatory review

CME Group (NASDAQ:CME) says it will launch two new monthly rental-index futures using Silicon Data. The contracts are set to follow hourly rental prices on Nvidia’s H100 and B200 GPUs, with the symbols GPU1 and GPU2. Customer testing is set for August 30.

CME executive Pete Keavey called compute “the currency of the AI age.” He said futures could give companies tools to manage GPU costs and help with planning. The launch is set for October 5, but it still needs a regulatory review.

Intercontinental Exchange (NYSE:ICE) is working on two separate models. One approach, with NATIVX, adjusts compute based on energy use and network. The other uses Ornn’s pricing, covering several GPU types.

CFTC review areaCore questionWhy it matters
Cash-market depthHow much trading and liquidity is in compute?Thin trading can swing prices around.
Reference pricesIs it possible to check an index independently?If a benchmark fails, settlement can be unreliable.
ConcentrationDo a small number of providers set price or supply?Fewer suppliers could make it easier to move prices.
SettlementShould these contracts deliver compute capacity or just pay cash?Physically delivering capacity can be tough to manage.
Customer protectionWhat protections or disclosures will customers get?New users might not realize how risky derivatives are.
Perpetual futuresWhat’s the approach for contracts with no set expiry?Leverage and funding can make things more complicated.

The CFTC said compute trading is still fragmented and opaque, often happening bilaterally. The agency expects early contracts will likely be cash-settled. Getting the exact chips, software, and network access delivered would be tough.

It’s not just data-center operators who would use this. AI developers might look to hedge a future rental bill. Capacity providers could limit exposure to lower prices. Traders could get a new way to bet on compute demand.

DateDevelopmentRegulatory position
May 19ICE and Ornn say they’re working on GPU compute futuresNeeds regulatory review
July 1ICE and NATIVX roll out energy-normalized contractsNeeds regulatory review
August 11CME gives specifics on H100 and B200 rental futuresNeeds regulatory review
August 19CFTC puts out compute-derivatives requestPublic comment starts
August 21Request published in the Federal RegisterOfficial notice out
August 30CME has its customer testing setTesting only
October 5CME aims for launchWaiting on review
October 20CFTC shuts comment windowNo final rule given

The consultation isn’t a green light. It also doesn’t decide if a single benchmark can cover a market divided by hardware age and location. Those splits could bring major basis risk for people using it.

Risks: Compute indices face risks tied to thin quoting, heavy reliance on a few data providers, and fast-moving hardware cycles. Using leverage can make losses bigger. Even if a contract trades well, it may not match a buyer’s real cloud costs.

The CFTC is now asking exchanges, providers and customers for evidence. Their responses will decide if pricing GPU-hours can work as a real commodity market. For AI buyers, that could change how budgets for future capacity are set.

BEZ KABLI • EXTENDED COVERAGE

Further analysis

Has the CFTC approved AI compute futures?
No. The CFTC has opened a public consultation on how compute derivatives should work and be supervised. Comments are due October 20, 2026. CME's planned October 5 launch remains subject to regulatory review.
What would the first CME compute futures track?
CME plans separate contracts tied to Silicon Data's hourly rental-price indices for Nvidia H100 and B200 GPUs. Each contract represents one month of rent. The trading codes are GPU1 and GPU2.
Who could use these contracts?
AI developers and other large compute buyers could hedge part of a future rental bill. Capacity providers could hedge falling prices. The protection will be imperfect when a user's actual hardware, region or service terms differ from the index.
Why is compute harder to standardize than oil or wheat?
Compute prices depend on chip generation, memory, location, availability, software and network links. The market is also fragmented and often bilateral. That makes reliable benchmarks and physical delivery difficult, so the CFTC expects many early contracts to settle in cash.
What are the main risks?
Sparse price quotes and provider concentration could weaken an index. Hardware can also become obsolete quickly. Futures add leverage, and a contract may not move with a buyer's actual cloud bill.

Artur Ślesik

Artur Ślesik is a technology and financial markets journalist at Bez-kabli.pl, covering artificial intelligence, semiconductors, technology stocks and emerging innovations. A graduate of Warsaw University of Technology, he combines a technical background with market analysis to explain how new technologies are shaping industries, businesses and investment trends worldwide.