Files

Polymarket joins Kalshi, CME, ICE, Architect and prospective exchange Liquid Compute in pursuing different ways to trade or establish future prices for AI compute. The push comes as the CFTC’s new request for public comment raises questions about whether today’s fragmented and often opaque compute market is sufficiently liquid, standardized and resistant to manipulation to support a mature derivatives market.

What AI compute markets are actually trading

AI compute refers to the processing capacity used to train and run artificial intelligence models. Much of the highest-value capacity today comes from GPUs, or graphics processing units, high-powered chips capable of performing huge numbers of calculations simultaneously. Nvidia‘s H100, H200 and newer B200 GPUs are among the chips most commonly used in large-scale AI infrastructure.

The emerging market is generally not about trading ownership of the chips themselves. AI companies, cloud providers and data-center operators buy or rent access to computing capacity, often priced by the GPU-hour, and the derivatives now being developed are designed to put a market price on what that access may cost in the future.

That could allow an AI company expecting to need large amounts of GPU capacity months from now to hedge against rising rental costs, while an infrastructure provider could manage the risk that future prices fall. The CFTC said in its request for comment that compute futures could aid risk management and price discovery and allow financial markets to “aggregate and reveal information about the future of the AI economy.”

Different exchanges are approaching that task in different ways. Conventional futures can track an index of GPU rental prices over time, while prediction market contracts can ask whether the price of a particular type of compute will be above or below a specified level on a future date. A series of contracts across different expiration dates can also be used to construct a forward curve, showing where traders collectively expect compute prices to be weeks or months ahead.

Selig wants US to set the compute benchmark

Selig put compute alongside prediction markets, crypto and other emerging products in his remarks at the CFTC’s Innovation Advisory Committee meeting Thursday, framing the initiative as “Winning the AI Race: Roadmap for Compute Market Dominance.” He tied the effort to the White House’s AI Action Plan, which calls for improving the financial market for compute to expand access to large-scale computing resources.

The CFTC is working with the Department of Commerce on that effort and plans to use feedback from its new request for comment to develop what Selig described as a “gold standard regulatory framework” for compute markets.

“Just as American markets helped establish the gold standard for trading the commodities that powered the industrial economy, we will do the same for the commodity that will power the intelligence economy,” Selig said in a separate statement Wednesday.

His Bloomberg comments a day later made clear that the objective extends beyond simply allowing exchanges to list new products. Selig said he wants the underlying price discovery and benchmarks established domestically, with compute trading on transparent U.S. markets and open order books.

“We’re working very hard together with Secretary (Howard) Lutnick and the Department of Commerce to set the standard here in the U.S.,” Selig said. “We want the best of the best, and we’re gonna lead in these markets.”

CFTC sees hurdles in building the market

For all of Selig’s enthusiasm, the CFTC’s request for comment makes clear that regulators still see major challenges in turning compute into a mature derivatives market. The agency said compute pricing remains fragmented and is often set through “opaque bilateral transactions,” limiting the amount of public price data available to exchanges and regulators.

The Commission also questioned whether compute yet has the fungibility, standardization and liquidity typically associated with commodities underlying futures markets. Prices can vary widely depending on the GPU model, provider, region and contract terms, while dominant suppliers may have enough pricing power to influence the cash market or a benchmark derived from it.

One of the RFC’s sharpest questions asks whether trading should be permitted in a derivative whose settlement price relies on data the CFTC “may not be able to observe, verify, or surveil.” The agency is also asking what protections would prevent a compute provider from influencing an index by changing posted rates, directing capacity to or away from a venue or choosing whether to execute transactions during a settlement window.

The CFTC is seeking input on those issues along with customer protections, market surveillance and perpetual compute futures. Comments are due Oct. 20, after which Selig has said the agency will continue developing its regulatory framework for the market.

Prediction markets move into compute pricing

Polymarket US’s new filings cover binary contracts tied to future values of AI compute indexes. One specifically certified contract asks whether the Ornn Data H100 SXM GPU Price Index will be at least $2.50 per GPU-hour on March 15, 2027. A wider class certification would allow Polymarket US to list similar contracts tied to H100, H200, A100, RTX 5090 and B200 GPU indexes. The broader filing also covers Ornn indexes measuring the price of AI-model usage by the token.

The contracts can ask whether an index will be above, below, at least, at most, between or exactly a specified value at a future point. Polymarket said the products may be listed beginning Aug. 24, although it remains to be seen whether Polymarket will launch them immediately or whether the CFTC will intervene while its compute comment process is underway. That possibility is not purely theoretical: in July, the CFTC stayed a CME self-certified 24/7 crude-oil contract while a related agency comment process was underway.

Kalshi has been pursuing a related strategy. In July, the exchange said it was using weekly and monthly prediction markets extending as far as a year out to construct a compute forward curve, giving traders a view of where GPU rental costs are expected to move. 

“We are using prediction markets to build the forward curve, which will provide the market a view of what compute costs will be in the future for different grades and time-frames of GPUs,” Kalshi Chief Risk Officer Udesh Jha told Bloomberg.

Jha said the curve could eventually support additional products, including futures and options. “It’s a key enabler for a lot of subsequent hedging, risk management and even speculative activities,” he said.

Exchanges race to price AI compute

Prediction markets are only one part of the emerging competition. CME Group plans to launch H100 and B200 rental-index futures on Oct. 5, pending regulatory review, using benchmarks from Silicon Data. ICE has announced GPU compute futures based on Ornn indexes and a separate partnership with NATIVX for energy-normalized compute futures. Architect Financial Technologies is building a U.S. exchange for futures and options tied to GPU rental costs and other AI supply-chain inputs.

Another prospective exchange has been built around that idea from the start. DeFi Rate reported in February on the startup then operating as Pluto, whose pending designated contract market and clearinghouse applications are publicly listed under the names PMEX Markets and PMEX Clearing. The company has since rebranded as Liquid Compute, which describes itself as the “financial layer for AI compute” and is currently inviting market participants to request access while its U.S. exchange applications remain pending.

“The aim of the exchange is to turn compute into a financial asset just like oil, gold, (or) other commodities,” CEO Ronit Jain told DeFi Rate at the time. Liquid Compute has since highlighted an H100 OTC forward trade involving Wintermute and said this week that it is “live and booking swaps now.” The company has not publicly identified the entity or jurisdiction through which those transactions are being conducted.

The race now is to build the transparent, trusted pricing infrastructure needed for a mature AI compute market. The CFTC wants that market centered in the U.S., but its new comment process shows regulators are still deciding what standards those prices and products will need to meet.