China and U.S. exchanges race to launch AI futures

China’s biggest futures venue is moving to give AI tokens a price, while U.S. exchanges are racing to do the same for compute power. The shift marks a new phase in AI finance: the inputs behind model training and inference are starting to look less like technical overhead and more like commodities that can be hedged, priced and traded.

The Shanghai Futures Exchange is in the early stages of designing contracts tied to AI tokens, according to people familiar with the plans. In the United States, CME Group and Intercontinental Exchange are building parallel products aimed at the cost of compute, not the output of software. CME said it planned to launch a compute futures market later in 2026, pending regulatory review, in partnership with Silicon Data. ICE followed with plans for GPU compute futures based on Ornn’s Compute Price Index.

The commercial logic is straightforward. AI builders face volatile input costs as demand for chips, cloud capacity and inference surges, and a futures market would let them lock in prices ahead of time. That matters for cloud providers reserving capacity, enterprise users budgeting for AI deployments and infrastructure investors trying to understand where margins may widen or compress as usage rises.

China’s case is especially forceful because the scale of token demand has exploded. The National Data Administration said daily token consumption exceeded 140 trillion in March 2026, more than 40% above the end of 2025 and more than 1,000 times the 100 billion recorded at the start of 2024. Liu Liehong, who heads the agency, has framed that growth as evidence of a new value system for AI, with a shift from simple chatbots toward more agentic systems that consume far more compute.

That kind of volume is exactly what commodity markets need before they can financialize an input. Electricity futures, bandwidth contracts and other standardizable derivatives showed how markets can turn hard-to-store resources into tradeable benchmarks once the underlying unit is measurable and the contract design is clear. AI tokens and GPU compute may now be following that path.

The new market, however, brings fresh risks. A tradable benchmark can help users hedge price shocks, but it can also invite speculation, concentration and basis risk if the index drifts away from actual buying conditions in local data centers or cloud contracts. If AI infrastructure becomes a financial market, pricing power could move away from engineers and procurement teams toward exchanges, liquidity providers and the firms that control the benchmarks.