A new market for AI compute

AI is turning compute into a strategic resource, and the scramble to secure GPU capacity is starting to look a lot more like a commodity market than a traditional cloud-services business. As data-centre developers, lenders and energy companies try to price the next wave of AI demand, a new question is coming into focus: Can the industry build the kind of benchmark and hedging tools that already exist for oil, gas and power?

Host Ed Crooks is joined by Peter Keavey, Global Head of Energy and Environmental Products at CME Group, and Carmen Li, Founder and CEO of Silicon Data. Together, they explore the case for a futures market in GPU compute: a financial product designed to bring more transparency, liquidity and risk management to one of the fastest-growing corners of the AI economy.

Carmen explains how the market works today. Most users are not buying chips outright; they are renting access to GPU capacity by the hour, often through longer-term agreements with hyperscalers, neo-cloud providers and data-centre operators. That market is already large, global and increasingly active, but it remains fragmented and opaque, with prices varying by provider, chip type and contract structure, and much of the trading still happening through bilateral deals and requests for quotes.

Peter sets out the logic for moving from that over-the-counter world to an exchange-traded one. In his view, a GPU futures contract could do three things at once: reduce counterparty risk through central clearing, concentrate liquidity in a transparent order book, and create forward benchmark prices the wider market can use. The proposed product is financially settled against an index of spot prices, translating an hourly rental market into a standardised monthly contract that could eventually extend several years forward.

The bigger issue, though, is energy. Power is not the whole cost of GPU compute, but it is the most volatile variable input, which means a GPU hedge could eventually sit alongside gas and power hedges for data-centre operators, lenders and infrastructure investors. The discussion keeps returning to what that means for markets such as Texas and Virginia, where the AI build-out is already shaping decisions on generation, grid access and where capital should go next.

Both guests stress that this is still a young market, but already a volatile one. Rental rates have swung sharply as chip scarcity eases and then tightens again, while banks, traders and developers are trying to make long-dated decisions without a reliable forward curve. If this market develops the way Keavey and Li expect, GPU futures would not just serve traders: they could become an important signal for anyone trying to judge how durable the AI boom really is, and how much energy the system will need to support it.

The growth of AI and digital infrastructure is turning computing capacity into a critical business cost. To manage this, CME Group, in partnership with Silicon Data, is introducing Compute futures.

This new marketplace will provide the opportunity to manage volatile compute costs and treat processing power as a true, tradable commodity.

- Transparent benchmarking: Access real-time data to price GPU capacity accurately.

- Regulated trading: Execute strategies on a trusted, regulated exchange.

- Risk management: Hedge against price volatility to plan long-term data center growth with certainty.

Learn more.