As demand grows for ways to hedge the costs of artificial intelligence, exchanges have begun exploring how to turn computing power into a tradable commodity. In August, CME announced a partnership with Silicon Data. Likewise, Intercontinental Exchange joined with Ornn in May as they seek to establish a benchmark for this nascent market. The contracts are under regulatory review.
Jeff Sprecher, chair and chief executive of ICE, said futures markets normally develop around an established spot market, something that does not yet fully exist for compute.
“My business is to do futures, and it’s very hard to do futures if there isn’t an underlying spot market,” Sprecher said during the Exchange Leaders: The View from the Top panel at FIA’s Futures & Options Expo in Chicago on 3 November.
Unlike commodities such as oil, compute is not easily interchangeable. Sprecher said companies building applications around one cloud provider cannot necessarily shift their workloads elsewhere simply because computing power is cheaper there.
“We’re hooked to all the major platforms, and it is not easy to move business from AWS to Azure once you start building in one of the clouds or picking one of the models, in the case of AI. They are not that portable where you can just say, oh, let's go over here and we'll use this compute and that compute,” he said.
There have been markets where futures have developed before a mature spot market, he said, but companies that buy large amounts of computing power would ultimately need to use the contracts as a hedge. “So I’m bullish but cautious,” he said.
Kalshi, meanwhile, has experimented with prediction markets designed to produce forward-looking prices for compute.
Lynne Fitzpatrick, president and chief financial officer of CME Group, said clients are already looking for ways to manage the exposure.
“It’s an asset class where we’re getting a lot of questions from clients, and they’re looking for a way to hedge this exposure because, as Jeff said, the markets just don’t exist for them,” she said.
The potential users range from corporates and banks to hyperscalers and chip producers, she said. Establishing a reliable benchmark is critical if the contracts are to reflect the costs those users actually face.
“The important part is: what is the underlying index that you’re pricing these on and is it broad enough that you are getting enough inputs to make it really, really reflective of the underlying market?” Fitzpatrick said.
More than one benchmark could ultimately develop, she added, pointing to commodity markets where different indices coexist to meet different needs.
Underscoring the incipient nature of the markets, Sprecher noted how the eventual market may need to go further than simply capturing the rental price of an individual GPU.
The overall cost of compute incorporates other inputs, including electricity and memory, he said, noting that exchanges already operate futures markets in power, natural gas and carbon emissions. “What people really care about is the total cost of compute,” he said.
That could eventually create demand for combinations of different products rather than a single benchmark. “It’s a basket, and I think it’ll trade like a basket,” Sprecher said.
Kalshi has tested a different approach in establishing a forward price for computing capacity.
Luana Lopes Lara, co-founder and chief operating officer, said Kalshi had launched prediction markets on the price of compute and found that the forward-looking curve implied by the contracts generated positive feedback. The company subsequently delisted the markets while it worked with regulators.
She said questions included how to ensure the underlying indices could not be manipulated. Kalshi remains interested in the market through prediction contracts as well as data, futures and perpetuals.
“We think it’s going to be an absolutely massive, massive market,” Lopes Lara said.
For the exchanges, the challenge now is not simply designing a contract. They need benchmarks businesses trust, enough liquidity for meaningful price discovery and participation from companies with genuine exposure to the underlying cost.
Ultimately, the test will be whether the companies actually paying for compute use the market.
“It would have to be the hyperscalers and the people that are directly paying for compute [that] would have to get into those markets to try to hedge,” Sprecher said.
Read an Explainer on Compute futures.
