AI Compute Is About to Become a Tradable Commodity
CME Group is preparing to launch futures contracts tied to GPU rental prices, opening the door to a new financial market for the AI boom
Illustration: Rafael Minguet Delgado via pexels.com
CME Group plans to launch compute futures contracts on October 5, pending regulatory approval, allowing companies and investors to trade contracts tied to the price of AI computing capacity. The contracts will initially be based on rental prices for Nvidia’s H100 and B200 GPUs.
The development was highlighted by Chamath Palihapitiya, the venture capitalist and entrepreneur, in a new deep dive published on his website. Palihapitiya argues that as AI infrastructure spending reaches enormous scale, compute is beginning to resemble commodities such as oil – a critical resource whose price can have a major impact on the businesses that depend on it.
Today, companies building and operating AI infrastructure have few ways to protect themselves against swings in compute prices. GPU rental costs can spike when demand surges, while older hardware can rapidly lose value when a faster generation arrives. Data centers can take years and billions of dollars to build, making future compute prices a significant financial risk.
But turning compute into a standardized commodity won't be simple.
Silicon Data, whose price index will underpin CME’s contracts, tested identical workloads across 3,500 GPUs at 11 cloud providers and found significant performance differences between GPUs of the same model. H100 performance varied by as much as 34.5% in one test.
That raises a fundamental question for the emerging market: what exactly are you trading if one GPU isn't necessarily equivalent to another?
If that problem can be solved, Palihapitiya argues, compute could develop into a major new asset class – potentially reaching trillions of dollars in value.