CME Group plans to introduce compute as a tradeable commodity, calling it 'the new currency for the AI economy'. Pete Keavey, CME executive, stated this mirrors oil's evolution from spot trading to global derivatives. Larry Fink, BlackRock CEO, predicted 'a new asset class will be buying futures of compute'.
The AI compute market is projected to grow from $360bn in 2025 to $2.3tn by 2030. However, 70-75% of new futures contracts historically fail to sustain trading volume. For compute derivatives to succeed like oil, underlying prices must be volatile enough for hedging.
Nvidia's H100 chip rental rates spiked to $8/hour in early 2024, then fell below $2 by late 2025, demonstrating significant variance. CME, working with Silicon Data, chose Nvidia's H100 and B200 chips as contract references, with B200 renting for $5.86/hour versus H100's $2.77/hour. Contracts will extend 36 months, but GPU-hours aren't fully standardized due to performance variations across configurations, networking, software, and location.
Different benchmark providers like Orrn further complicate consensus.
Source: Financial Times Markets · Summarized by HeadlinesBriefing