The Chicago Mercantile Exchange (CME) Group plans to launch futures contracts based on computing power, marking a significant development in the emerging market for artificial intelligence (AI) infrastructure. The new product, announced in August, aims to position compute as a standardized, tradeable commodity, reflecting its growing importance in the AI-driven economy.
Pete Keavey, a CME executive overseeing the initiative, compared compute to oil’s role in the 20th-century economy, suggesting that compute futures could evolve similarly from spot pricing to a global derivatives market. Industry leaders share this optimism, with BlackRock CEO Larry Fink projecting that compute futures could become a new asset class.
The move is underpinned by projections from Boston Consulting Group estimating the AI compute market to expand from $360 billion in 2025 to approximately $2.3 trillion by 2030. Capturing even a fraction of this volume would represent a lucrative avenue for the CME Group.
However, launching a successful futures market involves overcoming significant challenges. Historical data shows that many new futures contracts—between two-thirds and three-quarters—fail to generate sustainable trading volume. For compute, the market must meet key criteria including sufficient price volatility, standardization of contracts, and a broad base of active participants.
Price volatility is critical for futures trading to be attractive for hedgers and speculators alike. The AI compute market exhibits such volatility, driven by supply constraints and complex logistics. For example, Oracle’s Project Jupiter data center, scheduled for completion in 2028, has faced delays related to natural gas pipelines and environmental permits. More immediately, rental rates for Nvidia’s H100 GPUs fluctuated significantly, spiking to $8 per hour at the start of 2024 before falling below $2 by late 2025. These swings could motivate market participants to seek hedging instruments.
Standardization presents another hurdle. Drawing lessons from the launch of oil futures tied to West Texas Intermediate crude, CME Group has opted to reference Nvidia’s H100 and B200 GPU models for its contracts. These choices aim to establish definitive benchmarks, with Silicon Data reporting rental rates of $2.77 per hour for the H100 and $5.86 for the B200. Contracts will extend up to 36 months into the future.
Nonetheless, GPU performance varies depending on factors such as cluster configuration, networking, software, and geographic location. Rapid technological turnover also complicates standardization, necessitating separate contracts for distinct chip models. Moreover, rental rate benchmarks differ between index providers—Silicon Data and Ornn, for instance, maintain separate methodologies that do not always align. Without consolidated benchmarks, the derivatives market risks slow adoption.
Finally, a successful futures market requires a diverse ecosystem of buyers and sellers. By contrast, the compute market remains concentrated: Nvidia dominates chip production, a small group of hyperscale cloud providers control much of the capacity, and a handful of AI research laboratories account for significant demand. This concentration could limit market liquidity and raise concerns about potential manipulation.
As the CME prepares to list these compute futures, market observers are closely watching the development. The contracts could offer much-needed transparency and price discovery in the rapidly evolving AI sector. Yet, some analysts caution that the market may prove more important as an indicator of AI infrastructure economics than as a fully functioning trading venue, drawing parallels to the ABX mortgage-backed securities index, which both illuminated and exacerbated risks during the subprime mortgage crisis. How the compute futures market evolves may have far-reaching implications for investors, technology providers, and the future trajectory of the AI economy.