CME plans to launch H100 and B200 computing power futures on October 5th, but this is still pending regulatory review. The contracts trade not chip prices, but rather the hourly rental cost of GPUs as compiled by Silicon Data, with each contract corresponding to one month's rent.
BCG Research believes the short-term impact is the creation of a publicly available forward curve for computing power: AI companies can lock in costs, cloud service providers can hedge against declining rents, and banks can more accurately assess data center cash flow. This is expected to reduce financing costs, drive more investment in computing power, and make forward prices a new indicator of whether AI demand is cooling. H100 rents have fallen from approximately $8 per hour to $1.96, indicating that genuine hedging demand does exist.
However, it differs fundamentally from mortgage-backed securities in 2008: futures only transfer price risk and do not directly create credit risk. What is truly alarming is that if computing power futures further become the basis for pricing GPU mortgage loans, data center valuations, and securitized products, it could lower financing thresholds and amplify leverage; once rents fall, it could create a negative feedback loop through collateral devaluation, margin calls, and financing contraction. In addition, computing power is affected by region, network and service quality, and there is still a significant basis risk between the standardized index and the actual rent.
