For decades, the global economy has relied on futures markets to mitigate volatility. Airlines lock in fuel prices, farmers protect their crop yields, and manufacturers stabilize the cost of raw metals.
Now, a new venture aims to apply this same financial infrastructure to artificial intelligence.
Silicon Data, a firm specializing in tracking pricing across various GPU marketplaces and cloud providers, has partnered with CME Group to develop the world’s first futures contracts for AI computational power. Pending regulatory approval, these contracts would allow enterprises to hedge against the volatile costs associated with training and deploying AI models.
Institutional interest is already mounting. Shortly after the announcement, asset managers ProShares and Rex Shares filed proposals for exchange-traded funds (ETFs) linked to these proposed contracts, including inverse and leveraged products.
Founder and CEO Carmen Li believes this market could eventually eclipse some of the largest commodity markets in existence. In a recent interview, Li suggested that the market could grow larger than oil futures, predicting that the energy and compute demand driven by AI will eventually surpass all other energy uses combined.
Treating Compute as Essential Fuel
The concept is based on a straightforward analogy: AI companies are becoming as dependent on compute power as airlines are on jet fuel.
Because most firms do not own the high-end graphics processing units (GPUs) required for modern AI, they rent capacity from cloud giants and emerging “neocloud” providers. However, as demand spikes, pricing fluctuates wildly, making it nearly impossible for businesses to accurately forecast operational expenses.
“We are currently seeing a high point of uncertainty,” explained Seoyoung Kim, a finance professor at Santa Clara University. “Many companies are unsure of their compute needs for the coming year, suppliers are uncertain about the capacity they should order, and manufacturers like Nvidia are operating without a clear picture of exact production requirements.”
To solve this, Silicon Data has developed GPU price indexes that track hourly rental rates for specific chips. These benchmarks are intended to function like West Texas Intermediate (WTI) crude oil, serving as the standard underlying asset for derivatives.
Such a market requires both hedgers and speculators. Companies fearing rising costs would buy contracts for protection, while providers with excess capacity could sell contracts to shield themselves from price drops.
The utility of these benchmarks is already gaining corporate traction. For instance, SpaceX referenced Silicon Data’s GPU rental-rate data in its prospectus during its process to go public.
The Role of Speculation
Beyond corporate hedging, these contracts would naturally attract speculators—traders who do not need the actual GPU capacity but wish to bet on the direction of compute pricing.
While critics argue that speculation can decouple prices from actual demand and increase volatility, proponents suggest it is essential for providing liquidity and improving price discovery.
“Speculators are a vital part of the ecosystem,” Li noted. “You need natural hedgers, market makers, and speculators. Their ability to express an opinion on the market’s direction is perfectly acceptable.”
Li, a Harvard MBA, argued that traders with insights into supply-and-demand dynamics help establish a transparent price floor and ceiling for the entire industry. The recent ETF filings by ProShares and Rex Shares suggest that investors are already viewing AI compute as a distinct, tradable asset class rather than just a technical overhead cost.
The Challenge of Standardization
Unlike a barrel of oil, AI compute is not a uniform physical commodity. Silicon Data notes that Nvidia’s H100 chip alone has over 50 different configurations, with costs varying based on memory, networking, utilization rates, and the physical location of the data center.
For a futures market to be viable, participants must trust that a single benchmark accurately represents these diverse variables.
“We normalize the daily pricing data on our platform to a base H100 case,” Li explained. “It is a highly complex normalization process that occurs before the index is even calculated.”
Professor Kim noted that standardization has historically been the primary hurdle for futures markets. Just as corn futures specify a precise grade of grain, compute markets must define exactly what is being traded.
Kim added that the Commodity Futures Trading Commission (CFTC) will likely subject the product specifications, settlement procedures, and benchmark construction to rigorous scrutiny before the market can officially launch.