Startup Silicon Data, in partnership with CME Group, is aiming to transform GPU computing power into a tradable, standardized commodity, pioneering the world's first AI compute futures market.
Previously, on May 12th, CME and Silicon Data announced the development of compute futures, designed to provide AI developers and financial institutions with a tool to hedge against price volatility in computing power. This product will be priced based on a GPU benchmark index compiled by Silicon Data and is currently awaiting regulatory approval.
Within days of Silicon Data and CME Group announcing their collaboration, asset managers including ProShares and Rex Shares filed applications for ETFs linked to the proposed contract, including leveraged and inverse products.
Carmen Li, founder and CEO of Silicon Data, anticipates that as AI's energy consumption eventually surpasses all other uses combined, the compute futures market will "exceed" the size of the oil futures market. Whether this vision becomes reality will first be tested by regulatory attitudes.
On core issues such as the standardization of computing power, contract specification definition, and benchmark price construction, the U.S. Commodity Futures Trading Commission (CFTC) will conduct a rigorous review. Seoyoung Kim, a finance professor at Santa Clara University, noted that the CFTC will want a clear understanding of what the product actually is.
The Logic of Jet Fuel Applied to Computing
This concept originates from a straightforward analogy: the dependence of AI companies on computing power mirrors that of airlines on jet fuel.
Currently, most companies do not own the high-end Graphics Processing Units (GPUs) needed to run AI systems. Instead, they rent them on-demand from cloud service providers and emerging "neocloud" platforms.
As demand for AI infrastructure surges, the volatility in compute rental costs has intensified, making it difficult for businesses to forecast expenses effectively. Seoyoung Kim, the finance professor, explained that we are currently in a period of high uncertainty where many companies don't know how much computing power they will need next year, suppliers don't know how many GPUs to order, and even manufacturers like Nvidia are unsure how many to produce.
It is within this context that Silicon Data hopes to enable the hedging and management of compute cost risks through futures contracts, similar to how airlines hedge fuel costs or farmers hedge agricultural prices.
As with all futures markets, compute contracts will attract not only corporate users seeking to lock in costs but also speculators—traders who have a view on compute price movements but have no actual GPU needs themselves.
Proponents argue that speculators play a crucial role in enhancing market liquidity and improving price discovery. Critics, however, worry that speculative activity could amplify price swings, detaching prices from the fundamentals of real supply and demand.
Carmen Li of Silicon Data takes an open view on this, stating that speculators are also a vital part of the ecosystem. She notes that a market needs natural hedgers, market makers, and speculators. It's entirely reasonable, she says, for speculators to have their own market views and wish to express them.
Benchmark Index: Standardization is the Central Challenge
The operation of a futures market depends on a credible and uniform benchmark price.
Silicon Data has already constructed a GPU price index system that tracks the hourly rental prices for specific chips across different service providers. Its goal is to replicate the benchmark status that West Texas Intermediate (WTI) crude oil holds in the energy derivatives market.
However, standardizing computing power is far more complex than standardizing crude oil. According to Silicon Data, Nvidia's H100 chip alone has over 50 different configurations, with prices varying significantly based on processor specifications, memory, network bandwidth, utilization rates, and data center location.
Carmen Li explained that they normalize the prices entered into their platform daily to a standard base H100 configuration, describing it as a very complex normalization step that must be completed even before the index calculation itself begins.
Seoyoung Kim pointed out that standardization is a common challenge for all futures markets. Corn futures contracts specify the exact grades that can be delivered; similarly, the compute market must precisely define what is being traded.
The CFTC will conduct a comprehensive review of the contract specifications, settlement procedures, and the methodology for constructing the benchmark index. This review represents the critical hurdle for the contract's final approval.
Silicon Data's benchmark data has already begun to appear in significant corporate documents. SpaceX cited the company's GPU rental price data in its IPO filing, lending credibility to the firm's standing within the industry.
Comments