Can AI Compute Become The Next Big Futures Market?
Start-up Silicon Data has partnered with CME Group to launch what could be the world's first AI compute futures market, pending CFTC regulatory approval. The market aims to help companies hedge against volatile GPU rental costs, similar to how airlines use oil futures. Key challenges include standardizing GPU pricing benchmarks and building sufficient market liquidity.
Summary
The video explores a novel financial product being developed by start-up Silicon Data in partnership with CME Group: a futures market based on the price of AI compute. Drawing a parallel to oil futures used by airlines to hedge fuel costs, the concept applies the same hedging logic to GPU rental pricing, which can swing dramatically and unpredictably.
Carmen Li, CEO of Silicon Data, explains that most AI companies do not own their GPU infrastructure outright but instead rent access through cloud providers and neo-cloud platforms. This creates significant cost volatility. While large enterprises can negotiate multi-year fixed-rate contracts with major hyperscalers like Amazon, Microsoft, and Google, doing so locks them into specific chips, providers, and prices — a potential innovation killer in a fast-moving industry. Smaller players lack this option entirely.
Silicon Data has built GPU price indices that track real-time hourly rental costs across providers, with the Nvidia H100 chip serving as the primary benchmark. The proposed futures market would allow compute users (long positions) to lock in future prices, while compute providers (short positions) can protect against price drops. Speculators would also participate, contributing to price discovery and liquidity.
A significant challenge is standardization. Unlike a barrel of oil, a 'GPU hour' is not uniform — Silicon Data has identified over 50 different configurations of the H100 chip alone, each trading at different prices. The company normalizes over 150,000 daily traded prices to create a comparable index. The CFTC must ultimately approve the product, determining contract size, trading times, settlement mechanisms, and whether settlement will be physical or financial.
Key Insights
- Carmen Li claims that compute will eventually surpass all energy combined as the largest human resource, suggesting the AI compute market could dwarf even oil futures in scale.
- Li argues that large enterprises locking into multi-year fixed-rate GPU contracts with hyperscalers gain cost predictability but lose flexibility — a trade-off she describes as a potential 'innovation killer' in a fast-moving industry.
- Silicon Data has identified more than 50 different configurations of the H100 chip alone, each trading at a different price, which the company must normalize before any index calculation can occur — a complexity that distinguishes GPU hours from traditional commodity benchmarks like a barrel of oil.
- A market expert notes that liquidity is not automatic just because AI is widely used, pointing out that most individuals and smaller companies do not consume compute at a scale that would require a futures contract.
- Li describes a three-sided ecosystem required for the futures market to function: natural hedgers on both the user and provider sides, market makers, and speculators — all of whom are necessary for price discovery and to signal market confidence in liquidity.
Topics
Full transcript available for MurmurCast members
Sign Up to Access