Bernstein: AI Compute Market Is Becoming a Commodity, Paving Way for New Hedging Tools

The rapidly growing market for artificial intelligence compute power is evolving to resemble a commodity market, a development that could enable the creation of futures contracts and other tools for hedging risk, according to an analysis by Bernstein.

In a note published Friday, Bernstein analysts Gautam Chhugani and Madison Rezaei argued that the market for AI compute—the data center processing power, hardware, and infrastructure used to train and run AI models—is following a similar path to electricity.

The Electricity Parallel

Bernstein draws a direct comparison between AI compute and the power market. Like electricity, GPU processing hours cannot be stored for later use, and their value varies significantly based on location, the generation of the chip, and the provider.

"Electricity was once dismissed as untradeable as it cannot be stored, varies by location and quality - until standardized hubs and reference prices turned forward procurement into a financial market," the analysts wrote. They noted that because compute shares these physical constraints, its forward curve will reflect expected scarcity rather than storage costs, making hedging the only way to carry price risk.

Market Context and Demand

The need for such financial instruments is driven by the explosive demand for AI compute since late 2022. This has led to a surge in large-scale deals and a massive capital expenditure cycle.

Ad

• The New York Times recently reported that Meta Platforms was in talks to provide compute power to Anthropic in a deal potentially worth $10 billion.
• The Wall Street Journal reported that SpaceX is in discussions to provide compute to the U.S. Department of Defense.
• This follows a reported $920 million per month cloud deal SpaceX signed with Alphabet's Google.

In response to this growing spot market, major exchange operators like CME Group and Intercontinental Exchange have already announced plans to launch tradeable futures contracts tied to GPU compute prices, pending regulatory approval.

Creating a Benchmark

To facilitate a true financial market, Bernstein outlined a three-step process for creating a standardized benchmark for compute. The first, and most challenging, step is to define a standard unit of compute, as prices for the same chip can vary widely between traditional hyperscalers and specialized "neoclouds."

"The index provider needs to define the ‘unit’, for instance, NVIDIA H100 SXM 80GB (includes server, memory, bandwidth etc.)," the analysts explained. Once a unit is standardized, a reference index can be created, and forward curves can be derived from market observations, ultimately allowing for the development of structured financial products to manage risk in the trillion-dollar AI capex cycle.