Bernstein’s View on Compute as a Commodity
Bernstein analysts Gautam Chhugani and Madison Rezaei contend that the market for computational power – the data‑center processing, hardware and infrastructure required to train and deploy AI models – is evolving in the same way electricity markets did, moving from a non‑tradeable resource to a standardized, hedgeable commodity. They note that GPU hours cannot be stored, capacity varies by location, chip generation and tenant, and therefore forward curves will reflect expected scarcity rather than storage costs, making hedging essential.
Recent Multi‑Billion AI Compute Deals
The analysts cite several high‑value agreements that illustrate the scale of the AI capex cycle, which they describe as a trillion‑dollar market. The New York Times reported that Meta Platforms is in early talks with Anthropic for a compute supply arrangement valued at roughly $10 billion. The Wall Street Journal disclosed that SpaceX is negotiating a compute contract with the U.S. Department of Defense, and that SpaceX recently signed a cloud services agreement with Alphabet’s Google worth $920 million per month.
Exchange‑Level Initiatives
Earlier in the year, CME Group and Intercontinental Exchange announced plans to launch tradable futures contracts linked to spot prices for graphics‑processing‑unit (GPU) compute. These contracts are pending regulatory review but would provide market participants with a mechanism to hedge price risk in the emerging compute market.
Proposed Benchmarking Framework
Bernstein outlines a three‑step process to create a reference index for compute:
1. Define a single unit of compute – they suggest using a specific configuration such as the NVIDIA H100 SXM 80 GB, which includes the server, memory and bandwidth.
2. Specify tier and lease characteristics – differentiate between neocloud providers and traditional hyperscalers, and between on‑demand and committed‑term leases.
3. Normalize for region and configuration – apply adjustments to account for geographic and technical variations.
The resulting standardized data would serve as a reference index, with forward curves derived from observed market prices.
Implications for Market Structure
According to the analysts, the creation of benchmarks and distribution mechanisms is the key driver for the growth of compute markets, mirroring the evolution of electricity markets where standardized hubs and reference prices transformed procurement into a financial market. They anticipate both cash‑settled and physically delivered forward contracts, as well as structured financial products built on the underlying compute asset.