GPU Price Index

A GPU price index measures the cost to rent GPU compute across cloud providers, expressed as a USD-per-GPU-per-hour rate. In 2026, several organizations publish competing indexes covering H100, B200, H200, A100, and other GPU models. Two of these - the Silicon Data index and the Ornn Compute Price Index (OCPI) - serve as the settlement benchmarks for CME and ICE compute futures contracts respectively.

Why GPU Price Indexes Matter

GPU rental prices are not published on a central exchange. Each provider sets its own on-demand rates, updated without notice. Reserved and committed pricing is negotiated bilaterally and rarely disclosed. This opacity creates a data problem: buyers cannot easily benchmark what they are paying, and sellers cannot easily price long-term commitments against a reference rate.

A standardized GPU price index solves this by aggregating prices across providers, applying a consistent methodology, and publishing a single daily number. That number then serves as a reference for contracts, derivatives settlement, internal cost accounting, and market analysis.

Major GPU Price Indexes in 2026

Silicon Data GPU Rental Price Index

Silicon Data’s index is the settlement benchmark for CME compute futures. It aggregates GPU rental pricing across major cloud providers, weighting by transaction volume. Key characteristics:

• Covers on-demand pricing - not reserved, committed, or negotiated rates
• Daily publication as USD/GPU/hour
• Includes hyperscaler vs. neocloud sub-indexes, capturing the price spread between AWS/Azure/GCP and CoreWeave/Lambda/Crusoe
• Proprietary index - not publicly available for free
• Used for CME H100 and B200 futures settlement

Ornn Compute Price Index (OCPI)

The OCPI is Ornn’s GPU rental price index, built on printed transactions only - actual executed deals rather than listed offers. Ornn argues this methodology produces a more accurate settlement price for institutional participants whose own contracts are based on executed prices. Key characteristics:

• Covers H100, H200, B200, B300, and RTX 5090 - broader model coverage than Silicon Data initially
• Daily publication via Bloomberg Terminal
• Used for ICE compute futures settlement
• Transaction-based construction - reduces influence of stale or withdrawn offer prices

Methodology Differences: What They Mean for Futures Traders

The core difference between Silicon Data and OCPI is the data input: Silicon Data includes listed offers; Ornn uses only executed transactions. In liquid markets, these converge. In thin markets - where few transactions occur but many offers are posted - they can diverge significantly.

For a participant hedging actual compute costs, the settlement index should match the pricing basis of the underlying exposure as closely as possible. If you buy compute at spot/on-demand rates from multiple providers, Silicon Data’s composite may be a better basis. If your contracts are priced on executed transactions, OCPI may be the better hedge.

Live GPU Price Index Data

ComputeWatcher is building a live GPU price index tracker - model-level daily prices, provider breakdowns, and historical charts for H100, B200, H200, A100, and MI300X. Subscribe to the weekly digest for launch notification.