Computable

Anyone trying to price GPU compute today faces a market where every provider quotes a different rate and no agreed reference price exists. Existing GPU price indexes are closed black boxes, giving users a number without the underlying data, method, or code to verify it. This is unusual for an asset that is rented, resold, and financed at commodity scale, since mature commodity markets typically have a trusted reference rate. Traders, renters, and financers of compute are left without a verifiable benchmark to anchor contracts or decisions.

Computable GPU Index (CGI) is a USD price per GPU-hour computed from the published on-demand rental rates of a fixed panel of providers. Every 15 minutes the system collects published rates, screens them to on-demand rates only from the fixed panel, has each provider cast a weighted vote at its price, and averages the central band of those votes as an interquantile mean, so outlier votes in the outer thirds cannot drag the index. The index, method, and code are fully open: every data point is sourced and timestamped, every published value can be recomputed from the public record, and the collector and calculation can be cloned from a public GitHub repo. The index currently covers H100, H200, B200, and B300 GPUs, is viewable live with charts and history, and can be accessed by AI agents via an anonymous, read-only MCP endpoint with no key required.