What is Computable GPU Index (CGI)?
Computable GPU Index (CGI) is an open-source reference price index that tracks the market cost of GPU compute in USD per GPU-hour. Designed to bring standard reference pricing to compute markets, CGI derives its figures from published on-demand rental rates gathered from a fixed panel of cloud providers. The calculation utilizes an interquantile mean over weighted votes, averaging the central band of rates while filtering out the outer thirds to remain resilient against anomalies, glitches, and provider tail manipulation. Every data point is sourced, timestamped, and made publicly available, allowing users to independently inspect and reproduce every published price print.
Computable GPU Index (CGI) key features
- Computes a USD reference price per GPU-hour every 15 minutes using on-demand rental rates from a fixed provider panel.
- Implements an interquantile mean voting model that excludes the outer thirds of quotes to maintain fault tolerance and resist outliers.
- Provides real-time pricing charts and historical records across multiple hardware SKUs, specifically H100, H200, B200, and B300.
- Delivers an open-source codebase and open public data records, enabling users to recompute and verify any historical print locally via Git.
- Features an open Model Context Protocol (MCP) interface that allows AI agents to access real-time compute data anonymously without an API key.
Who is Computable GPU Index (CGI) for?
Computable GPU Index is intended for market participants who trade, rent, or finance compute resources and require a transparent, standardized reference rate. It also supports developers and autonomous AI agents seeking programmatic, keyless access to verifiable GPU pricing data for infrastructure budgeting, contract negotiation, and automated workload routing.