Computable

What it does

Computable operates a marketplace for buying and selling GPU compute allocated by calendar week. Buyers can purchase access to specific GPU nodes for exact time periods rather than committing to fixed-term leases. The platform aims to enable price discovery and resale of compute capacity, which currently happens through private bilateral negotiations.

The company also publishes the Computable GPU Index, an open-source price reference for GPU-hours. The index collects on-demand rental rates every 15 minutes from a fixed panel of providers, aggregates them using an interquantile mean method, and publishes a single USD price per GPU-hour for major GPU models including H100, H200, B200, and B300.

Who it is for

The marketplace targets buyers and sellers of GPU capacity. Founders want the ability to rent compute for exact durations without long-term commitments. Resellers benefit from secondary market liquidity. The GPU Index serves anyone pricing or benchmarking compute: researchers, traders, financiers, and AI infrastructure companies.

Pricing

The site does not show prices for the marketplace itself. Access to the GPU Index is provided free and requires no authentication.

How it stands out

Computable identifies a structural problem: GPU compute historically trades only through 24-month private leases with no visible pricing or resale options. The same hardware rents at 2x spreads depending on the counterparty. Founders from Jump Trading and Coinbase recognize this resembles energy markets before deregulation.

The marketplace's week-based structure differentiates it from standard long-term leases. The GPU Index is distinguished by being fully transparent: methodology, data, and code are published openly. Anyone can reproduce the index value from the public record. The calculation uses an outlier-resistant voting mechanism where no small group of providers can manipulate the result.

What a founder should check

First, verify whether buyers actually prefer weekly allocations over monthly or longer terms. Existing lease durations may reflect genuine customer needs rather than supplier lock-in. Second, assess whether the index gains adoption as a reference rate. Commodity indices require critical mass of trading activity and widespread acceptance to matter; initial adoption barriers are real. Third, examine the incumbent resistance. Major GPU providers control their own pricing and may prefer bilateral negotiations that allow price discrimination over transparent benchmarks.