Added
Updated
Computable
Computable runs a calendar-week exchange for dedicated H100 GPU compute.
Computable runs a calendar-week exchange for dedicated H100 GPU compute.
Computable's market lists H100 nodes sold by the calendar week with posted prices per week and a standing buyback quote on every position, so buyers pay for the weeks they use and turn unused capacity back into cash.
The company also sells cash-settled rate floors on GPU compute, priced from a single node upward against a published reference index, and publishes the Computable GPU Index as an open reference rate for the same market.
Computable operates a weekly GPU capacity market where buyers pay only for the weeks they need and sell unused weeks back at a posted quote. Dedicated H100 nodes list at published week-by-week prices from about 2.32 to 2.85 dollars per GPU-hour across the 2026-2027 season, with no flexibility premium and no multi-year commitment.
Listings are provider-attested rather than permissionless: capacity appears only after a provider contract and attestation, the provider guarantee travels with the claim, and nodes run with root access live within an hour of the window start.
GPU capacity is traded at commodity scale but priced like a one-off service. Providers quote on-demand rates that differ widely for identical hardware, and buyers sign multi-year leases they cannot resell, which leaves capacity idle and exposes both sides to sharp price moves.
Computable treats compute like energy markets before 2000 were treated, before visible spot prices and standardized contracts emerged there. The company publishes a reference price per GPU-hour, runs week-dated auctions with published clearing prices, and sells cash-settled rate floors so a buyer of a long GPU contract can hedge the downside.
Computable's advantage is liquidity in a market that had none. Long GPU leases lock tenants into weeks they may not use, while providers carry idle capacity, so both sides lose money on rigidity; Computable lets each sell unwanted weeks to the other instead of signing away the difference.
The company also publishes its Computable GPU Index as an open-source USD price per GPU-hour from a fixed provider panel, with methodology, data, and code public so any print can be reproduced. Competing rental marketplaces quote closed prices with a flexibility premium and no resale market, and its rate-floor hedging product settles against a disclosed reference index.
Computable sells GPU compute capacity by the calendar week. Its market lists dedicated H100 nodes with week-by-week prices through January 2027, lets buyers resell unused weeks at a posted quote, and supports locking future weeks at today's price.
The Computable GPU Index publishes an open-source USD price per GPU-hour, updated every fifteen minutes from a fixed panel of providers, and covers H100, H200, B200, and B300 SKUs. An MCP endpoint serves the live index to agents anonymously.
Rate floors are cash-settled puts on that reference index for Eligible Contract Participants, giving buyers of long GPU contracts a hedge against price declines.
Computable prices weekly H100 capacity on a published grid that runs from roughly 2.32 dollars per GPU-hour in the autumn weeks to 2.85 dollars per GPU-hour for late January 2027, with prices posted per week rather than negotiated.
Rate floors are priced by SKU, calendar months covered, floor level, and coverage size, with deeper floors and longer cover costing more; quotes are indicative and priced from a single node upward. Users pay up front for specific nodes on specific weeks with no leverage and no margin.
