Problem

AI teams increasingly need scarce GPU capacity for specific future windows, but conventional rental marketplaces charge a premium for flexibility. That makes it expensive to reserve capacity ahead of time, while buying too much capacity creates idle spend and leaves customers exposed to sharp rate increases. Computable’s central use case is letting a team reserve only the GPU capacity and calendar weeks it needs for a future AI run or project—for example, locking January capacity in July before another H100 price jump—without paying for unused periods.

The problem also exists on the supply side: capacity that is not needed immediately is difficult to monetize efficiently, and awkwardly timed reservations can leave short, unsellable gaps in a provider’s calendar. Computable addresses this mismatch by treating GPU time as a tradable, date-specific asset rather than a simple hourly rental.

Product / Service

Computable is a marketplace for buying, selling, and redeeming GPU hours by calendar week. Customers choose the GPU SKU, number of GPUs or nodes, contiguous week window, and datacenter region they want, then participate in a sealed-bid, pay-as-bid auction. The allocation algorithm considers both bid value and the amount of future capacity a reservation would fragment, while winning customers pay the prices they submitted. After payment and verification, a customer receives remote SSH access to a virtual machine occupying an entire qualifying GPU node during the purchased weeks.

The company positions the product as wholesale-priced capacity with instant liquidity: customers can acquire only the weeks they need and seek to sell unused commitments rather than absorbing idle cost. The terms impose a 10% bid deposit and a 2% platform fee; they also clarify that a customer may request a sale back to Computable, but that repurchase is discretionary and no buyback price is guaranteed. The intended benefit is therefore better forward price certainty and more flexible capacity planning than standard rentals, with liquidity subject to the platform’s terms.

Market

Computable competes in the intersection of GPU cloud infrastructure, compute marketplaces, and marketplace finance. Its closest marketplace comparisons include Vast.ai and Spheron, while RunPod and Lambda represent specialized AI-cloud alternatives; hyperscalers such as AWS, Google Cloud, and Microsoft Azure are broader substitutes for purchasing GPU capacity. Computable differentiates around trading capacity across future calendar weeks and explicitly pricing the value of preserving a sellable inventory calendar, rather than simply offering on-demand GPU instances.

The company appears to be at an early launch stage rather than at demonstrated scale. Y Combinator lists it as an active Summer 2026 company in Marketplace Finance, Cloud Computing, Trading, and AI, with a two-person team in New York City. Its public launch described a first auction for GPU capacity covering August through January, with bidding open until July 31, 2026, and the site showed live listings and week-by-week availability. The evidence does not disclose revenue, customer counts, or completed-auction results, so traction is best characterized as an initial market launch with public listings and early visibility, not proven commercial scale.