A small team of former trading-floor engineers wants to treat computing power like pork bellies.
Computable, a two-person startup backed by Y Combinator, opened a secondary market for GPU rentals this summer. The platform lets artificial intelligence companies buy and resell dedicated compute time on NVIDIA H100 chips, with prices listed weeks and months out like airline seats or hotel rooms. If your model training wraps up early, you can offload the remaining reservation. If demand spikes, you buy someone else's slot.
The mechanics are straightforward. Buyers reserve full eight-GPU nodes in weekly blocks. Prices were shown from $2.56 per GPU-hour for slots extending into early 2027, at least according to the company's live market page in late August, though more recent snapshots show rates around $2.80 per GPU-hour. Each reservation delivers root access to a single-tenant machine within an hour, equipped with NVLink interconnect and running Ubuntu. On-demand rentals cost slightly more, billed per second.
What distinguishes Computable from standard GPU rental services like Vast.ai or RunPod is the resale feature. "We put a public price on every week," Ray Song, one of the co-founders, explained in the company's launch announcement. The market grid displays buy and sell buttons for each calendar block. Computable also posts standing buyback quotes, creating at least the appearance of liquidity in a market historically defined by long-term contracts and capacity scarcity.
Song and a second co-founder come from trading backgrounds at Jump Trading, Coinbase, and Citadel Securities. Their approach treats computing infrastructure less like a utility purchase and more like a tradable commodity. The startup even offers cash-settled floor rates on a proprietary GPU rental index, allowing data centers and AI teams to hedge against falling rates. Those derivative products remain restricted to eligible contract participants.
Betting on Scarcity
The timing reflects broader anxiety in the AI industry. Cloud providers have started pushing compute contracts that stretch five years, according to a report in The Information. Startups face a worsening shortage of high-end chips, and many have little choice but to lock in capacity they may not need.
Computable's market arrived just as larger exchanges began circling the same opportunity. CME Group and a data provider called Silicon Data announced plans in August to list futures contracts tied to H100 and B200 rental rates, with an October launch pending regulatory approval. Intercontinental Exchange revealed a separate futures product in May, partnering with a firm called Ornn to settle contracts against the Ornn Compute Price Index.
"The market for compute is in desperate need of a globally accepted pricing mechanism and risk management tool," Trabue Bland, a senior vice president at ICE, said at the time. Other startups, including Liquid Compute and Compute Exchange, are building platforms for institutional buyers to trade forward capacity and reserved blocks. None have disclosed meaningful volume figures.
Computable has not publicly disclosed detailed trade data or the full methodology of its hedge-product index. The company participated in Y Combinator's most recent cohort and surfaced publicly with a post on Hacker News in July, drawing attention from developers frustrated by opaque GPU pricing. Song appeared as a speaker at an Energy Investors Forum event in late July, listed simply as "Co-Founder, Computable."
The company operates nodes in North America and wipes disks at teardown. Every reservation in what the startup calls its "Genesis Auction" is backed by physical hardware, according to the platform's delivery page. Whether that hardware proves sufficient as the market scales, or whether buyers will trust a derivatives layer built on top of it, remains an open question. For now, Computable offers something scarce in the GPU world: a way out.
