The San Francisco Compute Company, known as SF Compute, was founded in 2023 by CEO Evan Conrad and co-founder Alex Gajewski. The company tackles a structural inefficiency in the AI economy: GPU capacity is typically sold through long-term reservations and opaque contracts, which leaves enormous amounts of compute idle while other teams are starved for it. SF Compute's answer is a financial-style marketplace where compute is traded in real time at prices set by supply and demand.
Through the marketplace, buyers can purchase short-duration access to clusters of high-end NVIDIA GPUs, such as H100 and H200, paying market rates rather than committing to multi-year deals. Sellers, including data centers and companies with spare capacity, can list their hardware and monetize unused cycles. The platform manages the matching, scheduling, and settlement, effectively turning GPU time into a liquid, spot-priced commodity and giving the market a transparent reference price.
This approach is particularly valuable for training and large batch workloads, where teams need a burst of capacity for a defined window and would otherwise overpay for idle reserved hardware. SF Compute says it manages more than $100 million in hardware spanning several thousand GPUs, and it has signaled plans to add newer chips like the B300 to its listings as they become available.
SF Compute raised an early $12 million round and then a $40 million Series A that valued the business at around $300 million, co-led by DCVC and Wing Venture Capital, with participation from existing investors Electric Capital and Alt Capital, plus others such as Soma Capital and Factorial Capital. Total funding is just over $50 million. The team has also recruited infrastructure veterans, including a former Voltage Park CEO as CTO and a Lambda and Sun Microsystems alumnus as chief business officer.
By treating compute as a tradable commodity, SF Compute occupies a distinct niche among neoclouds and GPU marketplaces. It is best suited to teams that want flexible, transparent, spot-priced access to large GPU clusters rather than fixed long-term contracts.