Skip to main content
TS

The San Francisco Compute Company

SF Compute builds and operates large-scale NVIDIA GPU clusters for AI training and inference, selling compute time on flexible contracts via a real-time marketplace.

The San Francisco Compute Company, known as SF Compute, is a vertically integrated supercomputing firm that builds and operates large-scale GPU clusters for AI training and inference workloads. Headquartered in San Francisco, the company designs its own infrastructure and full-stack systems software, operating from UEFI upward, rather than reselling capacity from third-party providers. It employs roughly 45 engineers and has raised $40m in Series A funding.

SF Compute sells time on clusters of NVIDIA H100s, H200s and B300s under flexible contracts, with H100 compute starting at $1.95 per GPU per hour and no long-term commitments or reseller markups. Customers can exit contracts by selling unused capacity back through a real-time marketplace, on which compute contracts are traded between buyers and sellers. The model is intended to give startups and enterprises access to supercomputing resources without lock-in.

Support is provided 24/7 through dedicated Slack channels. The company describes its operations culture as vertically integrated, building and running its own infrastructure, with an engineering focus spanning hardware-level firmware through cluster and marketplace systems. It states an ambition to build what it considers the defining infrastructure marketplace for the AI era.

1 Open job

We use cookies

We use cookies to understand how this board is used and to improve it. Analytics run only if you accept. Cookie Policy