SF Compute is a vertically integrated supercomputing company that builds and operates large-scale GPU clusters for AI training and inference workloads. The company is based in San Francisco and employs about 45 engineers. It raised $40 million in Series A funding.
The company offers GPU cluster time on hardware including NVIDIA H100, H200, and B300 GPUs, with pricing starting at $1.95 per GPU per hour. Contracts are flexible: customers can exit by selling back unused capacity. SF Compute also operates a real-time compute marketplace where these contracts are traded, enabling both startups and enterprises to access supercomputing resources without long-term lock-in.
Technical work spans supercomputing, GPU cluster operations, and AI training and inference infrastructure. The company provides 24/7 support via dedicated Slack channels.
