The San Francisco Compute Company

The San Francisco Compute Company runs large-scale, vetted GPU clusters, allowing customers to rent capacity ranging from 800 H100s for an hour to 8 H200s dedicated for a year, with no long-term contract lock-in since unused capacity can be sold back at any time. The company operates its clusters from UEFI on up and provides support comparable to a traditional cloud provider, along with a CLI for buying, selling, and managing GPU instances. Because hardware failure rates are much higher on GPU clusters than web servers, the company is built to detect and repack around failed nodes and refunds customers for failed-node downtime.