The San Francisco Compute Company

When AI companies buy access to GPU clusters to train large-scale models, they risk getting locked into overpriced long-term contracts and paying for time they don’t need. To fix that problem, and to increase the overall liquidity and resilience of the burgeoning data center economy, SF Compute created an order book where buyers can buy and sell compute by the hour and sublease contracted but unused resources to others. The company doesn’t own GPU clusters — it manages them on behalf of their owners, collecting a small fee for each transaction while helping heavy GPU users reduce their financial exposure.