SF Compute

SF Compute builds data centers, GPU clusters and cloud infrastructure for teams training and running AI models. It offers virtual machines, bare metal, managed Slurm and Kubernetes, with reservations ranging from one node for an hour to thousands of nodes for multiple years. Customers can buy capacity through its API or CLI and resell unused GPU time when their needs change. Its scheduler supports that resale market, while the Orderbook API exposes standing buy and sell orders so customers can inspect hardware availability and prices for a particular delivery window.

Founded in 2023 by Evan Conrad and Alex Gajewski, SF Compute is led by Conrad as CEO. The business grew from Conrad’s need for short-term compute: his then-company bought a GPU cluster for an earlier startup and subleased spare capacity. Alongside its infrastructure work, the company released Tiny Narrations, a dataset of 30,000 hours of synthetic story narration derived from TinyStories. It includes pre-tokenized audio for research into small audio models and multimodal scaling.

SF Compute connects GPU cluster vendors with customers who can commit to long contracts and resell portions of their capacity. The company charges a fee each time that compute is resold.