SkyPilot started as an open-source project at UC Berkeley's Sky Computing Lab, where our founders ran into a problem: cloud providers donated compute to the lab, but most researchers had a hard time migrating to new providers. Valuable compute went unused.

So we built SkyPilot. Instead of forcing developers to pick one cloud, what if we unified all of them into a single pool — a “sky” of compute?

Today, AI compute has become even more fragmented. Every AI team has to find compute anywhere they can. Many clusters, across neoclouds and hyperscalers. Reservations, mixed with bursty resources. Kubernetes. Slurm. VMs. Many accelerator generations. The problem we set out to solve became an industry-wide challenge.

SkyPilot turns fragmented compute into one AI supercomputer. With one platform to manage all AI compute and frontier workloads, AI teams can make the best use of their infrastructure — turning compute into intelligence, faster.