SkyPilot

SkyPilot is an open-source framework that runs AI workloads across any cloud, neocloud, or Kubernetes cluster, auto-provisioning the cheapest available GPUs; its 2026 SkyPilot Platform adds enterprise fleet management.

SkyPilot addresses the fragmentation of modern AI compute: teams juggle multiple clouds, neoclouds, Kubernetes clusters, and accelerator types to find capacity. SkyPilot abstracts these away, automatically selecting, provisioning, and managing compute and storage on whatever infrastructure is available and cheapest, so a single job definition can run anywhere. It handles spot-instance recovery, data sync, and multi-cloud orchestration for training, fine-tuning, RL, inference, and evaluations. The open-source project has been downloaded over 14 million times and is used by hundreds of organizations, including Nubank, Abridge, and H Company, to turn scattered resources into a unified 'AI supercomputer.' It supports development, agentic workloads, batch jobs, and multi-cluster production serving. In July 2026, SkyPilot launched from stealth with $20M in seed funding and introduced SkyPilot Platform, a unified AI compute platform built on the open-source core. The Platform adds large GPU fleet operations, standardized cluster management, workload orchestration, governance, and enterprise controls for frontier AI teams, while the framework itself remains free and open-source.