SkyPilot
Run AI workloads on any cloud — the intercloud broker from UC Berkeley.
About
SkyPilot is an open-source framework from UC Berkeley's Sky Computing Lab for running ML and AI workloads on any cloud with a single cloud-agnostic interface. It acts as an intelligent broker: given a workload's resource requirements, it finds the cheapest available GPUs across AWS, GCP, Azure, Kubernetes, and neoclouds like CoreWeave, Nebius, Lambda, and RunPod, with automatic failover across regions and clouds on capacity errors and managed spot-job recovery.
The project emerged from stealth as a company in July 2026 with a $20M seed round led by Lux Capital, backed by the CEOs of Databricks, Hugging Face, Vercel, and Replit, plus Google's Jeff Dean. It is Apache 2.0 licensed, BYOC (resources launch in your own cloud accounts), and widely used for training and batch inference. The honest caveat: it is an infrastructure broker, not an app platform — spot recovery restarts jobs from scratch unless your code checkpoints, and cost savings depend on your willingness to run multi-cloud.
What it can do
- One job spec runs across 20+ cloud and Kubernetes backends
- Automatic cheapest-GPU selection and cross-cloud failover
- Managed spot jobs with auto-recovery
- YAML task definitions, code sync, and job queueing
- BYOC model — resources stay in your own accounts
Setup
pip install skypilot, configure cloud credentials, launch with sky launch -t task.yaml.
Prerequisites
Python, cloud provider credentials (AWS/GCP/Azure or neoclouds).
Use cases
Works with
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 77; Trending score 88/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.