SkyPilot

SkyPilot is an open-source framework: it runs large language models (LLMs), AI, and batch jobs across any cloud, maximizing GPU availability and minimizing cost.

SkyPilot unifies your cloud infrastructure—Kubernetes, Slurm, VMs, and existing machines—into a single, optimized compute pool for AI workloads. The system automatically selects the cheapest resources across multiple regions and clouds (AWS, GCP, Azure, etc.), guaranteeing best-in-class GPU availability and absolute minimum cost without managed solution markups. Users deploy with a unified interface, maintaining control: all VMs and storage buckets remain securely within their cloud account. This architecture enables seamless scaling of services, like Qwen2.5-72B-Instruct, up to multiple replicas, all served via a single endpoint.

https://skypilot.co