SkyPilot

About SkyPilot

Skypilot.ai offers a platform that unifies fragmented AI computing infrastructure across multiple clouds, clusters, and hardware types into a single control layer. The company emerged from stealth on July 21, 2026, with a $20 million seed funding round led by Lux Capital, with participation from firms like Amplify Partners, Coatue Management, and Foundation Capital. The technology originated as an open-source project at UC Berkeley's Sky Computing Lab, where the founders identified challenges researchers faced in migrating workloads between different cloud providers.

The company was founded by a team of UC Berkeley researchers—Zongheng Yang (CEO), Zhanghao Wu (CTO), and Romil Bhardwaj (CPO)—along with their professor Scott Shenker and Ion Stoica, a co-founder of Databricks who serves as Board Chairman. Their research, which began with a 2023 paper titled "SkyPilot: An Intercloud Broker for Sky Computing," formed the basis for the company. The open-source version of SkyPilot has achieved significant traction, with over 14 million downloads and adoption by organizations like Shopify, Nubank, Abridge, and Hippocratic AI.

Skypilot.ai's business model is centered on a commercial enterprise platform built on its successful open-source foundation. This platform is designed for organizations managing large GPU fleets, providing a unified interface to run, manage, and scale AI workloads. It allows teams to treat infrastructure as code, defining and launching jobs across various environments—including over 20 cloud providers, Kubernetes, and on-premises systems—using a single YAML file. The platform addresses GPU scarcity and cloud vendor lock-in by acting as a neutral broker that can route workloads to the most cost-effective or available compute resources at any given moment. Key features include optimizing costs by leveraging spot instances, automatic failover and recovery from preemption, and stopping idle resources.

The enterprise-grade SkyPilot Platform extends the open-source capabilities with features tailored for production environments, such as fleet-wide GPU health monitoring, automated remediation, multi-team quota management, single sign-on (SSO), and SOC 2 compliance. It supports a wide range of AI tasks, including model training, inference, reinforcement learning, and production deployments. By abstracting away infrastructure complexity, SkyPilot enables AI development teams to focus on building models rather than managing underlying resources, reportedly allowing some clients to achieve significant improvements in GPU utilization and performance.