SkyPilot

SkyPilot is a San Francisco-based company that builds a unified AI compute platform for teams running large-scale AI infrastructure. Its technology presents compute spread across hyperscale cloud providers, specialized neoclouds, Kubernetes clusters and multiple generations of accelerators through a single control plane, so that organizations can develop, manage, monitor and scale workloads across providers without changing how those workloads are built or operated.

The commercial product, SkyPilot Platform, is a managed offering built on the SkyPilot open source project. It covers large GPU fleet operations, standardized cluster management, workload orchestration, governance and enterprise controls, and supports interactive development, agentic workloads, training, reinforcement learning, inference, evaluations and multi-cluster production serving. Additional capabilities include fleet-wide GPU health monitoring, automated remediation, high availability, team and quota management and single sign-on; the company states it has achieved SOC 2 compliance. Customers bring their own compute rather than purchasing capacity from SkyPilot.

The company frames its market around what it calls "custom intelligence" — agents, applications and post-trained models built on an organization's own data and run on infrastructure it controls — arguing that building such systems requires more compute than any single provider can supply.

Founding story

SkyPilot was founded by UC Berkeley researchers Zongheng Yang, Zhanghao Wu and Romil Bhardwaj together with Databricks co-founder Ion Stoica and networking and cloud computing researcher Scott Shenker, both Berkeley professors. The underlying technology emerged from years of AI systems research at UC Berkeley. Stoica has described the problem as one he encountered at Databricks, where expanding from one cloud to two took a year of engineering work; CEO Zongheng Yang interned at Databricks when it had roughly ten employees.

Business model

SkyPilot maintains an open source project and sells SkyPilot Platform, a managed commercial layer aimed at enterprises operating large GPU fleets. Customers bring their own AI compute from hyperscalers, neoclouds and Kubernetes clusters, and SkyPilot provides the orchestration, management, governance and enterprise control layer on top of it.

Commercial revenue comes from the managed SkyPilot Platform product, offered on top of the free open source project; access was opened to select new customers at launch.

Traction

The open source project has surpassed 14 million downloads and more than 280 contributors, and SkyPilot says it powers AI compute across hundreds of organizations, including Nubank, Abridge, Applied Compute, H Company, Archer and Hippocratic AI. Private preview customers managed more than 10,000 GPUs and supported over 200 researchers within a single organization.

Latest developments

On July 21, 2026 the company launched from stealth with $20 million in seed funding led by Lux Capital and simultaneously introduced SkyPilot Platform, its managed commercial product, opening access to select new customers.