SkyPilot raises $20 million seed round for AI infrastructure tooling

SkyPilot raised $20 million in seed funding and launched as a company to commercialize software for managing AI infrastructure across clouds and on-premises systems. Lux Capital led the round, with participation from other institutional investors, Databricks CEO Ali Ghodsi, Google chief scientist Jeff Dean and other angel investors.

The San Francisco startup did not disclose a valuation, revenue, customer count or current headcount. It said the capital will go toward development of both its open-source project and paid product, as well as hiring.

SkyPilot is built around an open-source project created at UC Berkeley by the company’s founding team. The group includes Ion Stoica, the Databricks co-founder and UC Berkeley computer science professor. The company is led by co-founder and CEO Zongheng Yang.

The pitch is aimed at a real operating problem for AI teams: compute is increasingly split across different environments. A company may train models and store data on-premises, run inference in a public cloud, and use clusters with different accelerators, schedulers and management layers. SkyPilot says its software gives developers one interface to operate those mixed environments.

What the software claims to automate

According to SkyPilot, its open-source tool reduces the code changes normally required to move AI workloads between infrastructure types. The company also says the software can automate routine maintenance work across differently configured clusters.

One example is bin packing, the process of placing workloads on servers in a way that uses available memory and compute more efficiently. SkyPilot says its tool can automatically adjust which workloads run on which machines. It also says the software can provision more infrastructure when a workload needs additional hardware and can resolve some technical issues without human intervention.

Those are useful claims if they hold up at scale, though SkyPilot has not disclosed third-party benchmarks, customer metrics or cost-savings figures. In AI infrastructure, where GPU supply, cloud pricing and cluster reliability are board-level concerns for model companies and enterprise AI teams, orchestration tooling is becoming a crowded and high-stakes category.

Paid product adds sandboxes and GPU management

SkyPilot plans to make money through a commercial product called SkyPilot Platform. The paid version includes tools for creating inference sandboxes: virtual machines where AI agents can run code while limiting security exposure. Developers can customize those environments with programming tools and other components, according to the company.

SkyPilot says the platform can start a new sandbox in less than one second by preloading virtual machines before an AI agent requests them. It also says the product can launch GPU clusters containing 5,000 chips in under a minute. Another feature, SkyPilot GPU Manager, is designed to monitor accelerator health and address issues it detects.

“Every organization is building custom intelligence around its own data and domains,” Yang said. “The challenge is that the AI compute needed to build it is fragmented across clouds. SkyPilot gives frontier AI teams a single platform to manage that infrastructure so they can build custom intelligence faster.”

The round gives SkyPilot capital and a set of recognizable backers, but the company still has to prove how much of the messy AI infrastructure stack customers are willing to hand to a new management layer.

This story draws on original reporting from SiliconANGLE.