SkyPilot raises $20M to unify AI compute
A Berkeley spinout backed by Databricks and Google's leadership exits stealth with a control layer that routes AI workloads across clouds and GPU generations.
SkyPilot came out of stealth on July 21 with a $20 million seed round led by Lux Capital, with Amplify Partners, Coatue Management, Foundation Capital, Race Capital, and The House Fund also participating. The company gives AI teams a single control layer over compute spread across hyperscale clouds, specialized neoclouds, Kubernetes clusters, and different GPU generations, so a training or inference job can move to wherever capacity is cheapest or available without a team rewriting its infrastructure code for each provider.
The founding team pairs Berkeley researchers Zongheng Yang, Zhanghao Wu, and Romil Bhardwaj with Databricks co-founder Ion Stoica and networking researcher Scott Shenker, and the funding round drew personal participation from a notable bench of technical leaders: Databricks CEO Ali Ghodsi, Google Chief Scientist Jeff Dean, Vercel CEO Guillermo Rauch, Replit CEO Amjad Masad, Hugging Face CEO Clem Delangue, and dbt Labs CEO Tristan Handy.
The pitch sits squarely in a gap operators keep hitting once they’re training or serving at any real scale: GPU capacity is scattered across providers with different APIs, pricing, and availability windows, and building your own scheduler to route around that is a project most teams didn’t sign up for. It’s an infrastructure layer alongside the routing decisions covered in neocloud GPU financing, one aimed at making multi-cloud GPU access a solved problem rather than a build-it-yourself one. Whether it earns “the Switzerland of AI compute” framing depends on whether it stays genuinely neutral as it signs deals with the same clouds it’s meant to arbitrage between, which is worth watching as the product matures past this seed stage.