SkyPilot secures $20 million to position itself as the neutral hub for AI computing.

      The company revealed itself from stealth mode on Tuesday. The founders developed it at UC Berkeley, in the same lab that gave rise to Spark, Databricks, and Anyscale. The team features Zongheng Yang alongside Databricks co-founders Ion Stoica and Scott Shenker.

      The high-profile investment

      Lux Capital spearheaded the $20 million seed funding round, as first reported by Fortune. Coatue, Amplify, Foundation, Race, and The House Fund also contributed. The list of angel investors is impressive, including Databricks’ Ali Ghodsi and Google’s Jeff Dean, along with the leaders of Vercel, Replit, and Hugging Face.

      They are wagering on a layer that Nvidia had previously sought. The company acquired Run:ai for around $700 million in 2024 to address a similar challenge. Analysts anticipate the AI orchestration market will expand from approximately $14 billion this year to over $60 billion by 2034.

      A fragmented computing landscape

      The issue that SkyPilot addresses is fragmentation. The demand for GPUs significantly exceeds the supply. Consequently, every AI team ends up contacting five to ten providers from the outset just to gather enough chips, as noted by SiliconAngle. They then face difficulties in utilizing them collectively.

      The heart of EU tech

      The latest updates from the EU tech sector, an insightful piece from our experienced founder Boris, and some dubious AI-generated art. It's free, delivered weekly to your inbox. Sign up now! SkyPilot functions as a control plane that consolidates everything into a single interface, spanning over 20 clouds, neoclouds, Kubernetes, and Slurm. It selects the most accessible hardware, allocates workloads to idle chips, and shifts jobs without requiring code changes. The open-source version has been downloaded over 14 million times, with leading deployments operating more than 10,000 GPUs.

      The advantage of neutrality

      SkyPilot is not beholden to any single cloud provider or chip manufacturer. They view Nebius and CoreWeave as partners rather than competitors. Their advantage is straightforward: it directs your workload to wherever it can be run most economically.

      This alters the pressing question facing the industry: can AI companies turn a profit? Yang highlights Cursor, which had negative margins until it ceased renting a competitor’s models and began training its own. He refers to that transformation as “custom intelligence,” which is now more economical due to open-weight models that are comparable to GPT and Claude.

      The drawback

      There is a noticeable gap. The code has been available for free on GitHub for years, raising the question of what prevents customers from using it without payment. Lux’s Brandon Reeves responds that the free version is, in his words, “probably only 1% complete.”

      The deeper investment lies with Stoica himself. His Berkeley lab has already produced two multibillion-dollar enterprises, making it a hub for students aiming to create the next one. SkyPilot offers not just a product but a promise. The entity that controls the computing layer, rather than any singular chip, may dominate the upcoming phase of AI.