For teams working on cutting-edge AI models, the sources of computing power are becoming increasingly diverse. However, this diversity also introduces challenges such as fragmented computing resources, inefficient management, and soaring costs. SkyPilot, an open-source AI computing power scheduling project developed by the Sky Computing Lab at the University of California, Berkeley, rose to prominence for its pivotal role in training the Vicuna model. With over 14 million downloads and adoption by industry giants like Meta FAIR, SkyPilot has established itself as a leader in the field. In late July 2026, SkyPilot officially transitioned into a startup, securing a $20 million seed funding round.

The open-source version of SkyPilot primarily caters to individuals and small teams, offering them efficient access to computing resources. Meanwhile, the commercialized SkyPilot Platform is designed to meet the needs of organizations requiring large-scale management of GPU clusters, some of which boast tens of thousands of cards. By venturing into the AI orchestration sector—a market projected to exceed $60 billion by 2034—SkyPilot is positioning itself at the forefront of innovation.

Maintaining a neutral stance in the competitive landscape, SkyPilot adopts a unique business model. Rather than building its own computing centers or reselling computing resources, the company focuses on optimizing computing power scheduling. This approach allows customers to leverage their existing infrastructure while benefiting from SkyPilot’s advanced orchestration capabilities.

Looking ahead, SkyPilot has ambitious plans to expand its reach. The company aims to cover heterogeneous computing power and support the entire AI lifecycle, from development to deployment. By doing so, SkyPilot hopes to empower organizations to build exclusive, customized intelligent models tailored to their specific needs.