RadixArk is a spin-out from UC Berkeley that maintains SGLang, an open-source engine designed to make the inference phase of large language models markedly faster and cheaper. The platform is already in use by teams at xAI and Cursor to boost training and real-time inference performance. Building on this foundation, RadixArk will keep SGLang free while offering paid cloud hosting and releasing new products such as Miles, a reinforcement-learning framework that helps models improve over time. The effort began in 2023 under Databricks co-founder Ion Stoica and has now transitioned into an independent commercial entity. Although much of its software remains open source, the company has introduced revenue-generating hosting services to capitalize on enterprise demand for AI infrastructure. RadixArk’s latest fundraising values the firm at roughly $400 million, giving it resources to expand product development and go-to-market efforts.

Total raised

$100M

Funding rounds

1

Latest round

Seed

Latest activity

May 2026