The open-source AI tool SGLang has officially become a standalone company called RadixArk, with a valuation of about $400 million. SGLang started in 2023 as a research project inside the University of California, Berkeley lab led by Ion Stoica, a co-founder of Databricks.
The tool helps speed up the inference stage of AI models – the part where a trained model produces answers or results, which can cut computing costs and boost performance.
According to people familiar with the situation, the new company raised money from backers, including venture firm Accel and early angel investors like Intel CEO Lip-Bu Tan. RadixArk was announced publicly last August, and the team behind it has now fully shifted away from the original open-source project.
Making large AI models run faster and more affordably
SGLang was built to help make large AI models run faster and more cheaply. Teams at companies such as xAI and Cursor have used the tool to speed up training and inference for their AI systems.
As interest in these performance tools has surged, part of the original SGLang team left the UC Berkeley lab and joined the new startup, RadixArk.
RadixArk plans to continue developing SGLang as an open-source engine while also building additional tools, such as Miles, a reinforcement learning framework that helps models improve over time.
While much of the company’s software remains free, RadixArk has begun charging for paid hosting services.
The move reflects a broader trend in the AI infrastructure world, where tools that started as academic or open-source projects are being turned into commercial businesses as demand grows.
Companies focused on AI inference, the part of AI that makes models work in real time, are attracting more funding as the market expands, says the report.
