A portion of the team behind SGLang, a widely-used open-source tool that helps companies such as xAI and Cursor speed up AI model training, has now moved to its newly established commercial venture. This company, named RadixArk, was first introduced in August.

Originally emerging as the SGLang project in 2023 within the UC Berkeley laboratory of Databricks co-founder Ion Stoica, RadixArk was recently valued at approximately $400 million in a funding round spearheaded by Accel, according to sources familiar with the deal. The startup had earlier secured angel investment from backers that included Intel CEO Lip-Bu Tan.

Ying Sheng, a major contributor to SGLang and a former engineer at xAI, departed Elon Musk’s AI firm to co-found and lead RadixArk as CEO, as she announced on LinkedIn last month. Sheng had previously worked as a research scientist at Databricks. Requests for comment sent to RadixArk’s Ying Sheng, Accel, and Lip-Bu Tan were not returned.

Both SGLang and RadixArk concentrate on enhancing inference processing—enabling AI models to operate more quickly and efficiently on existing hardware. Alongside model training, inference accounts for a significant share of the server expenses tied to AI services. Consequently, optimization tools in this area can deliver substantial cost savings almost instantly.

vLLM, a more established project for inference optimization, has also evolved from an open-source initiative into a startup. The newly formed company has reportedly discussed raising over $160 million at a valuation around $1 billion, according to a recent report. Andreessen Horowitz declined to comment.

Similar to SGLang, vLLM was incubated in Ion Stoica’s lab at UC Berkeley. Stoica, a professor at the university, is the renowned co-founder of Databricks and several other startups. His firm did not invest in either company.

RadixArk continues to advance SGLang as an open-source AI model engine. The startup is also developing Miles, a specialized framework built for reinforcement learning, which enables businesses to train AI models to improve progressively over time.

Startups offering inference infrastructure for developers have attracted a wave of funding in recent months, highlighting the ongoing significance of the inference layer in AI. Baseten recently obtained $300 million at a $5 billion valuation, as reported on Tuesday. This follows a comparable move by competitor Fireworks AI, which raised $250 million at a $4 billion valuation last October.