RadixArk Spins Out With $400M Valuation Amid AI Inference Boom
▼ Summary
– RadixArk, the commercial company behind the open-source AI tool SGLang, was recently valued at about $400 million in a funding round led by Accel.
– The startup, which originated from a UC Berkeley lab, focuses on optimizing AI inference processing to make models run faster and more cheaply on existing hardware.
– This is part of a broader trend where popular open-source AI infrastructure tools, like vLLM, are becoming highly valued, venture-backed startups.
– These inference optimization tools are critical as they can create enormous immediate cost savings for companies running AI services.
– The sector is seeing a major funding surge, with other startups like Baseten and Fireworks AI also securing hundreds of millions at multi-billion dollar valuations.
A clear trend is taking shape within the AI infrastructure sector, where widely-used open-source projects are evolving into venture capital-backed enterprises commanding valuations in the hundreds of millions. The newest player in this space is RadixArk, the commercial entity formed around SGLang, a tool gaining rapid adoption for its ability to make AI models operate more swiftly and cost-effectively. The startup has achieved an approximate valuation of $400 million in a funding round spearheaded by Accel, according to sources with knowledge of the deal. This marks a significant milestone for a company that was formally introduced just last August.
The development coincides with a shift of key personnel from the SGLang maintenance team to the newly established commercial venture. Originally created as SGLang in 2023 within the UC Berkeley laboratory of Databricks co-founder Ion Stoica, RadixArk has now spun out independently. Ying Sheng, a principal contributor to SGLang and a former engineer at xAI, departed Elon Musk’s company to become the co-founder and CEO of RadixArk. The startup had previously secured angel investment from figures such as Intel CEO Lip-Bu Tan.
Both the open-source SGLang project and its commercial counterpart, RadixArk, concentrate on a critical area: optimizing inference processing. This involves making AI models run faster and more efficiently on existing hardware. Given that inference constitutes a major portion of the server expenses for AI services, tools that enhance this process can deliver substantial and immediate cost savings for companies.
RadixArk is part of a broader movement. Another project, vLLM, which is a more established system for inference optimization, is also making the transition to a funded startup. Reports indicate the newly formed company is in discussions to raise over $160 million at a valuation nearing $1 billion, with Andreessen Horowitz reportedly leading the investment. Like SGLang, vLLM was also incubated in Ion Stoica’s UC Berkeley lab. Industry observers note that several major technology firms already rely on vLLM for their inference workloads, while SGLang has seen its popularity climb sharply in recent months.
On the product front, RadixArk continues to advance SGLang as an open-source AI model engine. Simultaneously, the company is developing a separate framework named Miles, which is tailored for reinforcement learning, a method that enables AI models to improve through experience. While the core tools remain freely available, RadixArk has begun generating revenue by charging for associated hosting services.
This activity underscores a significant surge in funding for startups that provide inference infrastructure to developers, highlighting the layer’s enduring strategic importance. The sector is attracting considerable capital, as seen with Baseten’s recent $300 million raise at a $5 billion valuation and a similar $250 million round for competitor Fireworks AI at a $4 billion valuation last fall. The race to build the most efficient AI inference engine is clearly heating up.
(Source: TechCrunch)
