RadixArk raises funding at $400M valuation for AI inference tools

RadixArk, a startup that was announced last August, has raised a funding round led by venture capital firm Accel, according to two people familiar with the matter. The funding round values the company at about $400 million. The startup, led by co-founder and CEO Ying Sheng, operates as a commercial entity developing SGLang, an open source AI model engine and tool for optimizing inference. Open source software refers to software with its source code made available for modification. SGLang is designed to optimize inference processing, which is the process of running AI models on hardware. Ying Sheng, who previously announced her role as CEO on LinkedIn last month, leads the company. The startup previously raised angel capital from investors, including Lip-Bu Tan.

The company originated as SGLang in 2023 inside the UC Berkeley lab of Ion Stoica. Stoica, a Databricks co-founder and UC Berkeley professor, incubated both SGLang and its peer project, vLLM. SGLang has gained traction by helping companies run AI models faster and more efficiently, reducing the server costs associated with running AI services. While RadixArk continues to develop SGLang as a free open source tool, the startup has started charging fees for hosting services. The company is also building Miles, a specialized framework designed for reinforcement learning, which is a machine learning training method.

RadixArk’s funding reflects a broader trend of open source research projects quickly transitioning into highly valued commercial infrastructure companies. Other competitors in the inference infrastructure sector have recently secured significant capital:

• Baseten: Secured $300 million at a $5 billion valuation, according to a Wall Street Journal report on Tuesday.
• Fireworks AI: Raised $250 million at a $4 billion valuation.
• vLLM: The competitor project has had conversations about raising upwards of $160 million in funding at a valuation of about $1 billion, as reported by Forbes last month. However, vLLM co-founder Simon Mo characterized the information about this funding round as “factually inaccurate.”

Why it matters

The inference layer is becoming a critical bottleneck for AI efficiency, driving a trend where open-source research projects rapidly evolve into high-valuation commercial infrastructure companies.