The team behind the open-source inference engine SGLang has officially incorporated as RadixArk and closed a $100 million seed round at a $400 million post-money valuation. Accel led the round, with Spark Capital co-leading. Notably, chip rivals Nvidia (through NVentures) and AMD both participated, along with MediaTek.
SGLang is a middleware layer that sits between AI models and hardware, reducing inference costs by managing GPU memory more efficiently. The project kicked off in 2023 and already handles trillions of tokens per day for clients including Google, Microsoft, Nvidia, Oracle, and xAI.
RadixArk also operates another open-source project, Miles, which focuses on large-scale reinforcement learning training and has been used for MoE (mixture-of-experts) model training.
The company was founded by Banghua Zhu and Ying Sheng. The funding will be used to expand both open-source projects, support more model architectures and hardware platforms, and launch a managed infrastructure service for enterprises.