RadixArk, a Shenzhen‑based startup, secured $100 million in a round that values the company at $400 million. Its open‑source engine, SGLang, sits between AI models and hardware, promising to cut inference and training power needs by optimizing short‑term memory usage.

Founded by Banghua Zhu and Ying Sheng Lianmin Zheng, the firm targets the growing demand for cheaper, faster AI workloads. Competitors focus on chip design, while RadixArk delivers a software layer that can be deployed across existing GPU and CPU platforms, broadening its appeal.

Backed by Nvidia and other industry heavyweights, RadixArk positions itself as a cost‑effective alternative to hardware‑centric solutions. Investors see the potential for significant savings in data‑center energy bills, while customers gain faster deployment cycles without new silicon investments.

The funding signals confidence that software‑centric optimizations can compete with hardware innovations, reshaping how enterprises approach AI scalability.