RadixArk is building open infrastructure for frontier AI, anchored by SGLang, the high-performance open-source inference engine its founders created. SGLang is already deployed across hundreds of thousands of GPUs and serves trillions of tokens per day for major clients including Google, Microsoft, and xAI, making it one of the most widely used open inference stacks in the industry.
The company's mission is to democratize frontier AI infrastructure by offering an end-to-end platform that spans training, fine-tuning, reinforcement learning, and inference. The aim is to give companies of all sizes ownership and control over their AI models without having to rebuild core systems from scratch, lowering the barrier to operating advanced models at scale.
RadixArk was founded by Ying Sheng and Banghua Zhu, the creators of SGLang, who began the project in 2023 to serve models efficiently at scale. The founders' research roots and open-source track record give the company deep credibility in the inference-optimization space.
In May 2026, RadixArk launched with $100M in seed funding at a $400M valuation, led by Accel and Spark Capital, with backing from NVIDIA and AMD plus an extraordinary roster of angels including co-founders of xAI, OpenAI, and Hugging Face, the PyTorch creator, and the CEOs of Intel and Broadcom.
By commercializing SGLang while keeping it open, RadixArk positions itself as foundational infrastructure for the AI ecosystem, competing in the same arena as other inference and model-infrastructure platforms while leaning on its open-source community and hardware-vendor relationships.