RadixArk

RadixArk builds open-source AI infrastructure that makes frontier-level model deployment and training accessible to any developer or company. Its two flagship products are SGLang — a high-performance inference engine originally from UC Berkeley's lab, now serving trillions of tokens daily across hundreds of thousands of GPUs — and Miles, an enterprise-grade reinforcement learning post-training framework. SGLang handles the "serving" side: running a trained model cheaply and fast at scale. Miles handles the "improving" side: RL-based post-training pipelines at 10,000+ GPU scale. On top of both open-source foundations, RadixArk sells a managed cloud infrastructure layer — handling clusters, APIs, multi-tenancy, and monitoring — so that AI companies don't have to rebuild this stack themselves. Customers range from individual developers and startups to Google, Microsoft, and xAI. Founded in 2025 by Ying Sheng (former xAI inference engineer) and Banghua Zhu (former Nvidia), the company operates from Palo Alto.

$400M post-money valuation · 30 employees · 20 open roles

Careers

RadixArk

Open roles