RadixArk has officially launched with $100 million in seed funding, positioning itself as a new player in the fast-growing AI infrastructure market. The Palo Alto-based company, founded by Ying Sheng and Banghua Zhu, plans to expand access to advanced training and inference systems that are typically available only to the world’s largest technology companies. The round values RadixArk at $400 million post-money and signals strong investor confidence in open infrastructure for frontier AI.
Major Seed Round Backed by Leading Investors
The funding round was led by Accel and co-led by Spark Capital, with participation from NVentures, NVIDIA’s venture capital arm, as well as AMD, MediaTek, HOF Capital, Walden Catalyst Ventures, Salience Capital, A&E Investments, LDV Partners, WTT Investment, and others. RadixArk also attracted prominent angel investors, including xAI co-founder Igor Babuschkin, Intel CEO Lip-Bu Tan, Broadcom CEO Hock Tan, OpenAI co-founder John Schulman, Hugging Face co-founder Thomas Wolf, and PyTorch creator Soumith Chintala. The company said the capital will support the growth of SGLang, broader hardware and model support, and the development of large-scale AI infrastructure for training, post-training, and inference.
Building on the Momentum of SGLang
RadixArk was created by the team behind SGLang, an open-source inference engine first developed in 2023 and now used across the AI ecosystem. The technology is described by the company as powering trillions of tokens daily for organizations including Google, Microsoft, NVIDIA, Oracle, AMD, Nebius, LinkedIn, xAI, Thinking Machines Lab, and humans&. Its adoption across hundreds of thousands of GPUs has helped establish SGLang as a widely used open-source framework for serving large models efficiently.
Addressing Infrastructure Bottlenecks in AI
The company is entering the market at a time when AI infrastructure remains expensive, complex, and concentrated among a relatively small group of well-resourced firms. RadixArk argues that many startups, enterprises, and research teams are forced to rebuild similar systems internally, creating duplication, slowing progress, and limiting experimentation. Its stated goal is to make frontier-grade infrastructure more affordable, open, and accessible so more organizations can build and control their own AI systems.
From Inference to Full Model Lifecycle Support
Unlike platforms focused mainly on providing compute access for existing models, RadixArk plans to offer an end-to-end infrastructure layer for AI development. The platform is expected to support proprietary model training, fine-tuning of open models, reinforcement learning, large-scale deployment, and production inference. Alongside SGLang for inference, RadixArk is also building on Miles, its open-source framework for large-scale reinforcement learning.
RadixArk’s launch reflects a broader industry push toward open, shared infrastructure as AI systems become more central to product development. With backing from major venture firms, chipmakers, and influential AI leaders, the company is aiming to become a foundational platform for teams building advanced AI applications. Its success will depend on whether it can turn the momentum around SGLang into a scalable commercial platform that gives developers, startups, enterprises, and research labs greater control over frontier AI infrastructure.