Our $100M Seed to Build Open Infrastructure for Frontier AI
Today, we are officially launching RadixArk with $100 million in Seed funding at a $400 million post-money valuation. The round was led by Accel and co-led by Spark Capital, with participation from NVentures, Salience Capital, A&E Investment, HOF Capital, Walden Catalyst, AMD, LDVP, WTT Fubon Family, MediaTek, Databricks and more.
Our Mission: Making AI infra open and accessible
AI has moved quickly, but the infrastructure behind it has not yet been built the way it should be.
Frontier labs have some of the strongest internal systems in the world. But even there, infrastructure is often built under the pressure of delivering the next model, supporting the next architecture, or scaling the next workload. The system is often not designed from first principles to be the most reliable, efficient, and generally useful foundation for the broader AI ecosystem.
Outside of these labs, the problem is even more painful. Every company building AI is forced to rebuild the same foundations again and again: inference engines, RL frameworks, orchestration layers, evaluation pipelines, and production tooling. As models, hardware platforms, and use cases become more diverse, this duplication becomes increasingly expensive for everyone.
RadixArk exists to change this. We believe AI infrastructure should be built once, built deeply, and shared broadly. The best systems should not remain locked inside a few organizations. They should be open, reliable, production-ready, and improved with the community.
Rooted in Open Source: SGLang and Miles
Our work starts from two open-source foundations: SGLang for inference and Miles for reinforcement learning and post-training.
- SGLang (Inference Engine): Since its creation in 2023, SGLang has quickly become a de-facto open-source standard for inference. Today, it powers trillions of tokens daily for industry leaders such as Google, Microsoft, NVIDIA, Oracle, AMD, LinkedIn, xAI, Thinking Machines Lab, humans&, and others.
- Miles (RL Framework): Our open-source framework for large-scale training serves as the second foundation of RadixArk, powering reinforcement learning at scale. Although Miles was introduced only recently, it has already been adopted by industry teams for large-scale MoE training, with a focus on efficiency and stability.
We come from the open-source community, and we intend to keep building with it. We will use this capital to grow SGLang and Miles, support new model architectures and hardware platforms, and build managed infrastructure for teams developing AI at scale. In an era where AI makes ordinary productivity cheaper, the rare work is still the same: patient refinement, independent thinking, and an uncompromising pursuit of correctness. We want this to be the culture of RadixArk: focused, humble, fearless, and meticulous about the details that make infrastructure truly work.
RadixArk is an infrastructure-first company. We care about performance, correctness, reliability, hardware efficiency, and long-term system design. To us, infrastructure engineering is not a support function. It is a core creative discipline that determines how fast models can improve, how efficiently they can run, and how widely frontier AI can be built.
The next generation of AI should not be limited by access to private infrastructure. More builders should be able to own their models, their systems, and their future.
That is what RadixArk is here to build. If this mission resonates with you, and if you believe AI infrastructure should be open, deeply engineered, and built for everyone pushing the frontier, we hope you will consider joining us.
The RadixArk Team