Founded by the creators and core maintainers of SGLang, the open-source inference engine powering trillions of tokens daily for Google, Microsoft, NVIDIA, Oracle, AMD, Nebius, LinkedIn, xAI, Thinking Machines Lab, and humans&, RadixArk emerges to build frontier AI infrastructure for all

PALO ALTO, Calif.–(BUSINESS WIRE)–RadixArk, the company democratizing access to frontier AI infrastructure, launched today 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 (NVIDIA’s venture capital arm), Salience Capital, A&E Investments, HOF Capital, Walden Catalyst Ventures, AMD, LDV Partners, WTT Investment, and MediaTek.

“We believe the next generation of AI won’t be defined by who owns the biggest private infrastructure, but by who builds the most meaningful applications on top of shared, world-class systems.”

Share
Other investors include Igor Babuschkin (Co-Founder of xAI), Lip-Bu Tan (CEO of Intel), Hock Tan (CEO of Broadcom), John Schulman (Co-Founder of OpenAI and Thinking Machines Lab), Soumith Chintala (PyTorch creator and CTO of Thinking Machines Lab), Olivier Pomel (Co-Founder of Datadog), Thomas Wolf (Co-Founder of Hugging Face), William Fedus (Co-Founder of Periodic Labs), Robert Nishihara (Co-Founder of Anyscale), Eric Zelikman (Co-Founder of humans&), and Logan Kilpatrick (Gemini Product Lead). The company will use the capital to grow SGLang, accelerate support for emerging model architectures and frontier hardware, and build large-scale inference and training infrastructure for the next generation of AI applications.

RadixArk was founded by Ying Sheng and Banghua Zhu, AI infrastructure and modeling veterans from xAI and NVIDIA. In 2023, Sheng and others created SGLang, an open-source inference engine for serving models at scale. SGLang quickly became a de facto open-source standard, stewarded by a global community of thousands of contributors across hundreds of companies, universities, and research organizations. SGLang is now deployed across hundreds of thousands GPUs worldwide and generates trillions of tokens daily for Google, Microsoft, NVIDIA, Oracle, AMD, Nebius, LinkedIn, xAI, Thinking Machines Lab, and humans&.

Today, the most sophisticated AI infrastructure is only available to a handful of companies. Neo-labs must rebuild core training and inference stacks from scratch, while infrastructure teams at every company from enterprises to startups are understaffed and underresourced. The result is enormous waste from duplicated effort, siloed research insights, and impeded progress for the entire AI ecosystem. By treating infrastructure as a first-class priority, RadixArk delivers the foundational open systems needed to build the next generation of AI.

“Our mission is simple yet ambitious: make frontier-level AI infrastructure open and accessible to everyone,” said Ying Sheng, co-founder and CEO of RadixArk. “We believe the next generation of AI won’t be defined by who owns the biggest private infrastructure, but by who builds the most meaningful applications on top of shared, world-class systems. We aim to make these systems orders of magnitude cheaper and more accessible, so everyone can build on them.”

RadixArk will go beyond traditional inference solutions that offer compute access for off-the-shelf or open-source models. Instead, the company is building an end-to-end platform that supports the full lifecycle of model development, including training proprietary models, fine-tuning open models, running reinforcement learning, and deploying and running inference at scale. By standardizing on a single platform, RadixArk customers maintain ownership and control of their models while having access to best-in-class infrastructure primitives.

“RadixArk is building the open foundation for the next era of AI—where companies don’t just consume models, they train and manage them as a core part of product development,” said Ivan Zhou, partner at Accel. “By democratizing training and inference infrastructure, RadixArk enables any engineer to experiment and innovate at the frontier, fully owning how AI powers their products.”

RadixArk’s platform is built on battle-tested, open foundations across the AI stack. Inference runs on SGLang, the fastest and most flexible open engine for modern models, while reinforcement learning is powered by Miles, the company’s open-source framework for large-scale training. SGLang was incubated at LMSys, a nonprofit organization founded by researchers from Stanford, Carnegie Mellon, UC Berkeley, and other universities.

“Some of the most important software of the last decade started as open-source projects run by small groups of researchers who refused to compromise on quality. SGLang sits in that lineage: born at LMSys, maintained by thousands of contributors, now the de facto standard for modern inference. RadixArk carries that same spirit into a company, and we’re honored to help Ying and Banghua scale it,” said Arpan Shah, general partner at Spark Capital. “Frontier AI is at risk of becoming the private infrastructure of a handful of companies. RadixArk is the counterweight: a belief that the next generation of AI products will be built on open, shared systems that any team can run, tune, and own.”

SGLang has day-0 support for virtually every open model family (Llama, Qwen, DeepSeek, Kimi, GLM, GPT, Gemma, Mistral, etc.) and hardware provider (NVIDIA GPUs, AMD GPUs, Intel CPUs, Google TPUs, etc.). Together, these frameworks are the starting point for a suite of managed infrastructure and tooling that supports everyone building AI systems, from individual developers to startups, enterprises, and research labs.

“SGLang is the absolute best inference framework for large language models,” said Igor Babuschkin, Co-Founder of xAI and RadixArk angel investor. “It was a crucial part of the infrastructure at xAI, because it enabled folks to run large models faster and more efficiently than many alternatives. I’m excited to see Ying and Banghua expand that vision with RadixArk.”

“Durable technology shifts are built on infrastructure that empowers entire ecosystems, not just individual companies,” said Lip-Bu Tan, CEO of Intel and RadixArk angel investor. “RadixArk has a compelling mission to build the next generation AI infrastructure stack, and SGLang is already emerging as a dominant inference engine for large models. I’m glad to support the company as an early investor.”

About RadixArk

RadixArk is an AI infrastructure company building open, scalable systems for training, deploying, and running frontier models. Founded by the creators and core maintainers behind SGLang — the open-source inference engine serving trillions of tokens daily — RadixArk is building an end-to-end infrastructure platform, treating inference, training, and post-training as core first-class citizens. The company builds on two open-source foundations: SGLang for inference and Miles for reinforcement learning. On top of these, RadixArk ships managed infrastructure and tooling that enable developers, startups, enterprises, and research labs to build and operate advanced AI systems with greater speed, control, and performance.

Founded by AI infrastructure and modeling veterans Ying Sheng and Banghua Zhu, RadixArk has raised $100 million in funding from Accel, Spark Capital, NVentures (NVIDIA’s venture capital arm), Salience Capital, A&E Investments, HOF Capital, Walden Catalyst Ventures, AMD, LDV Partners, WTT Investment, MediaTek, Igor Babuschkin (Co-Founder of xAI), Lip-Bu Tan (CEO of Intel), Hock Tan (CEO of Broadcom), John Schulman (Co-Founder of OpenAI and Thinking Machines Lab), Soumith Chintala (PyTorch creator and CTO of Thinking Machines Lab), Olivier Pomel (Co-Founder of Datadog), Thomas Wolf (Co-Founder of Hugging Face), William Fedus (Co-Founder of Periodic Labs), Robert Nishihara (Co-Founder of Anyscale), Eric Zelikman (Co-Founder of humans&), and Logan Kilpatrick (Gemini Product Lead). Learn more at radixark.com.

Contacts
Press Contact
Kira Wolfe
Command for RadixArk
kira@heycommand.com

(c)2026 Business Wire, Inc., All rights reserved.