May 7, 2026Funding-SeedInfrastructureOpen Source

RadixArk Raises $100M Seed — SGLang Goes Commercial

SGLang has been the open-source inference standard for almost three years. Google, Microsoft, NVIDIA, Oracle, AMD, xAI, LinkedIn, Thinking Machines all run it in production, hundreds of thousands of GPUs, trillions of tokens daily. The creators just spun out a company.

RadixArk is led by Ying Sheng and Banghua Zhu (ex-xAI, ex-NVIDIA). Seed round of $100 million at a $400M post-money valuation. Accel led. Spark co-led. NVentures (NVIDIA's VC arm), AMD, MediaTek, Databricks, HOF Capital, Walden Catalyst, LDV all in. Notice the strategic logos — chip vendors and a foundation-model platform on the same cap table is unusual.

The bet: open-source inference engine plus a parallel RL framework called Miles. SGLang is the inference rail; Miles handles large-scale post-training, already adopted for MoE training at scale. RadixArk wants to be the company that maintains both, plus managed infrastructure for teams that don't want to roll their own.

This is the vLLM playbook with a twist. vLLM raised inside Anyscale's commercial wrapper. SGLang took longer to commercialize but built deeper hyperscaler distribution first. The five-year question: do hyperscalers (Google, AWS, Azure) keep paying for managed SGLang, or do they fork it like they forked Lucene? RadixArk is betting on the maintainer-as-vendor pattern that Databricks, MongoDB, and Elastic proved out.

Source: https://www.radixark.com/blog/radixark-launches-100m-seed
← Previous
DeepSeek Finally Takes Outside Money — $45B First Round
Next →
OpenSearch-VL Open-Sources the Whole Multimodal Search Stack
← Back to all articles

Comments

Loading...
>_