In the current fierce competition at the AI model layer, Silicon Valley's three core semiconductor companies, NVIDIA, AMD, and Intel, have unusually set aside their market rivalry and...Total seed funding of $1 millionAll of these investments have been poured into RadixArk, a startup team originating from the open-source community. This company, known for its open-source inference engine SGLang and reinforcement learning framework Miles, not only achieved a valuation of up to $4 million in its early stages, but this "dream team" of investors also suggests that a dramatic shift is taking place in the underlying landscape of global AI infrastructure.
SGLang and Miles: The "Dual-Wielders" Mastering Reasoning and Training
RadixArk's ability to impress tech giants lies in its possession of two of the most dominant weapons in the open-source community, covering the two main lines of AI development:
• Inference Standard SGLang:In just two years, SGLang has become the de facto industry standard. It enjoys immense popularity on GitHub and is deployed on over 40 GPUs. More importantly, its "Day-0 compatibility" feature means that SGLang can always support both MoE (Hybrid Expert) architectures and long context models from the outset, pushing hardware performance to its limits. Heavy users include Google, NVIDIA, and even xAI.
• Miles, a powerful training tool:A framework specializing in large-scale reinforcement learning (RL) training. On the day of DeepSeek-V4's release, SGLang and Miles became the world's first open-source technology stack to simultaneously support its inference and RL training, demonstrating unparalleled engineering discipline and technical acumen.
The founding team of RadixArk is equally remarkable. Its CEO, Sheng Ying, is the initiator of LMSYS Org and has extensive practical experience in system development; while its CTO, Zhu Banghua, previously served as NVIDIA's chief research scientist. This dream combination of "researching the open-source ecosystem + underlying hardware systems" ensures that the technology stack not only progresses quickly but also runs stably.
The "Same Table" Logic of Chip Giants: Solving Structural Mismatches
While NVIDIA, AMD, and Intel are market competitors, why are they willing to jointly invest in the same inference engine? This reflects the structural pain points of the current AI industry: "expensive hardware computing power and a fragmented software ecosystem."
• NVIDIA's defense and expansion:For NVIDIA, investing in SGLang can maximize the utilization of high-end GPUs such as the H100 and B200. Rather than building a closed ecosystem, it's better to ensure that the most powerful open-source engine runs smoothly on its own GPUs, making customers willing to pay for the expensive hardware.
• The Breakthrough Battle Between AMD and Intel:For AMD and Intel, who are struggling to break NVIDIA's CUDA monopoly, a "neutral, open-source, and powerful" software standard is urgently needed. Only through an intermediary layer like RadixArk can AMD's MI300 series or Intel's Gaudi chips have a platform to perform fairly, thereby attracting developers to join.
