RadixArk raises $100M seed

AI infrastructure is the layer most people never see, but it decides how expensive modern models are to run. That matters now because the bottleneck is no longer just getting a model — it’s keeping the thing fast, stable, and affordable once real traffic shows up. RadixArk is the newest bet on that problem. On May 5, 2026, the company launched with a $100 million seed round and a plan to turn SGLang, its open-source serving engine, into a broader platform for training and inference at scale. (finance.yahoo.com) ### What is RadixArk actually selling? RadixArk is building software for the messy middle between AI models and the hardware they run on. Its roots are in SGLang, an open-source inference engine created in 2023 that helps developers serve large models more efficiently. The company says it wants to grow from that(finance.yahoo.com)neering orgs. (finance.yahoo.com) ### Why does SGLang matter so much? Because inference is where the bills pile up. Every time a model answers a prompt, memory and compute get chewed up, and a lot of that work is redundant when prompts share context. SGLang’s pitch is that it cuts waste in that process, which lowers cost and improves throughput. That is why the project spread so fast across production AI stacks. (techfundingnews.com) ### How big is the traction? Pretty big for something that started as open source. RadixArk says SGLang is deployed across hundreds of thousands of GPUs worldwide and generates trillions of tokens daily. It also says users include Google, Microsoft, NVIDIA, Oracle, AMD, LinkedIn, xAI, and others. Those are huge names, and even if usage depth varies by company, the signal is clear — this is not a toy developer tool. (finance.yahoo.com) ### Who founded it? The company was founded by Ying Sheng and Banghua Zhu, described as AI infrastructure and modeling veterans from xAI and NVIDIA. That founder profile helps explain both the product focus and the investor list. This is not a startup trying to discover a wedge after the funding — it is a team commercializing a tool that already has real adoption inside demanding AI environments. (finance.yahoo.com) ### Who put in the money? The round was led by Accel and co-led by Spark Capital. NVentures, NVIDIA’s venture arm, also joined, along with AMD, MediaTek, HOF Capital, Walden Catalyst Ventures, and others. The company says the financing values RadixArk at $400 million post-money. Separate reporting also notes backing tied to Databricks and Broadcom leadership. (finance.yahoo.com) ### Why raise that much at seed? Because this category eats capital fast. If you are building software that has to support new model architectures, new chips, and giant enterprise workloads, a normal seed round does not go very far. The money is supposed to fund SGLang’s growth, broader hardware support, and a b(finance.yahoo.com)en AI workloads. (finance.yahoo.com) ### What does this say about the market? It says investors still believe there is room below the model layer and above the chips — a valuable control point where software can make AI much cheaper to run. That is a crowded race, with projects like vLLM chasing similar territory, but RadixArk has one thing many infra startups do not: usage first, company second. That usually gets attention. (techfundingnews.com) ### Bottom line? This is a funding story, but really it is a cost story. RadixArk is trying to turn a widely used open-source engine into the infrastructure stack more companies rely on when frontier models leave the lab and hit production. The $100 million seed says investors think that layer could become one of the most valuable parts of the AI stack. (finance.yahoo.com)