Featherless Secures $20M in Series A Funding to Scale Serverless AI Inference Infrastructure

Featherless just dropped a clean $20M in Series A, and the name fits because they are taking weight out of a system that has been getting heavy real fast. San Francisco keeps minting companies that promise scale, but this one is obsessed with removing friction where it actually hurts. Open models, real workloads, no babysitting infrastructure. One API, thousands of models, and a quiet suggestion that maybe the future of AI should not feel like negotiating with a black box.

Eugene Cheah, Harrison Vanderbyl, and Wesley George did not stumble into this lane. They bring 30 years of building and breaking software systems, and it shows in the way Featherless thinks about inference. Serverless is not a buzzword here, it is the product. Developers get to run language, vision, audio, and multimodal models without playing part time data center operator. Enterprises get sovereignty across the US and EU without signing their life away to a single vendor. The pitch is simple, the execution is not.

AMD Ventures and Airbus Ventures co-led the round, with BMW i Ventures, Kickstart Ventures, Panache Ventures, and Wavemaker Ventures stepping in like they know where this is headed. Airbus Ventures, Kickstart Ventures, and Panache Ventures are doubling down from the seed, which tells you the early conviction did not fade once the lights got brighter. $25M total raised in under 2 years is not luck, it is signal.

Featherless is already serving over 30,000 open models, and that number matters because choice is power. Pair that with native support for AMD ROCm and you start to see the angle. This is not just about running models, it is about owning how and where they run. Neutral infrastructure, optimized across inference, model, and workflow layers, built for people who do not want to be locked into someone else’s roadmap.

There is a lesson tucked in here for builders. They did not chase the loudest part of AI, they went after the part everyone complains about when the cameras are off. Cost, control, and complexity. Solve those and adoption follows. Wrap it in a developer experience that feels light, almost featherless, and suddenly scale is not a threat, it is the default setting.

What shows up here is a shift in leverage. Open models are no longer the experimental corner of AI, they are becoming deployable, governable, and commercially viable at scale. Capital is following that reality, and platforms that reduce friction between access and execution are starting to define how enterprises will actually adopt AI, not just talk about it.