Featherless AI Lands $20M Bet From AMD, Airbus in Race for Open-Source Inference
The convergence came fast. On April 29, 2026, BMW i Ventures announced a $300 million Fund III with an AI thesis. One day later, the Munich-based firm showed up as an investor in Featherless AI's $20 million Series A—alongside AMD Ventures, Airbus Ventures, and a cluster of other strategic players betting that the future of AI deployment won't belong to any single cloud giant or chip maker.
For Featherless, a San Francisco startup that's been quietly building serverless infrastructure for open-source models, the round represents validation of a particular thesis: that enterprises increasingly want flexibility over lock-in. AMD Ventures and Airbus Ventures co-led the financing, announced April 30, which brings the company's total disclosed funding to at least $25 million since a $5 million seed round barely a year earlier, in March 2025.
The timing carries weight beyond the BMW announcement. AMD itself had just committed $150 million to Nutanix in February, part of a broader push into AI infrastructure that now includes this stake in Featherless. Airbus Ventures, meanwhile, doubled down after leading the seed round—perhaps more than the founders expected when they started the company in 2023.
A Platform Built on Neutrality
What Featherless does sounds straightforward on paper: it runs over 30,000 open-source AI models through a single API, letting companies deploy inference workloads without marrying themselves to AWS, Google Cloud, or any proprietary model ecosystem. Founded by Eugene Cheah, Harrison Vanderbyl, and Wesley George—all veterans of the RWKV open-source community—the platform bills itself as hardware-agnostic, a claim that resonates differently now that AMD and other chip makers are trying to chip away at NVIDIA's dominance.
The company hosts infrastructure in both the U.S. and EU, addressing what it calls "sovereign AI" requirements, a term that's become shorthand for European regulators' insistence on data residency. Hugging Face lists Featherless as an official inference provider in its documentation, and the startup claims it's the fastest-growing partner in that ecosystem, though this is not confirmed by Hugging Face.
"Featherless.ai is at the forefront of a critical new phase" in open-source AI infrastructure, Sagi Paz, head of AMD Ventures, said in coverage published May 1. The company plans full support for AMD's ROCm open software platform, which could matter considerably if AMD gains traction against NVIDIA's CUDA ecosystem.
Kasper Sage, managing partner at BMW i Ventures, pointed to enterprise needs around performance, cost, and data locality—table stakes now for any company trying to sell AI tools into heavily regulated industries. Abby Hitchcock, a principal at Airbus Ventures, was more specific: "millions of specialized models" need reliable, cost-effective serving, models that can "load within seconds with high GPU utilization," she said in the same May 1 coverage.
Joan Yao, general partner at Kickstart Ventures, which backed Featherless at seed through the ACTIVE Fund, described the Series A as reinforcing "conviction" in the team's technical edge. Whether that edge holds as competition intensifies is the open question.
Deployment and Differentiation
Featherless plans to use the capital in predictable ways: scale global infrastructure, launch a dedicated marketplace for specialized open models, deepen AMD and ROCm integration, expand enterprise deployments. The company announced those plans April 30 alongside the round.
Recent product moves offer some clues about where it's headed. Managed runtimes for open-source AI agents arrived earlier this year; NVIDIA's NemoClaw stack launched on the platform April 21, 2026, with flat monthly pricing starting at $100. That pricing strategy—flat-rate, unlimited-token tiers posted publicly—is part of how Featherless differentiates itself from competitors whose bills can spiral unpredictably under heavy usage.
But the competitive landscape here is brutal and capital-intensive. Together AI raised $305 million in May 2025. Fireworks hit a $4 billion valuation by October of that year. Baseten closed a $300 million Series E in February 2026, and Modal Labs entered talks at roughly $2.5 billion the same month. Against that backdrop, a $20 million round is significant but hardly overwhelming.
Featherless did not disclose its Series A valuation, which is often a sign that either the number disappointed or that investors negotiated hard. LinkedIn data from mid-April showed just 2–10 employees, a figure that seems almost certainly outdated now, given active hiring posts and the capital infusion. Small teams can move fast in infrastructure software, but scaling a global inference platform with the kind of uptime enterprises demand is another matter entirely.
What the company does have—and what may ultimately matter more than headcount—is alignment with corporate venture arms that have their own strategic reasons to see a hardware-agnostic inference layer succeed. AMD wants an alternative to NVIDIA-dominated stacks. Airbus and BMW want AI tools that don't come with cloud vendor lock-in or geopolitical risk. Whether Featherless can deliver on that promise at scale, with a relatively modest war chest, is the story still unfolding.
For now, the bet is on the board. The hard part—building a business that outlasts the hype cycle—comes next.
