Featherless, a serverless AI infrastructure platform, has raised $20 million in a Series A round. The company gives developers access to over 30,000 open-source AI models through a single API, with no server setup or hardware to manage.
Most AI teams today depend on a small number of large providers. A recent MIT Sloan study found that on OpenRouter, the leading open model aggregator tracked in the research, closed models account for close to 80% of AI token usage and nearly 96% of revenue. The same study estimates that a broader shift to open models could save the AI industry approximately $25 billion annually. Users already on aggregator platforms could cut costs by over 70% while improving benchmark performance by more than 14%. The obstacle is not model quality. It is infrastructure. Most teams do not have the tools to run open models reliably at scale, so they stay with closed systems and absorb the costs. Featherless is building the infrastructure intended to change that.
The round was co-led by AMD Ventures and Airbus Ventures, with participation from BMW i Ventures, Kickstart Ventures, Panache Ventures, and Wavemaker Ventures.
AI inference is the process of running a trained model to generate a response or prediction. It is where most of the real-world cost and activity in AI actually sits. The global AI inference market was valued at $97.24 billion in 2024 and is projected to grow at 17.5% annually through 2030. Within that market, there is a clear shift underway toward open-source frameworks and modular architectures that work across different hardware providers rather than locking users into a single vendor's ecosystem.
There is also a political dimension to this. Governments and large enterprises are increasingly cautious about depending entirely on a handful of US-based cloud providers for AI access. Featherless runs infrastructure in both the EU and the US, which matters for organizations that need to keep data within specific borders, including those in regulated sectors and public institutions.
The company has outlined four investment areas. First, it will expand its model library so that newly published open models become available to developers quickly, without waiting for manual infrastructure work. Second, it plans to ship an open-source agent runtime. An agent runtime is a software foundation that lets developers build AI-powered applications and automated workflows without depending on closed APIs.
Third, Featherless will invest in what it calls its AI optimization stack. This is a system that handles inference, model, and workflow optimization as one integrated process rather than separately. Running 30,000 models cost-effectively requires that kind of integrated approach. A single-model optimization strategy does not scale to that number. The fourth area is enterprise: private deployment environments and regional data sovereignty options for organizations with more demanding compliance requirements.
Featherless was created by the team that built RWKV, an open-source model architecture that has drawn attention for offering an alternative to the transformer design used in most large language models today. Research work continues through Recursal Labs, the company's in-house research group. The founding team has roots in Canada, Singapore, and Australia, and the company operates offices in San Francisco, Toronto, Singapore, and several European cities.
The service itself is straightforward to use. A developer makes a request through a single API and Featherless runs the model on its infrastructure. There is nothing to install, no hardware to provision, and no model files to manage locally. The company says it is the fastest-growing inference partner in the Hugging Face ecosystem. Hugging Face is the main platform where AI researchers share and publish models publicly, and it serves as the primary source for Featherless's model catalog.
Featherless positions itself as hardware-neutral and independent from any major cloud provider. That independence is central to its pitch to developers and enterprises who want to avoid being tied to a specific vendor's pricing or access policies. Serving 30,000 models simultaneously requires a fundamentally different technical approach than a platform built around a small, curated model catalog.
AMD Ventures, the venture arm of chip manufacturer Advanced Micro Devices, co-led the round. The investment is connected to a technical partnership: Featherless and AMD are working together to ensure open-source models run natively on AMD's ROCm platform, an open software stack for running AI workloads on AMD hardware. Airbus Ventures co-led alongside AMD. BMW i Ventures, Kickstart Ventures, Panache Ventures, and Wavemaker Ventures also participated.
The mix of investors is worth noting. Industrial companies like AMD, Airbus Ventures, and BMW i Ventures sit alongside more traditional tech-focused venture firms. That combination suggests interest in open AI infrastructure extends into sectors beyond software, including aerospace and automotive.
