Hugging Face buys Pollen Robotics to push open-source humanoids

If you thought Hugging Face was only interested in hosting code and weights in the cloud, think again. The open-source AI giant is officially getting into the hardware business.

Hugging Face announced it has acquired Pollen Robotics, a French startup that has spent the last nine years building open-source humanoid hardware. The financial terms were not disclosed.

Following the acquisition, Hugging Face plans to commercially sell Pollen's latest robot, Reachy 2, while giving developers full access to inspect, modify, and improve the underlying codebase.

Meet Reachy 2, the $70,000 open-source lab partner

Founded in 2016 by former Inria researchers Matthieu Lapeyre and Pierre Rouanet, Pollen had previously raised modest funding of about €2.5 million ($2.83 million), backed by investors like Bpifrance.

This marks Hugging Face's fifth acquisition, following deals for tools like Gradio and XetHub, but it represents its first real leap into physical hardware.

The two teams are not strangers. After Hugging Face hired former Tesla Optimus engineer Remi Cadene in early 2024, it launched the LeRobot initiative to bring open-source frameworks to embodied AI. The LeRobot GitHub repo quickly amassed over 12,000 stars, frequently pairing its software with Pollen's designs.

Reachy 2 is built specifically for research, education, and AI experimentation rather than mass-market housekeeping. Moving on a wheeled mobile base, the upper body features stereo vision, spatial audio, and two arms with seven degrees of freedom that can carry up to 3 kilograms each.

The entire system can be teleoperated via VR gear and scripted in native Python. Research universities like Cornell and Carnegie Mellon have already put Reachy 2 to work in their labs. Hugging Face is offering the unit for purchase right now at $70,000.

The debate over open-source bodies

Not everyone in the robotics world thinks opening up full hardware and firmware access is a great idea.

Critics frequently argue that open-sourcing robots makes them lucrative targets for malicious hackers, warning that compromised physical actuators pose real safety threats that chatbots never could.

Hugging Face co-founder and chief scientist Thomas Wolf pushed back against that narrative, arguing that physical AI must be open, affordable, and transparent. The company maintains that an open community is actually the best defense for finding and patching vulnerabilities before bad actors can exploit them.

With CEO Clem Delangue predicting that over 100,000 personal robots will be pre-ordered this year, Hugging Face wants to make sure those future machines run on community-driven protocols rather than closed corporate stacks.

Why it matters

We think Hugging Face's move is a much-needed reality check for the humanoid robotics hype cycle.

While companies like Tesla and Figure dominate headlines with polished videos of closed-source bipeds, academic labs and independent researchers are often left stranded. Building custom hardware from scratch is brutally slow, while commercial testbeds remain prohibitively expensive.

At $70,000, Reachy 2 is still out of reach for weekend hobbyists, but for universities and AI labs, it offers an off-the-shelf, fully open platform to validate models in the real world. Paired with HF's ultra-cheap $100 SO-100 arm, the company is quietly constructing an entire hardware pipeline for embodied AI.

For the broader industry, Hugging Face is attempting to pull off the same playbook that won it the software space. By anchoring robotics models, teleoperation datasets, and physical blueprints inside its hub, it positions itself as the default operating system for physical intelligence.

The obvious risk here is execution. Building, shipping, and servicing delicate electromechanical hardware is infinitely harder than hosting Python scripts on AWS. If Pollen's supply chain falters or physical safety bugs emerge, Hugging Face may discover that open-source goodwill gets strained quickly when moving metal is involved.