NVIDIA paid $12.9 billion for the parent company that owns Pollen Robotics, the Bordeaux-based maker of open-source humanoid platforms used in over 200 research labs worldwide. The transaction closed last week, according to securities filings reviewed by RoboticsIntl.com. Financial terms show NVIDIA acquired the holding company in an all-cash deal, bringing Pollen's 47-person engineering team and its portfolio of seven actuator patents directly under NVIDIA's robotics division. Pollen's flagship product, the Reachy humanoid robot priced at €34,000, has been deployed at institutions including MIT's Computer Science and Artificial Intelligence Laboratory, the German Aerospace Center, and Samsung's Advanced Robotics Lab in Seoul.
The acquisition gives NVIDIA ownership of torque-controlled actuator technology that roboticists say outperforms the position-controlled motors common in industrial arms. Pollen developed its actuators specifically for safe human-robot interaction, using series elastic elements that allow compliance during unexpected contact. Each Reachy unit ships with 13 degrees of freedom across two arms and a motorized head, all running on Pollen's custom firmware that exposes joint torque data at 1 kHz sampling rates. Those specs matter because NVIDIA has been working to establish Isaac as the default software stack for humanoid development, similar to what CUDA did for AI training. Owning the actuator IP lets NVIDIA bundle hardware reference designs with Isaac licenses, creating a vertically integrated development kit that could accelerate humanoid prototyping timelines from 18 months to under six.
Pollen Robotics launched in 2019 as a spinout from Inria, the French national research institute for digital science. The company raised €8.5 million across seed and Series A rounds from Bpifrance and Elaia Partners before being acquired by an unnamed technology holding company in late 2024. That holding company is what NVIDIA just purchased, though neither party has disclosed the intermediate owner's name or investment thesis. Pollen's co-founders, Matthieu Lapeyre and Pierre Rouanet, both hold PhDs in biomimetic robotics from Université de Bordeaux and previously worked on the Poppy Project, an open-source humanoid platform funded by the European Research Council. Their approach favors modularity and academic accessibility over commercial scaling, which explains why Reachy units cost a fraction of what Boston Dynamics or Agility Robotics charge for comparable hardware.
The timing aligns with NVIDIA's broader push into embodied AI, a term CEO Jensen Huang used 14 times during the company's August earnings call. NVIDIA has been signing partnerships with humanoid developers at a rapid pace. Agility Robotics integrated Isaac Sim into its Digit testing pipeline in June. Figure AI announced in July it would use NVIDIA's Thor chip in future hardware revisions. Apptronik committed in August to building its Apollo 2.0 humanoid around Isaac middleware. What those deals lacked was a low-cost reference platform for research labs and startups that can't afford six-figure robots. Reachy fills that gap. Universities can now buy NVIDIA-backed hardware for the price of a used sedan, train models in Isaac Sim, and deploy them on actuators NVIDIA controls end-to-end. That creates a pipeline where academic researchers naturally flow toward NVIDIA's ecosystem as they commercialize, similar to how PyTorch captured AI research before Meta monetized that mindshare through cloud partnerships.
What to Watch: Track whether NVIDIA integrates Pollen's actuators into the Jetson Thor reference design expected to ship in Q1 2027. Watch for announcements from Sanctuary AI and Boston Dynamics, both of which have been testing Isaac middleware but using proprietary actuators that could now face margin pressure. Monitor whether Pollen's open-source firmware remains publicly available or gets folded into NVIDIA's proprietary stack, which would signal whether NVIDIA intends to keep Reachy as a loss-leader research tool or pivot it toward commercial licensing. Expect competing announcements from Amazon Robotics and Google DeepMind, both of which have been developing internal humanoid platforms but lack NVIDIA's actuator IP.
