NVIDIA Acquires Hugging Face: The Open-Source Era of Physical AI Is Accelerating
Yesterday, NVIDIA officially announced it will acquire Hugging Face for approximately $12.93 billion (official announcement). This partnership essentially welds together the two tracks of “open-source AI” and “robotics” — NVIDIA brings the compute, simulation, and deployment infrastructure, while Hugging Face brings the world’s largest open-source model community: over 18 million developers sharing 3 million+ models, 500K+ datasets, and 1 million+ apps, with more than 200,000 enterprises building AI on top of it.
Jensen Huang made it clear: Hugging Face will remain an open platform supporting multi-cloud, multi-accelerator development and deployment. The open ecosystem isn’t being co-opted — it’s being amplified.
And Seeed Studio happens to be standing right beside both companies — we are an Elite Partner for NVIDIA Jetson edge AI hardware, and a deep participant in the Hugging Face ecosystem. So today, we’d like to take this opportunity to talk seriously about what we’ve been doing in this wave of Physical AI, and where you should start learning and getting hands-on with embodied AI.
Why We Embrace “Open Source + Hardware Accessibility”
Open-sourcing a model is only the starting point. Between a developer going from “seeing a model” to “running it on their own robot” lies an entire pipeline — data collection, simulation, training, and deployment — and every single step is a place where people give up.
What Seeed does is to connect all these links: provide more affordable AI hardware and well-documented references to learn from. Backed by our mature supply chain and agile manufacturing platform, we gradually turn global innovation full of uncertainty into mass-producible, deliverable products.
Seeed offers a complete NVIDIA Jetson-based product lineup that provides developers with accessible, easy-to-develop embodied AI platforms, along with a series of vivid open-source robot case studies — from reBot and Reachy Mini to MicroDuck.
From open-source hardware and open-source AI stacks, to out-of-the-box platforms, validated accessories, and hands-on tutorials — we believe the best way to learn Physical AI is to work through the pitfalls yourself, experiencing and learning step by step in practice, and building it with your own hands.
reBot: A “Truly Open-Source” Robot Arm, and an Ultra-Detailed Free Embodied AI Course
reBot Arm B601-RS is our fully open-source 6-axis robot arm built for getting started with embodied AI: 2.5 kg payload, ±0.1 mm repeatability, driven by RoboStride motors. reBot also has a sibling version, the B601-DM (Damiao motors, 0.2 mm precision, 767 mm reach) — same structure, two motor options, your choice.
The open source we’re talking about is “true open source”: beyond code — sheet-metal and 3D-printed structural source files, a BOM detailed down to every screw’s spec and purchase link, the Python SDK, and support for ROS1/2, NVIDIA Isaac Sim, and the Hugging Face LeRobot framework — all released without reservation. The hardware blueprints are all yours. For production-line players who like machining their own parts, you’re welcome too — we hope more creative makers will join this ecosystem and unlock infinite possibilities for the reBot arm together.
reBot has been natively integrated into the Hugging Face LeRobot ecosystem (reBot in the LeRobot framework), supporting teleoperation, data collection, training, and deployment for both single-arm and dual-arm setups. Yes, that Hugging Face — the one NVIDIA is acquiring. Your robot arm’s datasets and policy models live, from day one, in the world’s largest open-source AI community.
Even more important is the getting-started course: built on the reBot Arm B601-RS and NVIDIA Isaac, the Physical AI & VLA Full-Workflow Development Course walks you through a complete sim-to-real VLA workflow: real-world demonstration data collection → simulation data augmentation → VLA post-training → deployment to NVIDIA Jetson for real-world inference. For developers who want a head start in the Physical AI era, starting with this arm and this course is the right move.
Seeed: The Agile Manufacturing Behind MicroDuck and Reachy Mini
The recently viral MicroDuck — the $399 bipedal robot duck jointly launched by Hugging Face and Pollen Robotics that sold 5,000 units right after launch — is manufactured by Seeed. And before that, Pollen Robotics’ desktop companion robot Reachy Mini is likewise manufactured and sold by us.
This is Seeed’s other role in the open-source ecosystem: bridging the community and the manufacturing side, empowering the open-source community with the mature hardware capabilities of China’s Greater Bay Area.
The open-source community provides ideas and designs; we turn them into products that can ship, scale to volume, and spread widely across the developer community — from prototyping and small-batch pilot runs to rapid scale-up, all covered in one chain.
A Comprehensive Product Matrix Built on the NVIDIA Jetson Platform, Empowering All Kinds of Robot Applications
We’ve also built a complete path for Physical AI development. As an NVIDIA Jetson Elite Partner, Seeed has built a full product line around the Jetson platform, covering everything from entry-level to industrial:
- reComputer Classic series (now supporting the Jetson AGX Orin module): the classic desktop edge AI development workstation, ideal for rapid prototyping, model validation, and deployment
- reComputer Mini series (supporting Jetson AGX Orin / Orin NX / Orin Nano): compact yet powerful — a better platform for inference deployment on robot arms and wheeled robots
- reComputer Rugged series: rugged solutions for industrial environments — wide temperature range, dust-proof, and reliable
- reComputer J60 series: built for the real-time inference and training scenarios of humanoid robots with higher computing demands, oriented toward multimodal perception and interactive understanding
From real-time inference of quantized GR00T models on Orin Nano / Orin NX, to Isaac Sim simulation training, to the reBot B601 + VLM + Cosmos embodied intelligence system we showcased at the recent World Robot Conference (recap here) — this Jetson product map is the bridge that takes Physical AI from simulation to reality.
Start Your Physical AI Journey Today
NVIDIA teaming up with Hugging Face means the combination of open-source models with cutting-edge compute and simulation infrastructure will move faster than anyone expected. And what Seeed continues to do is what we’ve been doing for over a decade: making technology accessible to everyone.
For developers ready to hop on board, the route is already paved:
- Get started: one Physical AI & VLA course that takes you from data collection all the way to real-world deployment (scan the QR code below to join the group and get the complete learning materials)
- Hardware: one reBot Arm B601 robot arm — truly open-source, fully documented, with native LeRobot support
- Compute: one Jetson edge computing kit to run models in the real world
The future of Physical AI will be built together by all of us. We’re glad to walk alongside NVIDIA and Hugging Face — and we can’t wait to see more imagination and prototype ideas in the community!
👥 Join the Physical AI Community Group
If you’re following:
Embodied AI | Physical AI | Edge Computing | Agentic Robotics | Open-Source Robot Arms | LeRobot | Robot Teleoperation | Robot Vision | VLA
You’re welcome to join our Physical AI Community Group on Facebook and exchange, practice, and explore with more developers globally.
About Author
