Nvidia's Jensen Huang Plays With Microduck Robots as Preorders Pass 20,000
Nvidia CEO Jensen Huang played with Microduck robots during his GTC Berlin keynote, and preorders for the devices have passed 20,000 units, TechCrunch reported. The company behind the robot says it expects deliveries in time for Christmas.
The Microduck is a 25-cm-tall biped robot, according to Reuters. Hugging Face debuted the device in August 2026 at a price of $399. Axios described it as a small, armless robot capable of picking up small objects. The Microducks are open-source devices produced by Pollen Robotics, which Hugging Face acquired in 2025, per TechCrunch.
Huang appeared on stage with four Microduck units during the GTC Berlin 2026 keynote, according to NDTV. Reuters reported that he spoke to the robots during a keynote focused on advances in AI and computing, and that he shared the stage with the tiny, duck-shaped, AI-trained machines developed by a French startup. In the same appearance, Huang hailed Europe's AI potential.
The software side is where Nvidia's stack comes in. Hugging Face and Pollen Robotics use NVIDIA Isaac Lab, Isaac Sim and Newton to build the Microduck, Nvidia's announcement states. Isaac Lab is the training framework, and Isaac Sim is a simulator that runs the robot in a virtual world before it moves in the physical one. The robots were trained on that stack to perform basic tricks, including walking, sitting and standing. During the keynote, Huang showed a video of an agent that set up a training course and simulated the robot in Isaac Sim, as described in Nvidia's keynote blog. A separate Nvidia post says the keynote also showed a Microduck running on NVIDIA Isaac Lab, Hermes Agent and Nemotron 3.5 Bolt, in a local AI demonstration.
In my view, the most useful detail for developers is the pipeline rather than the plush-toy framing. A sub-$400 open-source platform trained largely in simulation, with agents generating training courses and evaluation loops, gives hobbyists and small teams a practical route into embodied AI (AI that acts in the physical world) without building their own hardware from scratch. Whether preorders convert into shipped units by Christmas, and whether the open-source community extends the robot's behaviors beyond basic locomotion, will be the practical tests of that model.
The broader context here is the convergence of consumer hardware and the simulation tooling that has mostly lived in research and industrial settings. Nvidia is positioning Isaac and Newton as the common layer beneath both, and the Microduck is one of the few public examples where that layer is visible on a product a consumer can order. The keynote placed the robot alongside claims about European AI capacity, which frames it as more than a novelty. What the robots will actually do in homes, classrooms and labs will become clearer as the delivery cycle begins.
