Nvidia physical AI stack powers next wave of robots at CES 2026

Nvidia is using CES 2026 to make a very specific argument: the next phase of AI isn’t just bigger text models, it’s “physical AI” — robots and machines that perceive, reason and act in the real world. The Cosmos 2.5 world models, Isaac GR00T N1.6 vision‑language‑action model, Isaac Lab‑Arena benchmarking framework, and the Blackwell‑based Jetson T4000 module together form an end‑to‑end stack for training and deploying generalist robots. When you see Boston Dynamics, Caterpillar and a Porsche‑designed NEURA humanoid all standardizing on this ecosystem, it starts to look like CUDA for embodied AI. ([investor.nvidia.com](https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Releases-New-Physical-AI-Models-as-Global-Partners-Unveil-Next-Generation-Robots/default.aspx?utm_source=openai))

For the AGI race, this matters because it extends the “foundation model + data + hardware” flywheel into the physical world. These open models and datasets on Hugging Face lower the barrier for labs and startups to train sophisticated robot policies without owning their own fleet and test track. Nvidia’s partnership with Hugging Face around LeRobot, and with Microsoft’s Azure Robotics Accelerator, suggests it wants to be the default substrate for the emerging ecosystem of general‑purpose robots in factories, logistics, healthcare and homes. If that consolidation continues, the same concentration concerns we see around GPU supply could reappear around physical AI platforms, but in the near term it will likely accelerate experimentation and deployment.