NVIDIA built its reputation on graphics cards, then became the company practically running the AI boom on the back of its chips. In 2026, the company has been talking about robots almost as much as it talks about chatbots. At CES in January, CEO Jensen Huang said the “ChatGPT moment for robotics” had arrived. Here’s what that actually means, without the keynote-slide jargon.
The basic idea: robots need a brain, not just a body
Building a robot’s physical hardware, arms, legs, grippers, has gotten a lot easier over the last decade. What’s been missing is the “brain” that lets a robot figure out how to actually do a new task without being painstakingly programmed for that exact job. NVIDIA’s pitch is that it can supply that brain as a product, the way it already supplies the chips that power AI chatbots, rather than every robotics company having to build their own from scratch.
What GR00T actually is
GR00T is NVIDIA’s foundation model for humanoid robots, essentially a general-purpose AI brain that a robot maker can adapt to their specific machine, instead of training one from zero. It’s the robotics equivalent of how most companies now build on top of an existing AI model rather than training their own from scratch.
Why robots need a training ground before the real world
A robot that learns by trial and error in the real world is slow and expensive, mistakes mean broken hardware or actual safety risks. NVIDIA’s answer is Newton and Isaac, physics simulation tools that let a robot practice a task thousands of times in a realistic virtual environment before it ever touches a real object. Cosmos, announced alongside these, generates realistic simulated environments and scenarios to train against, effectively a digital obstacle course a robot can fail in safely, over and over, until it gets good.
This isn’t just a lab demo anymore
What’s notable about 2026 specifically is how many real companies are actually using this stack rather than just NVIDIA showing it off on stage. Boston Dynamics, Caterpillar, Universal Robots, Franka Robots, and NEURA Robotics have all been named as users of NVIDIA’s robotics tools. Caterpillar, in particular, has been applying it to edge AI on actual construction job sites, not a hypothetical future use case.
The hardware behind it
NVIDIA also used 2026 to push new chips specifically aimed at this space, including the Blackwell Ultra and Rubin platforms for heavier compute, and the smaller Jetson T4000 for running AI directly on a robot at the edge, rather than relying on a constant connection back to a data centre.
What this means for everyday life, realistically
Despite the showreel talk of robots that clean homes or assist in hospitals, most of the real, working deployments right now are industrial, warehouses, factories, construction sites, rather than anything sitting in a living room. The honest read: this is infrastructure being built for a wave of more capable robots, not evidence that one is showing up at your door next year.
