NVIDIA Bets Big on Humanoids: Physical AI Takes Center Stage at GTC 2026

If there was any doubt that NVIDIA has moved beyond chips and into the very DNA of the robotics industry, GTC 2026 erased it. Jensen Huang took the stage in San Jose with a bold vision: physical AI — the idea that the next great frontier of artificial intelligence isn't just in data centers, but in machines that move, grasp, and operate in the real world. And NVIDIA wants to power all of it.

110+ Partners and Counting

The headline number from GTC 2026 is staggering: NVIDIA announced collaborations with more than 110 global robotics companies, spanning humanoid manufacturers, simulation tool providers, and AI software developers. The roster reads like a who's-who of the emerging humanoid industry — from Boston Dynamics and Agility Robotics to Figure AI, 1X Technologies, Apptronik, Fourier Intelligence, Sanctuary AI, UBTECH, and Unitree. By weaving itself into this many partnerships simultaneously, NVIDIA is executing the same platform strategy that made it dominant in cloud AI: become the standard infrastructure layer that everyone builds on top of.

Isaac GR00T: A Foundation Model for Robot Bodies

Central to NVIDIA's robotics push is Isaac GR00T (Generalist Robot 00 Technology), a foundation model designed specifically for humanoid robots. Unlike large language models that process text, GR00T is trained on multimodal data — video, sensor streams, and teleoperation demonstrations — to give robots a general understanding of how to perceive and interact with the physical world. At GTC 2026, NVIDIA expanded the GR00T ecosystem with new synthetic data generation pipelines and improved sim-to-real transfer techniques, dramatically reducing the amount of costly real-world data collection needed to teach a robot a new skill.

Isaac Sim and Isaac Lab: Training Grounds for the Real World

NVIDIA's Isaac Sim, built on the Omniverse platform, received major updates at GTC 2026. The simulation environment now supports faster-than-real-time physics rendering, enabling robotics teams to run millions of training iterations overnight that would take months to gather on physical hardware. Paired with Isaac Lab — a reinforcement learning framework tuned for legged and dexterous robots — the stack gives companies a full pipeline: design a humanoid, train it in simulation, fine-tune on real hardware, and iterate. Several partner companies demonstrated robots at GTC that had been trained almost entirely in Isaac Sim before their first real-world steps.

"The robotics industry is at an inflection point. Physical AI will be as transformational as the internet — and NVIDIA intends to be the computing platform that makes it possible."

— Jensen Huang, CEO, NVIDIA, GTC 2026 Keynote

Why This Matters for the Humanoid Race

The humanoid robotics market is intensely competitive, with dozens of companies racing to deploy bipedal, dexterous machines in factories and warehouses. But building a humanoid is only half the challenge — training it to perform reliably in unstructured environments is the other. NVIDIA's pitch is that its GPU hardware, simulation tools, and foundation models compress that second half dramatically. By standardizing on NVIDIA's stack, robotics companies can avoid reinventing simulation infrastructure and instead focus their engineering on the hardware and application layer. For the broader industry, this could mean faster time-to-deployment and a rising tide of capable humanoids hitting the factory floor sooner than anyone expected.

The Bigger Picture: Physical AI as a New Computing Era

Jensen Huang framed physical AI not merely as a robotics trend but as the third wave of AI after perception (image recognition, speech) and generative AI (LLMs, diffusion models). Each wave demanded new hardware, new software, and new infrastructure — and NVIDIA has positioned itself to supply all three for this one. With Blackwell GPUs powering training clusters, Jetson Thor modules targeted at humanoid onboard compute, and the Isaac software stack tying it together, NVIDIA is betting that the machines of the next decade will run on its silicon from the warehouse floor to the cloud.