NVIDIA Unveils Cosmos 3 and GR00T N2 as Robotics Giants Adopt Physical AI Stack
Jessie A Ellis Mar 16, 2026 22:16
NVIDIA announces major robotics partnerships with ABB, FANUC, and KUKA while releasing new world models and simulation frameworks at GTC 2026.
NVIDIA dropped a significant expansion of its robotics platform at GTC 2026, announcing partnerships with industrial giants controlling over 2 million installed robots worldwide while unveiling Cosmos 3—the company's first unified world foundation model for robot intelligence.
The stock traded at $181.84 on March 16, up 0.88% as investors digested the scope of the announcement.
The Industrial Alliance
ABB Robotics, FANUC, YASKAWA, and KUKA—the four companies that dominate global industrial robotics—are now integrating NVIDIA Omniverse libraries and Isaac simulation frameworks into their virtual commissioning solutions. That's not a pilot program. These firms are embedding NVIDIA Jetson modules directly into robot controllers for real-time AI inference at the edge.
"Physical AI has arrived—every industrial company will become a robotics company," Jensen Huang said during the keynote. The statement sounds like typical tech hyperbole until you consider who showed up to validate it.
What's Actually New
Three releases matter here. Cosmos 3 unifies synthetic world generation, vision reasoning, and action simulation into a single foundation model. Think of it as giving robots the ability to imagine scenarios before encountering them.
Isaac Lab 3.0 enters early access with the Newton physics engine 1.0, enabling faster large-scale robot learning on DGX-class infrastructure. The upgrade adds multiphysics simulation for complex dexterous manipulation—robots handling cables, assembling components, tasks that previously required extensive manual programming.
GR00T N1.7 is now available with commercial licensing, while Huang previewed GR00T N2, claiming it helps robots succeed at new tasks in unfamiliar environments more than twice as often as leading vision language action models. GR00T N2 currently ranks first on MolmoSpaces and RoboArena benchmarks, with availability expected by year-end.
Real Deployments, Not Demos
Skild AI partnered with Foxconn to deploy generalized robot intelligence on NVIDIA Blackwell production lines—meaning AI-driven dual-arm manipulators are now handling actual chip manufacturing assembly. Samsung's assembly robots are using the Newton physics engine to master cable handling in simulation before deployment.
Healthcare robotics got attention too. CMR Surgical is using Cosmos-H simulation to train its Versius surgical system, while Medtronic explores NVIDIA IGX Thor for surgical robotics requiring mission-critical precision.
The Ecosystem Play
Microsoft Azure, Nebius, CoreWeave, and Alibaba Cloud are all integrating various pieces of NVIDIA's physical AI stack. Disney built Kamino, a GPU-accelerated physics simulator on NVIDIA's Warp framework, to train robot policies for its animatronic characters—including an Olaf robot debuting at Disneyland Paris on March 29.
The Hugging Face partnership connects NVIDIA's claimed 2 million robotics developers with 13 million AI builders through the LeRobot open source framework.
For investors tracking AI infrastructure buildout, this announcement represents NVIDIA extending its dominance from training and inference into the physical world. The question isn't whether robots will run on NVIDIA silicon—that's increasingly settled. The question is how fast manufacturing, logistics, and healthcare can absorb these systems at production scale.
Image source: Shutterstock- nvidia
- robotics
- physical ai
- gr00t
- cosmos
