Darius Baruo
Jan 29, 2026 17:28
NVIDIA releases new open-source physical AI models and frameworks at CES 2026, with Caterpillar, LEM Surgical, and NEURA Robotics deploying systems built on the stack.
NVIDIA unveiled a comprehensive suite of open physical AI models and frameworks at CES 2026, with major industrial partners already deploying systems built on the technology. The announcement comes as NVDA trades at $189.50, up 1.8% on the day, with the company’s market cap sitting at $4.63 trillion.
The new toolkit spans the entire robotics development pipeline—from simulation and synthetic data generation to edge deployment. Caterpillar, LEM Surgical, and NEURA Robotics demonstrated working systems at the show, marking a shift from lab prototypes to production-ready machines.
Industrial Deployments Take Center Stage
Caterpillar’s Cat AI Assistant now runs on NVIDIA’s Jetson Thor edge module, bringing voice-activated controls into heavy equipment cabs. Operators can adjust safety parameters and receive step-by-step guidance through natural language. Behind that interface, Caterpillar uses Omniverse libraries to simulate job-site layouts and multi-machine workflows before pushing changes to actual fleets.
LEM Surgical brought something more consequential: an FDA-cleared robotic surgical system already in routine clinical use for spinal procedures. The Dynamis system uses Jetson AGX Thor for compute and NVIDIA’s Cosmos Transfer world model to generate synthetic training data. The dual-arm humanoid robot mimics surgeon dexterity while reducing physical strain on surgical teams.
NEURA Robotics is training its 4NE1 humanoid and MiPA service robots in OpenUSD-based digital twins through Isaac Sim and Isaac Lab. The company has partnered with SAP to integrate Joule AI agents with its robots, using NVIDIA’s Mega Omniverse Blueprint to simulate complex operational scenarios before real-world deployment.
The Open-Source Play
NVIDIA’s strategy centers on making these tools freely available. The company released Isaac Lab-Arena for policy evaluation, the Alpamayo portfolio for autonomous vehicles, and integration with Hugging Face’s LeRobot ecosystem. Developers can now access Isaac GR00T N1.6 models directly within LeRobot workflows.
This open approach isn’t altruistic—it’s ecosystem lock-in through different means. When developers build on NVIDIA’s simulation frameworks, they’re more likely to deploy on NVIDIA hardware. The Hugging Face partnership puts GR00T models in front of that platform’s robotics community, and Reachy 2 humanoid compatibility with Jetson Thor creates another hardware pathway.
AgiBot is using Cosmos Predict 2 as the backbone for its Genie Envisioner platform, generating action-conditioned videos that help policies transfer more reliably to physical robots. Intbot is using Cosmos Reason 2 to give social robots contextual awareness beyond scripted interactions.
What Traders Should Watch
CEO Jensen Huang called this “the ChatGPT moment for physical AI” during his CES keynote on January 10. That’s marketing hyperbole, but the industrial deployments suggest real traction. An FDA-cleared surgical robot and Caterpillar integration represent enterprise revenue streams, not just research partnerships.
The Alpamayo release on January 5 specifically targets autonomous vehicle edge cases—the rare scenarios that have stalled large-scale AV deployment. If NVIDIA’s simulation tools can help solve the long-tail problem that’s plagued the industry, the company captures value from every automaker and robotics firm building on its stack.
Near-term catalysts include adoption metrics from the Cosmos Cookoff developer challenge and any production announcements from partners like NEURA Robotics or AgiBot. The surgical robotics angle deserves particular attention—healthcare applications typically command premium margins and face less regulatory uncertainty than consumer robotics.
Image source: Shutterstock
