NVIDIA Expands Physical AI Ecosystem with New Models and Industrial Partnerships
AI-generated image: synthetic visual, not an actual depiction of events, people, or locations.
Key Takeaways
- NVIDIA introduced Cosmos 3 and Isaac Lab 3.0 to provide a unified framework for synthetic world generation and robot learning.
- Industrial giants including ABB, FANUC, and KUKA are using NVIDIA digital twins and Jetson modules to optimize production lines.
- The GR00T N1.7 model is now available for commercial licensing, while the upcoming GR00T N2 model shows significant improvements in task success rates.
- Strategic partnerships in healthcare and logistics are bringing physical AI to surgical systems and autonomous warehouse operations.
NVIDIA revealed several software updates and strategic alliances at GTC 2026 to promote the adoption of physical AI. Key releases include the Cosmos world foundation model and enhancements to the Isaac simulation suite, which provide the computational framework for robots to navigate and interact with their surroundings autonomously. CEO Jensen Huang described the company's full-stack platform as a foundational element for the global robotics industry, supporting infrastructure in factories, logistics, and transportation.
Major robotics firms, such as FANUC, ABB Robotics, KUKA, and YASKAWA, are adopting NVIDIA Omniverse and Isaac tools to improve virtual commissioning. By using high-fidelity digital twins, these companies can refine production lines before physical implementation. They are also deploying NVIDIA Jetson modules within their robot controllers to facilitate real-time AI processing at the edge.
The new Cosmos 3 model integrates synthetic data generation with vision-based reasoning and action simulation. Additionally, Isaac Lab 3.0 has entered early access, utilizing the Newton 1.0 physics engine to support large-scale training on DGX infrastructure. These advancements target the development of general-purpose robot intelligence, allowing machines to master new tasks with minimal retraining.
Humanoid developers, including Boston Dynamics, Figure, and Agility, are leveraging the Cosmos and Isaac platforms for robot training and validation. NVIDIA also released the GR00T N1.7 model for commercial use and previewed GR00T N2, a next-generation foundation model. According to the company, GR00T N2 doubles the success rate of robots in unfamiliar environments compared to previous vision-language-action models and currently leads benchmarks on MolmoSpaces and RoboArena.
