Nvidia Unveils 'Physical AI' Foundation Models for Humanoid Robotics

Nvidia announces Cosmos, an AI model that understands physics, alongside the Vera Rubin superchip to power the next generation of humanoid robotics.

At the CES 2026 keynote, Nvidia CEO Jensen Huang introduced 'Cosmos,' a groundbreaking AI foundation model designed specifically for the physical world. Unlike previous models focused on text or static images, Cosmos is trained to understand the laws of physics, enabling robots to perceive and interact with their environments with human-like intuition.

The announcement also featured the 'Vera Rubin' superchip platform, which is now in full production to provide the massive computational power required for real-time physical AI. This hardware-software synergy allows for the training of humanoid robots in virtual 'digital twin' environments before they are deployed to perform complex tasks in factories, hospitals, and homes.

Industry experts believe this marks the end of the 'chatbot era' and the beginning of the 'agentic hardware' era. With major partnerships announced with firms like Siemens, Nvidia is positioning itself as the central nervous system for a new generation of autonomous machines that can troubleshoot assembly lines and assist in domestic chores without human intervention.

Published on January 9, 2026