Nvidia is positioning itself as the Android of generalist robotics, unveiling a comprehensive ecosystem at CES 2026 designed to become the foundational platform for physical AI. The company introduced a suite of new robot foundation models, simulation tools, and edge hardware aimed at accelerating the development of versatile, task-generalizing robots capable of learning and adapting in real-world environments. At the core of this strategy are several new open-source foundation models available on Hugging Face. Cosmos Transfer 2.5 and Cosmos Predict 2.5 are world models that generate synthetic data and evaluate robot policies in simulation, enabling faster and safer training. Cosmos Reason 2 is a vision-language model (VLM) that empowers AI systems to perceive, understand, and act in physical spaces. Complementing these is Isaac GR00T N1.6, a next-generation vision-language-action (VLA) model specifically built for humanoid robots. GR00T uses Cosmos Reason as its cognitive core, enabling whole-body coordination and the ability to perform complex, simultaneous tasks like moving and manipulating objects. To support this AI stack, Nvidia launched Isaac Lab-Arena, an open-source simulation framework hosted on GitHub. It consolidates task scenarios, training tools, benchmarks like Libero, RoboCasa, and RoboTwin, and standardized evaluation environments into a unified platform. This addresses a major bottleneck in robotics: the high cost, time, and risk associated with testing advanced robotic behaviors in physical settings. Isaac Lab-Arena allows developers to safely experiment and iterate in virtual environments before deploying on real hardware. Nvidia’s ecosystem is further strengthened by OSMO, an open-source command center that integrates the entire robotics workflow—from data generation and model training to deployment—across desktop and cloud environments. This infrastructure acts as the connective tissue between tools, enabling seamless collaboration and scalability. Powering the entire stack is the new Blackwell-based Jetson T4000, the latest addition to Nvidia’s Thor family of edge AI chips. Designed for on-device computation, the T4000 delivers 1,200 teraflops of AI performance and 64GB of memory while operating efficiently at just 40 to 70 watts. This makes it a cost-effective, high-performance option for real-time robotic control and AI inference in the field. Nvidia is also deepening its collaboration with Hugging Face, integrating its Isaac and GR00T technologies into the LeRobot framework. This connection links Nvidia’s 2 million robotics developers with Hugging Face’s 13 million AI researchers and builders, lowering the barrier to entry. The open-source Reachy 2 humanoid robot now natively supports the Jetson Thor chip, allowing developers to test and deploy models without being locked into proprietary systems. The company’s strategy is already gaining traction. Robotics has become the fastest-growing category on Hugging Face, with Nvidia’s models leading in downloads. Major robotics firms—including Boston Dynamics, Caterpillar, Franka Robots, and NEURA Robotics—are already adopting Nvidia’s hardware and software stack. By creating an open, accessible, and powerful platform, Nvidia aims to become the default infrastructure for the next generation of generalist robots, much as Android became the standard for smartphones. The goal is not just to sell chips, but to define the entire ecosystem of physical AI.