Nvidia lays out full-stack push to become the Android of general-purpose robotics
From foundation models and simulation software to edge hardware, Nvidia’s CES 2026 announcements signal a bid to dominate the emerging physical AI ecosystem.
ADVERTISEMENT
Nvidia has unveiled a broad suite of robot foundation models, simulation frameworks and edge computing hardware at CES 2026, marking a decisive step toward positioning itself as the default platform for general-purpose robotics as artificial intelligence increasingly moves from the cloud into the physical world.
The announcements reflect a wider shift underway in the AI industry, where advances in sensors, simulation and multimodal models are enabling machines to reason, plan and act across real-world environments rather than perform narrowly defined tasks. Nvidia is betting that this transition will mirror the smartphone era, where a small number of platforms came to dominate development.
At the core of the company’s strategy is a new set of open robot foundation models, released on Hugging Face, designed to support generalist robotics. These include Cosmos Transfer 2.5 and Cosmos Predict 2.5, world models built for synthetic data generation and policy evaluation in simulated environments, as well as Cosmos Reason 2, a vision-language model intended to help machines understand and respond to physical surroundings.
Nvidia also introduced Isaac GR00T N1.6, its latest vision-language-action model tailored for humanoid robots. Built on top of Cosmos Reason, GR00T enables coordinated, whole-body movement, allowing humanoids to manipulate objects while moving, a capability that has been difficult to achieve at scale.
To support training and testing, Nvidia launched Isaac Lab-Arena, an open-source simulation framework hosted on GitHub. The tool is designed to reduce the cost and risk of validating increasingly complex robotic skills by allowing developers to train and benchmark robots in virtual environments before deploying them in the real world. The framework brings together task libraries, training workflows and benchmarks such as Libero, RoboCasa and RoboTwin, aiming to establish a shared industry standard.
Complementing the software stack is Nvidia OSMO, an open-source orchestration system that connects data generation, training and deployment across desktop and cloud setups. On the hardware side, Nvidia introduced the Blackwell-based Jetson T4000, a new edge computing card delivering up to 1,200 teraflops of AI performance with 64 GB of memory while operating within a 40–70 watt power envelope.
Also read: Honda Motorcycle December sales jump 45% YoY to 4.46 lakh units
Nvidia is also expanding its collaboration with Hugging Face to lower the barrier to robotics development. The integration brings Nvidia’s Isaac and GR00T tools into Hugging Face’s LeRobot framework, linking Nvidia’s robotics developer base with Hugging Face’s broader AI community. The open-source Reachy 2 humanoid robot is now compatible with Nvidia’s Jetson Thor platform, giving developers flexibility to experiment without being locked into proprietary systems.
Early indicators suggest the approach is gaining traction. Robotics has become the fastest-growing category on Hugging Face, with Nvidia’s models among the most downloaded. Several industrial and robotics firms, including Boston Dynamics, Caterpillar, Franka Robots and NEURA Robotics, are already building on Nvidia’s platform, reinforcing the company’s ambition to become the foundational layer for physical AI.
