Nvidia is trying to move artificial intelligence out of the data center and into machines that can operate in the physical world. Its new Project GR00T is aimed at humanoid robots, a category drawing work from companies including Google, Figure, Microsoft, Tesla, Boston Dynamics, and others.
The project fits into a larger race around embodiment: giving AI systems a body that can move, interact, and learn from the world around it. Nvidia’s bet is that general-purpose humanoid robots will need foundation models, specialized computing, simulation tools, and a wider partner ecosystem to become more capable.
What Nvidia announced
At Nvidia’s annual GTC conference keynote on Monday, CEO Jensen Huang described humanoid robotics as a major direction for the company’s AI work. In a statement, Huang said, “Building foundation models for general humanoid robots is one of the most exciting problems to solve in AI today.”
Huang also said, “The next generation of robotics will likely be humanoid robotics,” and, “We now have the necessary technology to imagine generalized human robotics.”
The centerpiece is Project GR00T, which stands for “Generalist Robot 00 Technology.” Nvidia describes it as a general-purpose foundation model for humanoid robots. The goal is for GR00T to act like an AI mind for robots, helping them learn skills and handle different tasks as needed.
Nvidia researcher Linxi “Jim” Fan called the project “our moonshot to solve embodied AGI in the physical world.” The wording matters because AGI, or artificial general intelligence, is described in the source as a poorly defined term that generally refers to hypothetical human-level AI, or beyond, that can learn any task a human could without specialized training.
Why humanoid robots are the focus
Robots do not have to look like people to be useful. The source article points to examples such as robot vacuum cleaners, autonomous weed pullers, industrial units used in automobile manufacturing, and research arms that can fold laundry.
Nvidia’s argument for humanoid robots is practical. Huang said, “In a way, human robotics is likely easier,” adding, “And the reason for that is because we have a lot more imitation training data that we can provide robots, because we are constructed in a very similar way.”
That means human movement can become training material for robot movement. A humanoid form also fits the physical world people have already built: tools, furniture, stairs, appliances, and other objects and interfaces are designed around human bodies.
GR00T is being designed to understand natural language and emulate human movements. Nvidia says that could help robots learn coordination, dexterity, and other abilities needed to navigate and interact with the real world more like a person.
The technology stack around GR00T
Project GR00T is not the only robotics announcement Nvidia made. The company also introduced Jetson Thor, a computer platform based on Nvidia’s Thor system-on-a-chip as part of the Blackwell GPU architecture.
According to the source, the SoC reportedly includes a transformer engine capable of 800 teraflops of 8-bit floating point AI computation for running models like GR00T. That positions Jetson Thor as a hardware piece meant to support a new generation of humanoid robots.
Nvidia also announced updates to Isaac Sim, its robotics platform for training physical interactions in thousands of parallel simulated physical environments. The updates include several pieces:
- Isaac Lab for reinforcement learning.
- OSMO for compute orchestration.
- Isaac Manipulator and Isaac Perceptor, collections of robotics pretrained models and libraries.
Isaac Manipulator provides AI capabilities for robotic arms. Isaac Perceptor adds multi-camera 3D vision capabilities for manufacturing robots. Together, these tools are aimed at giving robotics researchers a head start while working inside Nvidia’s development platform.
The partner ecosystem Nvidia wants
Nvidia says it is working with several leading humanoid robot companies on GR00T and related robotics efforts. The names listed include Apptronik, Agility Robotics, Boston Dynamics, Figure AI, Fourier Intelligence, and Sanctuary AI.
That strategy reflects Nvidia’s broader position in AI: it can provide components and tools that other companies build products on top of. In robotics, that could mean foundation models, chips, simulation environments, and pretrained capabilities that help robot makers move faster.
Figure AI is one example named in the source. Its Figure 01 humanoid robot gained attention on social media after a video demonstration in which it responded to verbal commands through an OpenAI back-end. Figure AI also raised $675 million in a funding round from Big Tech investors, including Nvidia.
The source frames that kind of company as exactly the customer Nvidia hopes to serve with its robotics platforms in the future.
Why the moonshot label fits
The promise behind embodied AI is large: if a capable humanoid body were controlled by AGI, it could suggest autonomous robotic assistants or workers. But the source also notes that some experts think true AGI is a long way off, which makes Nvidia’s goal aspirational as much as technical.
Fan wrote that “The GR00T model will enable a robot to understand multimodal instructions, such as language, video, and demonstration, and perform a variety of useful tasks.” He also wrote, “We are collaborating with many leading humanoid companies around the world, so that GR00T may transfer across embodiments and help the ecosystem thrive.”
The ethical stakes are also part of the story. The source notes that if Nvidia’s tools eventually help achieve embodied humanoid AGI, the result could bring deep ethical issues, including the potential for massive human job displacement.
Fan has also tied Project GR00T to his newly founded GEAR Lab, short for “Generalist Embodied Agent Research.” He has specialized in using simulations of physical worlds to train AI models, and that work is now extending to robotics. As Fan wrote, “At GEAR, we are building generally capable agents that learn to act skillfully in many worlds, virtual and real,” followed by, “Join us on the journey to land on the moon.”