Nvidia jumps into humanoids with GR00T foundation model

Humanoid robots are having a massive moment, pulling in venture dollars while fighting off plenty of healthy skepticism. Now, Nvidia is stepping in to make sure all those metal bodies run on its digital brains.

At its annual GTC conference this week, CEO Jensen Huang unveiled Project GR00T, a general-purpose foundation model built specifically for humanoid robots, alongside dedicated silicon to power them.

A Marvel-sized robotics alliance

Nvidia basically recruited nearly every major player in the humanoid race under one tent. The partner roster includes 1X Technologies, Agility Robotics, Apptronik, Boston Dynamics, Figure AI, Fourier Intelligence, Sanctuary AI, Unitree Robotics, and XPENG Robotics. The only obvious absentee is Tesla, which is keeping its Optimus project strictly in-house.

Agility Robotics co-founder Jonathan Hurst noted that human-centric robots like Digit are poised to change labor forever, and modern AI will fast-track that transition. Sanctuary AI CEO Geordie Rose added that embodied AI is too critical to build in silos, justifying their long-term bet on Nvidia's ecosystem.

Custom silicon for the physical world

To give these digital brains a home, Nvidia also introduced Jetson Thor, a brand new computer engineered specifically for humanoid workloads and generative simulation.

The Thor SoC packs a next-generation GPU based on Nvidia's Blackwell architecture. Its transformer engine pumps out 800 teraflops of 8-bit floating point performance, built to handle multimodal foundation models on the fly. It also packs integrated safety processors and 100GB ethernet bandwidth to make hardware integration easier for engineers.

Nvidia didn't stop at two-legged machines. It launched Isaac Manipulator, bringing foundation models and an 80x path-planning speedup to robotic arms from players like Universal Robots and Yaskawa. It also debuted Isaac Perceptor, bringing multi-camera 3D surround vision to mobile logistics robots with partners like BYD.

Why it matters

True general-purpose robotics is still years away, and skepticism around humanoid utility remains completely valid. But for hardware builders and developers, Nvidia just handed them a turnkey platform, drastically lowering the barrier to entry for vision, dexterity, and movement models.

Strategically, this is classic Nvidia. Instead of betting on one mechanical design, it is positioning itself as the indispensable pick-and-shovel provider for the entire robotics industry. Whether Figure AI, Agility, or Boston Dynamics wins the warehouse floor, Nvidia gets paid on both the software and chip layers.

This setup puts enormous pressure on solo operators like Tesla, who must match an entire ecosystem's worth of shared R&D. The real test, however, is whether 800 teraflops inside a battery-constrained mechanical frame can survive real-world chaos, but for now, Nvidia is effectively dictating the terms of the embodied AI race.