NVIDIA platform powers more than 80% of humanoid robots
English translation from Unibot. Source published 2026-08-22.
More than 80% of the world's humanoid robots use the NVIDIA platform. Chosun cites this figure from ABI Research. The indicator indicates a high concentration around one supplier of processors, models and simulation tools that are used to train and operate humanoid machines.
Demonstration at exhibitions
NVIDIA demonstrated its reach at GTC Taipei 2026, held on June 1. In the image behind CEO Jensen Huang, there were more than 20 humanoids from different countries.
Companies represented include Agility Robotics and Figure AI from the US, Unitree and AgiBot from China, Neura Robotics from Germany, and Rainbow Robotics and LG Electronics from South Korea.
A similar picture was observed at Automate 2026 in Chicago on June 22. The humanoids were labeled Accelerated by NVIDIA. According to Chosun, this includes machines using NVIDIA processors, as well as robots trained in the company's virtual environments.
What's included in the platform
NVIDIA's humanoid platform goes far beyond chips. In March 2024, the company introduced the robotics foundation model GR00T. It became part of a broader system for teaching robots and performing learned skills on physical hardware.
The system combines AI models for robots, virtual training environments and processors for on-board operation. For humanoid manufacturers, this is a ready-made alternative to developing every element of the software and computing stack in-house.
Huang said NVIDIA works with almost every company developing robotics systems.
ABI Research's assessment requires clarification. Chosun's definition includes robots using NVIDIA hardware and those trained with its software. These are different forms of dependency. The report does not distinguish between humanoids who rely on NVIDIA for work and those who use its tools only at the development stage.
The balance of power in the industry
In the Morgan Stanley Humanoid 100 report cited by Chosun, 10 of the top 15 humanoid intelligence companies are based in the United States. The only Chinese company on the list is Baidu.
NVIDIA is not the only American developer of universal robot control models. Google DeepMind's Gemini Robotics 2 is said to control walking, balance, and hand manipulation throughout a humanoid's body. Physical Intelligence's π0.7 is designed to recombine learned actions when the robot encounters an unfamiliar task.
Chinese humanoid manufacturers, including AgiBot and Unitree, are developing their own robotic intelligence systems. But the report notes that their chips and virtual learning infrastructure remain heavily dependent on American technology.
Kim Ik-jae, director of KIST's AI and Robotics Research Institute, says China is trying to close the gap through industrial data and field testing, while the US maintains an edge in advanced robotic intelligence.
This concentration gives NVIDIA influence over competing humanoid programs. A more detailed breakdown of the 80% estimate by CPU usage, simulations and models would show how deeply this impact reaches into actual running machines.