Chinese humanoid robot makers - including Unitree Robotics, AgiBot, Galbot, Engine AI and UBTech Robotics - have become early adopters of Nvidia's latest product that enables their machines to have real-time, intelligent interactions with people.
The Jetson AGX Thor developer kit and production modules, which Nvidia launched on Monday, are computers designed to power robots across industries including manufacturing, logistics, transport, healthcare, agriculture and retail.
Built on an Nvidia Blackwell graphics processing unit (GPU) that features 128 gigabytes of memory, Jetson AGX Thor delivered up to 2,070 FP4 teraflops of AI computing power to run the latest artificial intelligence models - all within a 130-watt power envelope, the company said.
Do you have questions about the biggest topics and trends from around the world? Get the answers with SCMP Knowledge, our new platform of curated content with explainers, FAQs, analyses and infographics brought to you by our award-winning team.
FP4, the 4-bit floating point format introduced with Nvidia's Blackwell GPU architecture, is touted to reduce computational demand in AI training and inference. A teraflop, which represents 1 trillion floating-point operations per second, is a measure of a computer's ability to perform complex mathematical calculations involving decimals with high precision.
"Jetson Thor brings a huge leap in computing power, enabling robots to be more agile, make decisions faster and achieve a higher level of autonomy," said Unitree founder and CEO Wang Xingxing in Nvidia's statement.
Nvidia's latest offering for Chinese humanoid robot developers shows the US semiconductor giant's commitment to the mainland market, despite a tumultuous US-China tech war and Beijing's security concerns about its H20 chips.
A pair of robots from Unitree Robotics compete in a kickboxing match at the inaugural World Humanoid Robot Games in Beijing on August 15, 2025. Photo: Reuters alt=A pair of robots from Unitree Robotics compete in a kickboxing match at the inaugural World Humanoid Robot Games in Beijing on August 15, 2025. Photo: Reuters>
In a speech given in June last year, Nvidia founder and CEO Jensen Huang said the next step of the new AI era was to push forward robotics. He said Nvidia had built an advanced development platform to create and train AI for various robotic applications.
Nvidia's latest offering comes as a number of Chinese humanoid robot makers enter mass production this year.
China's robotics market is set to grow at an annual rate of 23 per cent to US$108 billion by 2028 from US$47 billion in 2024, solidifying the country's dominant position in the fast-developing sector, according to a note published in June by Morgan Stanley.
In a Monday post on its WeChat account, UBTech said the Jetson AGX Thor developer kit and production modules "seamlessly integrates cutting-edge embodied intelligence technology with exceptional edge computing capabilities".
UBTech also said its Walker S2, the world's first humanoid robot that can change its own batteries, uses Nvidia's Jetson AGX Thor.
"We've built Jetson Thor for the millions of developers working on robotic systems that interact with and increasingly shape the physical world," Huang said in Monday's statement. He touted the new product as "the ultimate supercomputer to drive the age of physical AI and general robotics".
The product's system-on-module solved one of the most significant challenges in robotics: running multiple AI workflows to enable robots to have real-time, intelligent interactions with people and the physical world, according to Nvidia.
Compared with its predecessor, the Jetson Orin, the Jetson AGX Thor delivered up to 7.5 times higher AI computing power and 3.5 times greater energy efficiency to run any generative AI model, the company said.
Other early Jetson AGX Thor adopters include industry leaders Agility Robotics, Amazon Robotics, Boston Dynamics, Caterpillar, Figure, Hexagon, Medtronic and Meta Platforms. Companies such as OpenAI, John Deere and Physical Intelligence are evaluating the product to advance their physical AI capabilities.
The Jetson AGX Thor developer kit starts from US$3,499. The T5000 production modules cost US$2,999 each for a buyer who orders at least 1,000 units.
This article originally appeared in the South China Morning Post (SCMP), the most authoritative voice reporting on China and Asia for more than a century. For more SCMP stories, please explore the SCMP app or visit the SCMP's Facebook and Twitter pages. Copyright © 2025 South China Morning Post Publishers Ltd. All rights reserved.
Copyright (c) 2025. South China Morning Post Publishers Ltd. All rights reserved.
Comments