《 Nvidia recruits troops and lays out trillion dollar business in physics, AI, and robotics in China ❗ ️️ Brothers and sisters, Nvidia and Huang Renxun have made another big move. This time, they are recruiting for talents. On June 30th, the NVIDIA Robotics team opened recruitment in China, with positions released in Beijing, Shanghai, and Shenzhen simultaneously, focusing on four lines: embodied intelligence, simulation, deployment, and solution architecture. The job keywords are very interesting: humanoid robot electrical solution, mechanical solution, full body sensing, full body control, mobile operation Isaac Lab…… On the surface, it appears to be a regular recruitment. But looking at the right position, this is Nvidia telling the market that physical AI is going to enter Chinese factories from the press conference. 01. What is Physical AI? In the past two years, the hottest stories of AI have been in big models, computing power, and data centers. The familiar terms for investors are GPU, server, liquid cooling, optical module, and electricity. But Nvidia is telling you now: AI only knows how to write, draw, and generate videos, it's not enough. The next step is for AI to be able to see, walk, grasp, and work. This is the "physical AI" repeatedly emphasized by Huang Renxun. In March 2025, Huang Renxun asserted that "generative AI has become the past, and the future belongs to 'agent AI' and 'physical AI'." Simply put, AI on the screen should enter factories, warehouses, hospitals, cars, and homes. It's not just about answering questions, but understanding the real environment and turning judgments into actions. So, the most crucial aspect of this recruitment is not the word 'robot', but the combination of positions. Hardware, electrical, sensing, control, simulation, and deployment are all recruited together - what Nvidia is adding is not a demonstration prototype, but a complete set of infrastructure for robots from training to landing. Nvidia's layout in the field of robotics actually began in 2014, with a clear core strategy - not to compete with machine manufacturers for market share, but to create an "Android ecosystem" in the field of robotics, providing a complete toolchain from computing power to development platforms to models. The GR00T humanoid basic model has been adopted by manufacturers such as Figure AI and Yushu Technology, and the Cosmos 3 world model is the world's first fully open-source physical AI training platform, which can shorten the training cycle from months to days. 02. What to watch in the robot market? ——Four value chains, A-shares and US stocks are so closely monitored Following Huang Renxun's layout, the market for robots cannot be solely judged by machine manufacturers. The whole machine factory is the easiest to be hyped up, but the competition is the most intense, and the gross profit margin is unstable. What we should really focus on are the four upstream value chains. 1、 Perception system Robots entering the real world must first perceive the world. Visual, sensory, tactile IMU、 Lidar - Without high-quality sensing, intelligence is like a castle in the air. A-shares worth tracking: Obi Zhong Guang（ http://688322.SH ）Over 10 self-developed chips have been taped out, and multiple 3D cameras have been integrated into the NVIDIA Isaac Sim platform, playing the role of a "visual perception base" in the humanoid robot reference platform. Adaptation verification has been completed with system level modules such as NVIDIA Jetson Thor; Keli sensor（ http://603662.SH ）Continuously promoting research and customer expansion around six axis force/torque sensors and joint torque sensors; An Peilong（ http://301413.SZ ）Hanwei Technology（ http://300007.SZ ）It has also been listed as a key target in the sensor sector by multiple securities firms. 2、 Execution system Servo motors, reducers, screws, and dexterous hands - the core of humanoid robots from display to work lies in whether their joints can operate with high precision, low energy consumption, and long lifespan. The cost of actuators accounts for one-third or even higher of the overall material cost of the machine. A-shares worth tracking: Green harmonics（ http://688017.SH ）Leading domestic harmonic reducer, the reducer and screw account for approximately 35% of Optimus' total cost; Wuzhou New Year（ http://603667.SH ）We have developed a full range of robot supporting bearings, which are compatible with multiple mainstream robot companies both domestically and internationally; Zhaowei Electromechanical（ http://003021.SZ ）Focus on core components such as dexterous hands and joint modules; Sanhua Intelligent Control（ http://002050.SZ ）Tuopu Group（ http://601689.SH ）Tier 1 supplier of Tesla Optimus, core target of actuator assembly. 3、 Edge computing power and control system Robots cannot wait for cloud answers for every action. It must complete fast inference, perception fusion, and motion control locally. Jetson Thor is Nvidia's answer sheet - with a maximum AI computing power of 2070 FP4 TFLOPS, 128GB of memory, and a power consumption of only 40-130W, specifically designed for humanoid robots and physical AI. NVIDIA has also partnered with Yushu Technology to launch the world's first open humanoid robot reference design, Isaac GR00T, which integrates Yushu H2 Plus body and Jetson Thor computing platform, with the goal of shortening the development cycle from a few months to a few weeks. This stage is currently led by Nvidia, with fewer pure A-share targets and more PCB and structural components supporting the industrial chain. 4、 Simulation and Data Ecology Future robotics companies are not competing on hardware, but on who has more high-quality motion data and who can reliably migrate virtual training to real factories. A-shares worth tracking: Suochen Technology（ http://688507.SH ）Zhongwang Software（ http://688083.SH ）Listed as a key target in the physical simulation software phase by institutions; Soft Communication Power（ http://301236.SZ ）There are layouts in the fields of digital twins and industrial simulation. In terms of the US stock market, the core target is Nvidia itself - it is a three in one player that combines physical AI underlying computing power, world models, and simulation platforms. Tesla (TSLA) is also worth paying attention to. 03. Final Summary Generative AI solves mental automation, while physical AI solves action automation. The former changes the office, while the latter changes the factory and supply chain. In the past decade, capital has been chasing the entrance to the digital world. In the next decade, capital will reprice - who holds the execution power in the real world. What Nvidia really wants to do is not sell a robot, but become the underlying platform for all robots. Physical AI is the inevitable path for AI technology to move from virtual to physical, and this track is just beginning. Whoever can bring robots from the laboratory into the workshop may get the ticket to the next round of industrial revolution. ---Mr. Xiaolong NVIDIA Physics AI Humanoid Robot Body Intelligence Huang Renxun Robot Industry Chain Obi Zhongguang Green Harmonic JetsonThor