The humanoid robot sector has posted notable gains recently, driving "physical AI" to gradually become a new area of market focus.
Over the past few years, artificial intelligence applications have been concentrated mainly in the digital world—large models, intelligent assistants, and content generation. Now, physical AI is pushing AI beyond the screen and into real-world scenarios such as factories, warehouses, cars, and homes, extending AI applications from processing information toward transforming the physical world.
Physical AI can be simply understood as giving AI the ability to perceive its real environment, understand tasks, plan actions, and execute operations. Typically, physical AI needs to combine cameras and various sensors to gather environmental information, use models and algorithms to determine the next action, and then rely on motors, actuators, wheels, or robotic arms to complete movements, adjusting based on results. Unlike AI that remains largely in text, images, or video, physical AI emphasizes the connection between perception, understanding, decision-making, execution, and feedback, requiring systems not only to judge "what it is" but also to clearly determine "how to do it" and reliably complete actions in real environments.
Overall, physical AI is a technology system formed by the integration of artificial intelligence, robotics, sensors, automatic control, and computing technology. Classified by application carrier, physical AI can include robots, intelligent mobility systems, industrial intelligent equipment, and life-service devices. Robots include industrial robots, quadruped robots, service robots, and humanoid robots; intelligent mobility systems include autonomous vehicles, unmanned delivery devices, and drones; industrial intelligent equipment is mainly used for production, inspection, patrol, and maintenance; life-service devices target scenarios such as households, commerce, and healthcare.
From the perspective of the industry chain, the upstream mainly consists of chips, computing power equipment, sensors, and precision components. Cameras, LiDAR, millimeter-wave radar, and force and tactile sensors help devices identify their surroundings, judge distances, and perceive contact states; processors provide computing power; motors, reducers, servo systems, ball screws, and actuators convert instructions into actual movements. The upstream segment is the foundation for physical AI to achieve perception and execution. As applications such as humanoid robots, autonomous driving, and unmanned delivery accelerate validation, performance requirements, reliability, and cost control for upstream core components will continue to rise.
The midstream includes AI models, control algorithms, simulation software, robot bodies, and system integration. Models are responsible for understanding the environment and tasks, control systems coordinate movements, simulation platforms can test different situations in virtual environments, and complete-machine and integration companies combine technologies into products that can be put into use. The midstream is a key link connecting upstream hardware with downstream scenarios. On one hand, models and algorithms need to adapt to different sensors and actuators; on the other hand, complete-machine companies need to carry out system integration and engineering implementation based on specific scenarios.
The downstream covers numerous industries including industry, business services, home services, and healthcare, and is the key to physical AI ultimately realizing value. For physical AI, future development depends on whether AI can move from understanding tasks to reliably completing them. The real environment is far more complex than the digital world. Robots need to cope with changes in lighting, moving objects, uneven ground, and human interference. They must not only do things correctly but also operate safely and continuously over extended periods.
Recently, Wang Xingxing of Unitree Technology said that with the release of GPT-6, physical AI is already showing signs of being broken through, and a breakthrough is very likely to occur in the next one to two years or two to three years. As related technologies iterate, technological breakthroughs are expected to gradually accelerate.
From a commercialization path perspective, physical AI may first land in scenarios with clear tasks and relatively fixed processes, such as factory handling, production-line assistance, warehouse sorting, and energy inspection. These scenarios have relatively clear operational requirements and strong environmental controllability, making it easier to form standardized products and replicable business models. In contrast, scenarios such as home services have more complex environments and more personalized needs, placing higher demands on robots' operational precision and generalization capabilities. Therefore, physical AI is expected to gradually penetrate and extend along directions such as factories, business services, and home services. Solving high-frequency, rigid-demand, and standardized tasks first, then gradually entering more open and complex scenarios, may be a more realistic commercialization path.
Currently, the industry scale is continuing to expand. According to data recently disclosed by the Ministry of Industry and Information Technology, in the first half of 2026, China's humanoid robot complete-machine products exceeded 400 models, with output exceeding 40,000 units, and full-year output is expected to exceed 100,000 units. Related products are also accelerating validation and deployment in scenarios such as warehousing and logistics, catering and retail, and home services, and related supply and application exploration are speeding up. With the increase in the number of complete-machine products, higher output, and faster scenario validation, humanoid robots are transitioning from the prototype display stage to small-batch application and engineering deployment. At the same time, coordination across the upstream and downstream of the industry chain is also strengthening, with core components, control algorithms, simulation testing, and system integration all iterating around real-scenario needs.
At the same time, policy is also promoting the combination of technology research and development with real scenarios. China's "15th Five-Year Plan" outline proposes encouraging technological innovation such as embodied intelligence and promoting deep integration of artificial intelligence with industry applications, and the supply of computing power, algorithms, and high-quality data has also been included in related arrangements. In June 2026, the Ministry of Industry and Information Technology and the State-owned Assets Supervision and Administration Commission of the State Council jointly launched a special action for real-scenario training of humanoid robots and embodied intelligence, focusing on fields such as industry, special operations, and services, promoting the construction of real training scenarios, tackling practical operation skills, and strengthening application validation and normalized deployment. The action proposes that by the end of 2026, more than 100 high-value application scenarios will be condensed and formed, driving the ability to achieve 10,000-unit-level deployment. Continued policy support in related directions is expected to further accelerate technological breakthroughs and the gradual broadening of application scenarios.
Therefore, the following directions may continue to be worth focusing on going forward. First, attention can be paid to humanoid robots and the robot industry chain, which is a relatively intuitive application direction for physical AI, covering complete machines, key components, and system integration. For specific components, focus can be placed on core component sectors such as reducers, servo systems, motors, ball screws, and actuators, which are expected to continue benefiting from the penetration of physical AI. At the same time, attention can be paid to the intelligent driving chain, focusing on segments such as millimeter-wave radar, LiDAR, steer-by-wire chassis, domain controllers, and intelligent cockpits.
Overall, physical AI is pushing AI from the digital world into the real world, and humanoid robots and intelligent driving are among its representative application carriers. As technological iteration, industry chain improvement, scenario validation, and policy support advance together, physical AI is expected to gradually move toward industrial implementation and continuously expand the boundaries of application in scenarios such as factories, business services, and home services.
Comments