On January 23, 2026, a blockbuster report caused a stir in the field of industrial automation. The report pointed out that the industrial robot industry is entering the “second renaissance” stage.
The report predicts that the compound annual growth rate of global industrial robot shipments will be as high as 12% in the next ten years.
Global industrial robots are moving from “flexible” to “intelligent”, and the key driver behind it is the thinking “brain” of physical AI.
Imagine that in the automobile factory before 2020, orange-yellow robotic arms performed spot welding, spraying and other operations accurately but coldly. They were unable to cope with unexpected situations, until the emergence of physical AI, that is, embodied intelligence technology, changed everything.
The Shengbo report clearly depicts the three stages of industry development.
From the fixed path stage in the 1980s, robots could only complete a single preset action and the application scenarios were limited; to the flexible path planning stage opened in the 2020s, robots can adjust paths in real time, unlock new applications, and realize the “first renaissance” of the industry; now in the complex task planning stage, robots have begun to have the ability to “think”, ushering in the “second renaissance.”
The report clearly points out that without the “second renaissance” driven by physical AI, the growth of the industrial robot industry will fall back to single digits. With it, it will start a period of rapid growth that will last for more than ten years.
So, what exactly is physics AI?
It is not a new category of “AI robots”, but a complete AI ecosystem built around robots. It has a four-layer core structure, which is like creating a complete “nervous system” for robots.
The bottom layer is the robot body and its digital twin. High-precision servo motors, reducers and embedded control algorithms are the foundation to ensure accurate and reliable movements.
Going up is the multi-modal AI-driven task planning software, which is the robot’s “brain model” that can perceive information and reason about planning.
The next layer is the sensor array, which is the “eyes” and “skin” of the robot. The 3D vision camera and tactile and torque sensors provide data for AI decision-making.
The top layer is the “world model”, which can simulate real physical laws and allow robots to learn by trial and error in the virtual world.
The report also highlights four new “cognitive trends.”
First, AI expands functions rather than subverts hardware, and the division of labor between “brain models” and traditional control algorithms is clear; second, “brain” and “world” belong to different battlefields, and “world models” are good at handling complex scenarios; third, demand for sensors will explode, which is the basis for mission planning and “world model” construction; fourth, leading manufacturers adopt a “dual strategy”, both self-developing core algorithms and seeking external cooperation.
Physical AI brings not only technological changes, but also the reconstruction of the value distribution of the industrial chain.
Investment opportunities are spreading from a single complete machine manufacturer to the entire industry chain.
Among complete machine manufacturers, those with core algorithm self-research capabilities and open cooperation will take the lead; core component manufacturers will enjoy dividends, and high-end suppliers will have a more stable position; sensor and AI software manufacturers are at the forefront, and their growth rate may exceed that of the complete machine market.
However, there is a huge gap from technology to application.
The pace of technology implementation may be slower than expected, and it will take a long time to adapt from the laboratory to the real workshop; the cost issue is critical, high-end sensors and AI software are expensive, and are initially limited to high value-added industries; the manufacturing industry’s own needs are the decisive force, and poor macroeconomics will hinder the popularization of technology; the lack of standardization is a hidden trap, and the inconsistent standards of different manufacturers will hinder the formation of ecosystems and large-scale applications.
The comparative data at the end of the report is meaningful. The automation rate of the welding process in the automotive industry is close to 90%, while the assembly process is less than 30%. This is the gap that physical AI wants to bridge.
When robots can handle complex assemblies as well as skilled workers, the remaining automation space will gradually open up.
Industrial robots are transforming from “steel arms” to “intelligent partners.” The first renaissance in the 2020s will allow robots to “move”, and the second renaissance around 2026 will allow robots to “think.” This factory revolution has quietly begun.
