全球自动化：人工智能……或早或晚——物理AI，一场数十年游戏的变革者

Global Automation: Artificial Intelligence... Sooner or Later -Physical AI, a game changer in a multi-decade game This report is part of a global series in which our analysts consider what their sector will looklike by the early 2030s when AI has been further integrated and commercialized. There arebroadly two schools of thought. The “Sooner” camp believes that AI is a game-changer fortheir sector, right now. The “Later” camp believes that AI is a game-changer for their sector…just not yet.We are in the “Sooner” camp — Physical AI is already a game changer infactory automation (FA), but it’s a multi-decade “game”. Jay Huang, Ph.D.+852 2123 2631jay.huang@bernsteinsg.com Weibin Liang, Ph.D.+852 2123 2666weibin.liang@bernsteinsg.com Physical AI is already a game changer because today 5-15% of leading FA players’revenue, and up to 50% of their new products, already feature AI functions (Exhibit 3), andit generally strengthens industry leaders’ competitive position. It’s a long game becausePhysical AI adoption will continue to ramp up over the next few decades; its profoundimpact is not a doubling of next-year growth rate, but a 5-10x increase of long-term TAM. Dien Wang, Ph.D.+852 2123 2622dien.wang@bernsteinsg.com Physical AI applications form a well-stacked pipeline, ranging from early research to pilotsand scaled deployment (Exhibit 1). It creates two types of additional opportunities in FA.Notably, it creates new industries - humanoid robotics was a completely unexpectedopportunity even just five years ago, and yet tracking to grow 100x from 2025 to early2030s (Exhibit 6; also see here and here). Then there is the subtler but far more common“Renaissance” type, where AI enables new functions of existing automation products hencepushing the boundary of adoption, such as in industrial robotics (Exhibit 8, Exhibit 12; alsosee here), machine vision, industrial control and simulation. It is how industrial robot densitycan increase about ten-fold in the long run (Exhibit 9), how machine vision can potentiallyreplace the majority of the 35-million quality inspectors worldwide (Exhibit 14, Exhibit 15),and how Cognex can achieve its strategic goal of doubling the customer base in five years. AI in FA is not a single omnipotent product, it is a technology concept to be realized in tensof categories of industrial products and hundreds of thousands of SKUs and use cases. Theprocess is an avalanche in slow motion. Call Physical AI the next “Industrial Revolution”, butrecall that the First Industrial Revolution (mechanization of production) took ~80 years; theSecond (electrification) started in the 1870s with its foundation laid by 1920, but is findinga new life today; the Third (automation and digitization) started in the 1950s and is still in fullswing. Be patient, and enjoy the longevity of the theme. The fact that AI enhances rather than displaces existing automation technologies makesincumbent FA leaders winners rather than losers to the theme. Empirically, Keyence andCognex carry the best AI-enabled vision products and enjoy the most revenue contributionfrom AI in the industry, and FANUC’s market share is significantly higher in AI-enabledindustrial robot applications than in the broader market. We refer to our previous note(here) for the detailed explanation of reasons. It is the “physical” layer that requires deepdomain know-how, protects incumbent leaders, and makes Physical AI distinct from manyother domains. One day, factories may have perfect AI agents advising them how to solvechallenging inspection and measurement problems, and the last word of the AI agent’sanswer will predictably be “call Keyence / Cognex and ask for model number XYZ”. DETAILS Future of Tech: Physical AI -- Bridging industrial and humanoid robotics 24 Mar 2026Humanoid Robotics: The path to 1 million annual shipment 26 Feb 2026Humanoid Robotics: Three steps forward in robotic brain development (Figure AI Helix 02) 4 Feb 2026The Long View: Robotics -- Physical AI and the next phase of industrial Robot Renaissance 22 Jan 2026Humanoid Robotics: Who plays where 12 Jan 2026Keyence and Cognex: One, Two, Three ... Making friends with AI 5 Sep 2025 EXHIBIT 1:Physical AI applications form a well-stacked pipeline, ranging from early research to pilots and scaleddeployment. EXHIBIT 3:Physical AI is already a game changer because today 5-15% of leading FA players’ revenue, and up to50% of their new products, already feature AI functions. EXHIBIT 6:We expect the share of humanoid robot shipments for industrial applications to rise quickly and peakat 50%+ between 2030 and 2035 with an overall market size of 1 - 10 million units p.a., but drop to ~10% in thefuture of 50 - 100 million units p.a. AI enables new functions for industrial robots, thereby pushing the boundaries of adoption. As a result, robot density couldincrease by up to tenfold in the long run (Exhibit 9). With a steady robot shipment growth forecast of 11% CAGR from 2025to 203