"Physical AI" entered the lexicon sometime around 2024. It arrived without apology. No one asked whether the term was needed. No one checked whether the engineering disciplines it describes had been practised, taught, and deployed for half a century. The term simply appeared in pitch decks and analyst notes, and suddenly a very old discipline had a very new name.
The costume
The rebrand follows a pattern. When capital needs a new category to invest in, it manufactures vocabulary. Cloud computing was servers in someone else's building. Big data was statistics with more rows. Digital twin was a simulation with a marketing budget. Now "physical AI" is robotics, machine vision, and sensor fusion repackaged for a venture audience that spent three years learning to say "large language model" and needs somewhere new to put the money.
Foxconn has been running vision-guided robotic assembly for years. Amazon has been deploying autonomous warehouse fleets since the Kiva acquisition in 2012. Fanuc, ABB, and KUKA have been building adaptive industrial robots for decades. What changed is not the capability. What changed is that generative AI created a narrative framework capacious enough to absorb robotics into its investment thesis.
The misdirection
The problem with renaming things is that it redirects attention. Call it "physical AI" and the conversation tilts towards the model. The world model. The training data. The compute. All upstream. All things that large technology companies are well placed to provide.
Call it what it is, robotics and automation, and the conversation stays where the value actually lives. On the factory floor. In the integration.
A robot that can navigate around obstacles in simulation is an engineering achievement. A robot that can navigate around obstacles on a production line while maintaining cycle time, respecting safety zones, and coordinating with fourteen other pieces of equipment is a deployment achievement. The distance between the two is not measured in model parameters. It is measured in Manufacturing Engineering hours.
The deployment gap
The FT recently ran a column arguing that "physical AI" could significantly enhance rich-world economies. The diagnosis was correct. The prescription was absent. "Factories need redesigning" appeared once, parenthetically, as if it were a minor logistical footnote rather than the binding constraint on the entire thesis.
The constraint is not the robot. The constraint is the capability to put it to work.
Foxconn and Amazon are not evidence that industrial AI scales. They are evidence that organisations with billion-dollar integration budgets and captive engineering teams can make it work. The relevant question for industrial policy is what happens when a fifty-person precision manufacturer in Nelson, Lancashire tries to do the same thing. The answer, overwhelmingly, is nothing. Not because the technology is unavailable. Because the integration capability does not exist.
The missing discipline
Every wave of manufacturing technology adoption follows the same pattern. The technology arrives. The hype cycle names it. Capital flows to the upstream providers. And then the wave stalls, because nobody funded the people who connect the technology to the process.
CNC machines: needed process engineers to programme them
Additive manufacturing: needed metallurgists who understood powder behaviour and post-processing
Industrial robotics: needed systems integrators who could read both the robot controller manual and the production schedule
"Physical AI" will need exactly the same thing. People who understand both the model architecture and the production physics.
The rebrand does not just obscure the history. It actively harms adoption by making the deployment problem invisible. If the conversation is about "world models" and "embodied intelligence," then the budget follows the vocabulary. It goes to the model builders, the simulation platform providers, the compute infrastructure. It does not go to the people who will actually make the thing work in a factory.
The real question
China's advantage in this space is not strategic intent, though it has that. It is density. A manufacturing base large enough to absorb new technology. An apprenticeship-to-deployment pipeline that still functions. Institutional capability that was never hollowed out because manufacturing was never culturally demoted.
The West's problem is not a shortage of robots. It is a shortage of the institutional memory needed to deploy them. Renaming the field does not fix that. It makes it harder to see.
The greatest technological transformations do not arrive through vocabulary. They arrive through integration. And integration is done by people with process knowledge, domain expertise, and the patience to make a machine work in context. Not in simulation. In production.
Manufacturing Engineers qualify production systems. They validate integration architectures. They design the feedback loops that turn a robot demonstration into a robot deployment.
Kaipability works at this interface, where the technology meets the process and the process meets the business case. If you are deploying automation and the conversation keeps drifting to the model instead of the line, that is the gap we close.