The New Factory Town
Physical AI Will Not Bring Back the Old Factory Town. It Could Create Something Better.
The Financial Times recently asked a provocative question: “Can AI robots save US manufacturing?”
It is the right subject, but perhaps the wrong question.
Physical AI will not restore the American factory economy of the 1950s and 1960s. It will not recreate enormous plants employing thousands of people in stable, repetitive jobs. Nor will it reverse fifty years of manufacturing job losses simply by installing more robots.
But it could make something else possible: a new American manufacturing geography built around smaller, more adaptive and more distributed production systems.
That possibility matters enormously for universities, smaller cities and the future of Knowledge Towns.
The central choice is not American workers or robots. Increasingly, it is robot-enabled American production or no American production at all.
From robots that repeat to robots that learn
Industrial robots are not new. Automakers have used robotic arms for welding, painting and assembly for decades. But these machines have generally performed fixed movements in highly controlled environments. They work brilliantly when every component arrives in exactly the right position and the same action is repeated thousands of times.
They are much less useful when products change, production runs are small or the physical environment is unpredictable.
Physical AI promises something different. By combining machine vision, language models, simulation and increasingly capable control systems, robots can begin to perceive their surroundings, respond to variation and learn new tasks.
The immediate opportunity is not necessarily the humanoid robot that can do everything a person can do. That remains a much more difficult technical challenge than the current enthusiasm sometimes suggests.
The nearer-term opportunity is more practical:
Inspection systems that identify defects humans might miss
Robotic cells that can be reconfigured for different products
Autonomous movement of materials within factories
AI-assisted machining and welding
Predictive maintenance
Digital twins that allow factories to simulate production before changing physical equipment
Systems that coordinate people, machines, suppliers and production schedules
These applications could change the economics of producing things in America.
Traditional factory automation was easiest to justify when a company manufactured millions of identical products. AI could extend automation into lower-volume, higher-value and faster-changing production. This is particularly important in aerospace, defense, medical devices, energy systems, advanced materials and specialized industrial equipment.
Hadrian, one of the companies highlighted in the Financial Times article, is pursuing autonomous factories for aerospace and defense manufacturing. Bright Machines is integrating robotics, software and production data into more programmable manufacturing systems. The ARM Institute in Pittsburgh is working on both robotics commercialization and the workforce required to operate these systems.
These companies are early signals of a larger industrial reorganization.
The wrong frame: workers versus robots
The public debate still tends to present automation as a contest between the worker and the machine.
This concern is understandable. Manufacturing once offered millions of Americans relatively secure, well-paid employment without requiring a four-year degree. Increased productivity has not prevented manufacturing employment from falling dramatically as a share of the American workforce.
We should not pretend that every robot creates more jobs than it eliminates. In some facilities, automation will increase output while reducing the number of production workers. The economic gains may accrue to owners and highly skilled employees while wages stagnate for everyone else.
But the worker-versus-robot frame leaves out the most important counterfactual.
For many manufacturers, the alternative to automation is not a fully staffed American factory. It is offshore production, an unfilled order, a strategically dangerous foreign dependency or a factory that is never built.
American manufacturers regularly report difficulty finding machinists, welders, technicians, production engineers and maintenance workers. In aerospace and defense, labor constraints and fragmented supplier systems are already limiting the country’s ability to increase production.
The relevant question is therefore not simply how many people work inside an individual factory after automation.
We should ask:
Does physical AI make domestic production economically possible?
Can it preserve industrial capabilities that would otherwise disappear?
Can smaller companies manufacture without having to build enormous plants?
Can it allow more regions to participate in advanced production?
Who captures the productivity gains?
What institutions will prepare workers for the jobs created around these systems?
Can automation support stronger communities rather than isolated industrial enclaves?
These are questions about economic systems and places, not just machines.
The factory is not the true unit of competition
The biggest weakness in most discussions of physical AI is that they stop at the factory door.
A robot-enabled factory still requires reliable power, high-capacity connectivity, appropriate buildings, technical workers, suppliers, housing, transportation, applied research and capital. It also needs an environment in which production problems can be solved quickly.
China’s manufacturing strength does not come from lower wages or greater robot adoption alone. It comes from the density of its production ecosystems. Suppliers, engineers, tooling companies, factories, logistics providers and skilled workers are located close enough to learn from one another and respond rapidly.
The United States will not reproduce that advantage simply by purchasing more robotic arms.
The effective unit of competition is the regional production ecosystem.
This is where physical AI intersects with the Knowledge Towns thesis.
America has hundreds of smaller cities with engineering traditions, industrial buildings, colleges, hospitals, defense facilities and potentially valuable infrastructure. Many have lost large employers but retain the institutional ingredients from which a new production economy could be assembled.
These places do not need to recreate the giant factory town. They can develop networks of smaller factories, shared facilities, research institutions, training programs, housing and specialized suppliers.
The physical-AI economy could be more geographically distributed than the digital economy because making things still requires land, buildings, energy and access to physical supply chains. Those requirements create an opening for places that were largely bypassed by the software economy.
A new role for the university
Most universities will respond to physical AI by creating a robotics course, adding an engineering concentration or purchasing equipment for a laboratory.
That is too small a response.
A university should not only prepare students to enter the physical-AI economy. It can help organize that economy regionally.
Many colleges already possess:
Land and buildings
Power and network infrastructure
Laboratories and machine shops
Faculty expertise
Relationships with regional employers
Housing and community facilities
Institutional purchasing power
The ability to convene business, government and philanthropy
At the same time, many institutions need new revenue, stronger enrollment propositions and a clearer role in their regions.
Physical AI presents an opportunity to convert the campus from an educational enclave into a permeable production platform.
A college-centered Physical AI Production District could include:
A shared robotics and manufacturing facility
Smaller manufacturers could test automation, produce short runs and train employees without making a large initial capital investment.Flexible production bays
Companies could move from university research to prototype and then into early-stage production without leaving the district.A simulation and digital-twin center
Regional firms could model factories, infrastructure and production processes before committing physical capital.Embedded apprenticeships
Students and incumbent workers would learn inside operating production environments rather than relying primarily on classroom instruction.A physical-AI venture studio
Engineers, operators and entrepreneurs could build companies around specific industrial problems identified by regional employers.Shared testing, inspection and procurement
Smaller firms could gain access to capabilities and purchasing relationships normally available only to large manufacturers.Secure compute and connectivity
Production data, robotics and digital twins will require sophisticated networks, edge computing and, in sensitive industries, strong data sovereignty.Housing and quality of place
Technicians, founders, students, experienced fellows and visiting engineers need somewhere attractive and affordable to live.
This is the Permeable University made tangible. Education, research, production and community life would no longer occupy separate worlds.
Why quality of place still matters
It may seem counterintuitive to discuss housing, walkability and public space in an analysis of industrial robotics. But the physical-AI economy will compete for scarce technical talent.
A highly automated factory may employ fewer people on the production line, but it will depend on technicians, integrators, engineers, software specialists and experienced operators. These workers will have choices about where to live.
A factory built beside a highway interchange, surrounded by parking and disconnected from daily life, may struggle to attract them. A production district connected to a university, housing, restaurants, recreation and a walkable center offers a different proposition.
This is one reason Knowledge Towns may become increasingly important to industrial policy. The quality of the production ecosystem and the quality of the place will become inseparable.
The workforce challenge cannot be an afterthought
The strongest case for physical AI is not that automation will painlessly create more jobs than it removes. We do not know that it will.
The stronger argument is that automation can increase the range of goods that America is capable of producing and create a wider ecosystem of technical, service and entrepreneurial work around production.
But those benefits will not be distributed automatically.
The new jobs may arise in different organizations, require different capabilities and appear years after other jobs disappear. A displaced production worker cannot simply be told to become a robotics engineer.
Workforce development must therefore be part of the initial investment design. Every publicly supported physical-AI facility should have an accompanying plan for apprenticeships, incumbent-worker training and advancement into better-paid technical roles.
Universities and community colleges should also create shorter pathways into these systems. A worker may need a ten-week module in machine operation, a six-month apprenticeship in robotic maintenance or a one-year technical credential, not another four-year degree.
We should also create pathways for experienced workers whose knowledge is essential but often undocumented. A 55-year-old production manager may understand more about how a factory actually works than a newly trained AI engineer. Physical AI systems will need that knowledge.
The goal should be to combine experienced industrial judgment with new technical capability.
Where this model could emerge
The most promising locations may not be the largest technology centers. They may be places with industrial knowledge, educational infrastructure and room to build.
Potential examples include:
Syracuse and the wider Micron corridor
Pittsburgh and the western Pennsylvania robotics ecosystem
Merrimack Valley and its advanced manufacturing base
Worcester and central Massachusetts
Burlington and Vermont’s aerospace and advanced-mobility cluster
Akron and northeast Ohio
Morgantown and the broader West Virginia industrial corridor
Waterville, Brunswick and the Maine Knowledge Corridor
College towns near defense installations and federal laboratories
Former industrial cities containing underused campuses and production buildings
Not every region should pursue the same industries. The point is not to install a generic robotics center in every town. Each place needs to begin with real regional demand and an existing or plausible production specialization.
The first move should be an honest Physical AI Readiness Assessment examining:
Anchor-customer demand
Industrial specializations
Power and connectivity
Buildings and development sites
University and community-college capacity
Skilled-worker availability
Supplier density
Housing and quality of place
Public and private capital
Ability to move from research to repeatable production
From that assessment, a region can identify the smallest viable production platform that creates immediate value and can expand over time.
A different American industrial future
Physical AI will not bring back the old factory town.
That economy was built around giant employers, large pools of production labor and a sharp separation between the factory, the college and the community. It created prosperity, but it also left places dangerously dependent on a small number of companies.
The next model can be more distributed and resilient.
Imagine a compact district where a university research team works with an established manufacturer on a new production process. Students learn in the operating facility. A startup uses shared equipment to make its first products. Experienced workers become trainers and production fellows. Suppliers occupy nearby flexible buildings. Housing, restaurants and public spaces make the district part of the town rather than an isolated industrial park.
The factory remains important, but it is no longer the whole economy.
The real competitive advantage comes from the rate at which the entire place can learn.
That may be the most important implication of physical AI. It is not merely a new generation of machines. It is an opportunity to recombine production, knowledge and place.
The future of American manufacturing will not be won by putting robots into yesterday’s factories.
It will be won by creating regional systems in which robots, people, universities, infrastructure and capital learn together.
