In 2025, the United States installed approximately 38,000 robots in its factories. According to preliminary BLS estimates, manufacturing employment fell by 68,000 between December 2024 and December 2025. American manufacturing production was able to grow without a proportional increase in labor, notably through productivity gains, without this demonstrating a specific effect of reshoring. The issue is both technical and distributive: technological choices affect the distribution of income and employment.
The Essential Points
- Automation coexists with fluctuations in manufacturing employment: approximately 38,000 robots installed in 2025, and according to preliminary BLS estimates, minus 68,000 manufacturing jobs between December 2024 and December 2025; this data alone does not demonstrate a reshoring effect.
- Automation captures productivity gains for the benefit of capital; certain regions hosting factories struggle to rebuild the blue-collar wage base they lost.
- Axelle Arquié’s thesis on AI as a rupture in the capital/labor relationship finds concrete confirmation here: relocated technical progress widens the gap between value produced and value distributed.
- A competing, liberal reading relativizes the issue: efficiency gains could finance new jobs, some outside manufacturing; the history of technological transitions argues for patience.
- The issue by 2030: whether this cycle can be architected differently through taxation, training, or negotiation, or whether automated rent becomes permanently established.
38,500 Robots, 90,000 Jobs: What the Gap Tells Us
The IFR (International Federation of Robotics) data for 2025 is clear. The United States installed approximately 38,000 industrial robots over the year, up 11% year-over-year. At the same time, the Bureau of Labor Statistics records, according to its preliminary data, a decline of approximately 68,000 manufacturing jobs between December 2024 and December 2025.
The two figures coexist without contradicting each other. It is precisely their coexistence that documents something. Factories are returning—automobiles, semiconductors, electrical equipment—but they differ significantly from those that left thirty years ago. Manufacturing productivity can allow for greater output with fewer hours worked, without this necessarily characterizing each relocated factory. An assembly workshop that employed 400 workers in 1995 employs 80 today for comparable, sometimes greater production.
This result fits into a long-term trend. American manufacturing employment lost approximately one-third of its workforce since the 1979 peak, some six million jobs. The deindustrialization of the 1990s-2010s is often attributed to relocations to China or Mexico, which is accurate for part of the flows. According to Acemoglu and Restrepo, automation related to routine tasks explains 50 to 70% of wage changes between groups of American workers observed from 1980 to 2016.
Factory relocations accelerate this mechanism in certain sectors.
Value Created Goes to Capital, Not Labor
Axelle Arquié, an economist who studies the redistributive effects of AI, formulates a thesis that 2025 data puts to direct test. She argues in her publications that AI raises questions about value distribution, because owners of machines, algorithms, and patents benefit widely from productivity gains, without an automatic mechanism for redistribution to labor.
American reshoring clearly illustrates the distributive question. When a battery factory opens in Tennessee or a chip foundry in Arizona, a substantial portion of the value produced returns to shareholders of the investing firm, to suppliers of robotic equipment, and to skilled maintenance jobs. Workers in the region only partially recover the positions lost since the 1990s. They sometimes access service-sector jobs, less stable, less well-paid, outside the scope of the new factory.
This mechanism results from an architectural choice. Companies invest in automation because the costs of installing and operating robots change, because capital is cheap for large firms, and because the framework includes wage and training incentives for certain projects, but no general mechanism for sharing productivity gains. The article published on the shortage of technicians created by robotization documents another facet of the same movement: robots eliminate positions and recompose required qualifications in ways that displaced workers and training systems had not anticipated.
The Liberal Response: The History of Transitions Argues for Patience
A competing reading exists and deserves to be taken seriously. It is advanced by economists like Tyler Cowen or Philippe Aghion, whose work on Schumpeterian growth reminds us of a solid empirical fact: the effects of automation on employment differ according to periods, sectors, and tasks; no rule guarantees that each wave creates more jobs than it destroys. The spinners of 1810 were replaced by mill foremen, then by mechanics, then by textile engineers. The cashiers of the 2000s were replaced by customer service agents, delivery workers, platform managers.
The argument holds over the long century. Over a decade, it is less convincing. The transition is painful for those who live through it, the gains diffuse for those who benefit from them, and the timeline is asymmetrical: job destructions are rapid, creations slow. Arquié shows that automation creates wealth overall. What remains to be determined is who waits, for how long, and with what income during the wait.
The standard liberal response, letting markets redistribute and investing in professional training, runs up against two recent American facts. Geographic mobility of American workers has declined steadily for forty years: workers displaced from Detroit or Pittsburgh do not easily resettle in Arizona to follow new factories. American vocational training remains fragmented, underfunded, and poorly coordinated with the real needs of industries undergoing automation. The compression of the minimum wage toward intermediate professions in France poses an analogous question in another context: when the wage base compresses from the bottom, negotiation over gains in the middle remains without clear answer.
Manufacturing Regions Face the Geography of Robots
Automation of reshoring does not deploy uniformly. New factories settle in Southern and Midwest states that benefit from massive federal incentives, the CHIPS Act and the Inflation Reduction Act, but they choose locations based on access to infrastructure, technical universities, and logistics networks, while also considering workforce availability, training, and retention.
Consequence: regions that lost the most during deindustrialization in the 1990s-2010s often do not benefit from reshoring. The return of production to American soil tends to concentrate in regions equipped with engineering infrastructure and industrial services, and often favors secondary metropolises rather than old working-class cities.
This geographic gap reinforces the thesis of capture by capital. The value produced in a semiconductor factory in Phoenix is distributed among shareholders of the group (often based outside the state), engineers recruited on a national market, and a handful of local maintenance technicians. The host city perceives tax revenues, but struggles to rebuild the density of intermediate jobs that traditional manufacturing produced. It is the same dynamic described in the article on heritage and inheritance: when gains concentrate in narrow, mobile capital, regions poorly endowed with initial capital remain structurally cut off.
The 2027-2032 Cycle According to Choices Made Now
The trajectory is not written. The 2025 data describe a state, not inevitability. Three forces will weigh on the distribution of gains from automated reshoring through the end of the decade.
The first is fiscal. Several proposals circulate in Congress to tax productivity gains linked to automation, a “robot tax” that would redistribute a portion of payroll savings toward reconversion funds or income guarantee schemes. Industrial groups’ resistance is strong, and no serious measure has yet been adopted. But political pressure is growing in manufacturing states, where elected officials are beginning to see that factories return without the working-class votes they hoped to reconstitute.
The second is union-based. The United Auto Workers demonstrated in 2023 that aggressive collective bargaining could wrest significant wage increases even in a sector under pressure from automation. The question is whether this model can extend to new factories, often implanted in states with weak union tradition, sometimes with temporary contracts or subcontractors that fragment the workforce. For now, unionization of new automated factories remains limited in many cases.
The third is technological, and it is the most uncertain. If robotization costs continue to fall at the current pace, a scenario emerges in which even technical jobs of maintenance and supervision, those that had seemed beyond the reach of automation, become themselves automatable. In this case, manufacturing employment could continue to decline even as American production continues to increase. Conversely, industrial AI that augments workers without replacing them, by reducing accidents, optimizing flows, enabling less specialized operators to pilot complex equipment, would open a more distributive scenario. The signals are mixed.
The two logics coexist in current American factories, depending on sectors and firms’ investment choices.
What the data allow us to say with caution: neither the dark scenario (total automation without employment) nor the rosy scenario (history repeats itself and everything ends well) are inevitable. What is certain is that the political and institutional framework will make a difference. The 1930s in the United States demonstrated this: the New Deal created mechanisms for redistributing gains from agricultural and industrial mechanization of the interwar period in a way that the market alone would not have produced. An equivalent for 2025-2035 does not yet exist. Its construction, or its absence, will be the real marker of the cycle to come.
The Great Political Omission of the Decade
Arquié raises a question that American public debate systematically avoids: what social contract governs the relationship between automated capital and displaced labor. Reshoring policies concentrated on industrial sovereignty, bringing back production of chips, batteries, essential medicines. This is a legitimate, documented, and partially achieved objective. They included objectives for good jobs, workforce development, apprenticeship, and prevailing wages, but did not organize a general mechanism for redistributing productivity gains.
This omission is not accidental. It results from asymmetries in mobilization capacities between industrial groups and fragmented workers in new industries. The CHIPS Act and the IRA created aid and national federal incentives that projects in the South and Midwest can benefit from. They included provisions on wages and unions, a real novelty. The prevailing wage clauses of the IRA primarily cover workers and mechanics employed by the taxpayer, contractor, or subcontractor for construction, modification, or repair; CHIPS requirements are distinct.
None of these clauses cover productivity gains generated by automation in the following years.
The test will come with social reports from 2027-2028, when factories financed by these programs will have reached cruising speed. If promised jobs materialize but a significant portion of productivity gains remain captured by capital, American industrial policy will have succeeded on sovereignty and partially failed on distribution. If social provisions hold and expand, a first model of distributive reshoring will be available. Data from 2027-2028 will allow us to decide.
Sources
- TechTimes / IFR World Robotics 2026 Analysis: US Factories Installed More Robots Than They Hired Workers in 2025, IFR Data Shows
- Axelle Arquié, L’IA, révolution industrielle du savoir : entre progrès technique, déstabilisation sociale et défi lancé à l’humain, Cairn: shs.cairn.info/publications-de-axelle-arquie–111394
- International Federation of Robotics (IFR), World Robotics 2026 Report, ifr.org (annual report, no stable page URL)
- US Bureau of Labor Statistics (BLS), Manufacturing Employment Data 2024-2025, bls.gov (monthly sector employment data)
