Preliminary Report

Leadership in physical AI will depend on productive deployment and resilient production capacity, not model capability alone. Like other general-purpose technologies, AI produces broad productivity gains when firms make complementary investments and reorganize production around it; in manufacturing, that means embedding intelligence in machines and putting those systems to productive use by augmenting human workers.

Manufacturers are waiting for robotic systems to become cheaper, more adaptable, easier to integrate, and better proven. At the same time, many improvements in performance and cost depend on the deployment volume that manufacturers provide. The market is trapped in a first-mover problem: each firm has reason to wait for someone else to absorb the uncertainty, while the ecosystem needs enough early deployment to generate production scale, operating evidence, service capacity, and confidence. Flywheels do not start themselves.

For manufacturers, fragmentation appears as a series of disconnected transactions. Technical assistance may come from one program, financing from another, cybersecurity guidance from a third, and workforce support from state or local institutions. Each operates under different timelines, eligibility rules, terminology, and measures of success. Firms may encounter many entry points but no authoritative pathway from problem identification through integration and repeat adoption.

The effects of automation on employment are shaped by three forces that operate simultaneously: job displacement, job transformation (complementary job redesign), and new job creation. The net employment effect in any given firm, occupation, or region depends on the relative magnitude of each, which in turn depends on the pace of adoption, the task structure of the work being automated, and the policy environment surrounding the transition. This is why robotics adoption produces starkly different labor-related outcomes across manufacturing subsectors and geographies rather than a uniform trend in one direction.

While automation has the potential to reinvigorate and expand the manufacturing base, a persistent mismatch in skills currently constrains its deployment. Manufacturers simply do not have the technicians to integrate, operate, and maintain robotic systems.

Robot deployment is not a simple process. Before the process can even begin, manufacturers must identify opportunities where a robot could improve their process. From there, manufacturers must then find the appropriate robot to purchase, and ensure they can get the capital needed to buy it. Both of these steps require a strong understanding of robot capabilities. Once a manufacturer has purchased their robot, they have to integrate it. This can require significant process changes, specialized engineering and software talent, and employee retraining.

A primary obstacle facing non-adopting firms, particularly SMMs, is a lack of knowledge. Many smaller manufacturers remain unsure of what robots can do, where to introduce robotics into their processes, how to select appropriate technologies, or how to navigate integration. Consequently, firms miss opportunities to improve their efficiency.

The demand side offers no offsetting relief. Industrial customers frequently engage in paid pilot programs or technical demonstrations to evaluate new robotic technologies. However, customers rarely commit to multi-year, large-volume purchase contracts without proven field reliability. This creates a deadlock: startups cannot secure financing to scale manufacturing without binding orders, and buyers will not issue binding orders without proof of scalable manufacturing capability.

Simulated data therefore extends real data rather than replacing it. A small set of high-quality real demonstrations can anchor a much larger set of virtual ones, but a simulation with no-real world grounding repeats its own mistakes at scale. Physical testing remains necessary to validate performance and safety under actual operating conditions.

The findings in this report point to a common conclusion. The United States does not lack robotics research, entrepreneurial activity, capable manufacturers, or public programs. It lacks a mechanism for converting those strengths into deployment and production at sufficient scale. Manufacturers hesitate because robotic systems remain costly, difficult to integrate, and insufficiently proven for many applications. Suppliers and integrators hesitate to invest because demand is fragmented. Workers and training providers face uncertain signals about future skill needs, while capital waits for evidence that only wider deployment can generate. Each actor’s decision can be rational in isolation while the system as a whole remains stalled.