How Physical AI Can Solve Reindustrialization in the U.S.

Building new factories or modernizing existing production sites is no mean feat. Despite decades of practice, many manufacturers struggle with it. Across the automotive, industrial equipment, heavy equipment and aerospace industries, more than nine in ten of these complex multi-year capital projects still finish late and cost more than planned. 

These overruns are not typically within an acceptable margin, which is usually factored in from the get-go: A recent Accenture study found that two in three manufacturers overspend on their large greenfield projects by almost 30%, driving additional costs of more than $600 million on average in a challenging financing environment where overspend lingers for years. 

Negative sentiment around these types of projects is growing, as our research indicates. It found that while 30-35% of heavy industry and energy companies discussed these investments in a predominantly negative tone in 2024, that percentage climbed to 50% in 2025.

The bottom line: At a time when rebuilding the manufacturing base in the U.S. is critical, companies are being challenged to do it quickly, at scale and with superior long-term cost/unit economics.

An emerging lever to help manufacturers overcome the clear difficulties in this area are the advances being made in with physical AI.

AI goes real world

Physical AI refers to artificial intelligence systems that integrate real-world physics and spatial dynamics into their models. They enable machines to perceive, understand and interact intelligently with the physical environment. Unlike traditional AI, which operates mainly in digital or abstract domains, physical AI bridges the gap between virtual intelligence and real-world execution. This allows users to simulate processes, machines and environments. 

I am already seeing examples of how this technology helps manufacturers expand and modernize their production capabilities. These examples fall into three buckets described below. 

Facility planning and commissioning

Early adopters’ projects indicate that manufacturers can reduce the commissioning time of new assembly lines before their physical construction by up to 50%. High-fidelity simulations of these lines can detect up to 90% of potential errors before anything is built or installed, preventing costly rework. 

For example, a pharmaceutical company, which recently started work on a new plant in the U.S., uses physical AI for facility planning and virtual commissioning. A digital twin simulates the layout of the facility to identify the most efficient use of space and potential bottlenecks to avoid costly rework. 

German motion technology company Schaeffler, for example, has been exploring how to simulate and identify the best layout for warehouses up-front. Their approach includes virtually positioning production lines, storage racks and kitting stations for dynamic material flow and seamless collaboration between humans, AGVs (automated guided vehicles) and robots.

Automation and robotics integration

Reindustrialization requires highly automated, software-defined factories to produce goods more efficiently to be competitive and, considering manufacturing’s skilled labor shortage, with fewer people. Physical AI-powered simulations allow manufacturers to plan, implement and continuously improve the automation equipment and robotic fleets needed to accomplish these tasks.

Foxconn is already using this technology to transition to a scalable AI-powered robotic workforce. This approach enabled Foxconn to cut deployment times for new robotics systems by 40%. Furthermore, physical AI-driven robotics improved cycle times by 20-30% and reduced error rates by 25%. Virtual validation eliminated costly trial-and-error in physical environments, reducing operational costs by 15%.

In dedicated pilots, Schaeffler is looking to determine the right degree of automation for each facility and application. The pilots cover scenarios with different types and sophistication of physical AI: primarily manual work, AMRs (autonomous mobile robots) supporting transport tasks and adaptive manipulators in highly automated facilities. 

Physical AI will assist manufacturers all the way to the last mile of automation, where general-purpose humanoid robots come into play. A good example of this can be seen at Skild AI, where they’re building a general-purpose robotics foundation model for training legged, wheeled and humanoid robots. 

Production and asset performance

The simulation capabilities physical AI brings allow manufacturers to produce and distribute goods more cost-effectively and safely. 

Applied to predictive maintenance, for example, physical AI can reduce machine downtime and maintenance expenses by at least 10% and, in some cases, by up to 40%. Caterpillar is one among those starting to use the technology to build digital twins of its factories and supply chains, for use in advanced manufacturing capabilities such as predictive maintenance and dynamic scheduling.

Further use cases include energy use and scrap rates, which we were able to reduce by 15-20% in our initial work with clients.

Another example comes from a U.S. consumer goods giant, which has begun using physical AI to analyze worker movement, picking rates and conveyor systems in its warehouses. Here, the simulation identified ways to improve the conveyor, which improved throughput by 20% by optimizing the conveyor flow. The company also saved 15% in capital expenditure by eliminating iterative trial-and-error redesign.

Physical AI can also make factories and warehouses safer for personnel. Belden, a network and data solutions provider, has developed a physical AI-based solution for worker safety in factories and warehouses. The solution puts up a virtual safety fence around robots to create safety zones without disrupting ongoing operations. If a human enters the zone, the robots are automatically stopped or rerouted. 

The reality here is that manufacturers need to get a grip on the spiraling costs of greenfield and brownfield projects, and they need to do so fast. While it is still early days for physical AI, I believe this technology will become an important enabler to manage industrial capital projects more efficiently, with dynamic simulations capabilities to assist manufacturers in tackling many issues they are facing today.

• How AI, Robots and Data Can Transform Manufacturing Amid Economic Uncertainty 
• Siemens Demonstrates Generative AI Copilot to Automate Engineering Tasks 
• Yokogawa Implements Multiple AI Agents at Gas Plant for Autonomous Control