Physical AI’s real prize: 280,000 old US factories

The physical AI conversation has largely fixated on new factories filled with humanoid robots, but the article points to a different reality: the biggest opportunity may be in the roughly 280,000 U.S. factories already standing. PepsiCo’s work with Siemens’ Digital Twin Composer, built on Nvidia Omniverse, shows how modeling a plant virtually before committing capital can lift throughput by 20% and flag up to 90% of design flaws before construction begins. That translates into estimated capital expenditure reductions of 10% to 15% — capacity extracted from assets PepsiCo already owns, not from new builds.

BMW’s approach reinforces this. Instead of designing a factory around humanoid robots, the automaker put Figure 02 robots onto an existing assembly line in Spartanburg, where they logged about 1,250 operating hours over 10 months in 2025 and supported production of more than 30,000 X3 vehicles. BMW has since moved a newer Figure model into logistics work at the same plant and started a separate humanoid pilot with Hexagon’s AEON robot in Leipzig. These robots plug into a factory already wired for AI, where BMW runs more than 600 AI use cases.

The investment flowing to startups like Augury and Harmoni suggests this retrofit-first model is gaining traction. Augury’s report found that 83% of 501 manufacturing leaders planned to increase AI investment this year, while Harmoni’s $10 million Series A round, led by Bessemer Venture Partners, is aimed squarely at factories still running traditional machinery. The stakes may hinge on whether these retrofits can demonstrate returns beyond early adopters like PepsiCo and BMW, and whether equipment financing and project lending shift accordingly. If AI can reliably unlock latent capacity in existing plants, the factory of the future may already have been built — the opportunity is making it intelligent.