PepsiCo lifts plant throughput 20% with Siemens digital twins

The physical AI opportunity is often framed as building new, highly automated factories. The deployments highlighted here point in a different direction: layering AI, sensors, and robots onto production systems that already exist. PepsiCo's work with Siemens' Digital Twin Composer, built on Nvidia Omniverse, let the company model a plant virtually before committing capital to physical changes. At a U.S. Gatorade plant, that approach lifted throughput 20% in three months and flagged up to 90% of design flaws before construction began.

BMW's test of Figure 02 robots at its Spartanburg, South Carolina plant follows the same logic. The robots spent about 1,250 operating hours over 10 months in 2025 retrieving and positioning sheet metal parts for welding, supporting 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 using Hexagon's AEON robot at its Leipzig plant in Germany. The automaker already runs more than 600 AI use cases across its business, including factory systems that inspect welds and flag defects in real time.

Startups are betting that retrofitting, not rebuilding, is where the larger market lies. Harmoni raised $10 million in Series A funding led by Bessemer Venture Partners on Sept. 9, targeting roughly 280,000 U.S. factories that still run traditional machinery and labor-intensive processes. The outcome hinges on whether that installed base adopts retrofit technology at scale, and whether the returns seen at PepsiCo and BMW can be reproduced across older plants. If they can, the economics of physical AI may be measured less by robot sales and more by how much additional capacity existing factories can unlock.