Physical AI Pushes Chipmakers Up the Value Chain

At TSMC’s European Symposium in Amsterdam, executives from key customers discussed how AI is reshaping their products and business models. STMicroelectronics CEO Jean-Marc Chéry emphasized the company... Read original →

Physical AI is forcing a paradigm shift in semiconductors—from component-centric to system-aware design. Advanced nodes like 3nm and EUV must now optimize for real-time agent inference in robotics and industrial systems, not just raw throughput, accelerating adoption of chiplets and 3D stacking beyond data centers. STMicroelectronics’ pivot to application enablement reflects a strategic hedge against geopolitical fragmentation and commoditization; by co-developing with automotive and automation clients, it locks in value through vertical integration. Cisco’s warning on infrastructure strain reveals fundamental bottlenecks in latency and thermal management, compelling foundries like TSMC (Taiwan, China) to fast-track co-packaged optics. Over the next 18 months, edge NPUs will fragment into highly customized variants, but compliance costs under the EU AI Act and U.S. export controls will rise sharply. While NVIDIA dominates training, ST and Nordic gain leverage in defining physical AI’s inference-control loop—potentially redrawing the industry’s profit map.