NVIDIA: FANUC, Sony, 7 more on physical AI platform

The announcements from NVIDIA and Qualcomm land in a market that sits where AI semiconductors, robotics, edge computing and industrial automation meet. Unlike conventional AI accelerators built mainly for cloud workloads, physical AI accelerators must support real-time perception, sensor fusion, inference, motion planning and control inside machines operating in the physical world. A typical architecture runs from sensors through vision processing, an AI accelerator, a world model, decision-making and motion control — which generates demand for dedicated chips that keep latency low and power draw modest.

Japan's large robotics and manufacturing base gives this technology a ready set of places to be deployed. The country is also building vast AI infrastructure for training the models that eventually run on physical AI systems. NVIDIA and Noetra announced a national AI factory with 13,750 Vera CPUs and 27,500 Rubin GPUs, providing about 140 MW of data center capacity, to support the FRONTia project and multimodal foundation models for AI agents, digital twins, robotics and physical AI. That sets up a two-layer chain: a central AI factory trains foundation models, which are then optimized for power-efficient accelerator chips deployed inside robots, vehicles and industrial machines.

Whether that chain translates into durable business for chip suppliers and robotics developers hinges on where physical AI accelerators are ultimately designed and standardized. Fujitsu, FANUC, Yaskawa and Kawasaki Heavy Industries started exploring a collaborative physical AI control platform in July 2026 for manufacturing, logistics and healthcare uses, and Rapidus's 2026 plans include chiplet packaging design and manufacturing technology for 2nm-generation semiconductors — signs that the competitive ground may extend beyond chip fabrication into packaging, memory and software stacks.