At present, Asian suppliers contribute roughly 90% of NVIDIA's production costs, spanning a wide range of sectors including chip foundry services, memory supply, and server assembly. NVIDIA's latest offerings, the Jetson Thor robotics platform and the DRIVE AGX Thor automotive system-on-chips, both built on the Blackwell architecture, are vying for TSMC's 3-nanometer wafer production capacity. Simultaneously, they rely on LPDDR5X memory sourced from Asian manufacturers. Although physical AI products do not necessitate TSMC's CoWoS advanced packaging technology, they still depend on 3-nanometer wafers and memory components from Asia. Owing to a constrained supply of LPDDR4 memory, NVIDIA has opted to prematurely discontinue the production of certain older modules, steering customers towards those utilizing LPDDR5X memory instead. In the meantime, TSMC's CoWoS advanced packaging business is experiencing rapid growth. However, chips manufactured at its Arizona facility still require shipment back to Taiwan, China for packaging. Despite NVIDIA's investments in the U.S. server manufacturing sector, the project has yet to reach large-scale mass production. Consequently, NVIDIA's physical AI product line continues to ramp up its component procurement in Asia, with growth outstripping the capacity of domestic U.S. production.