Nvidia says South Korea has Vera CPU customers, LG robot results ahead

Vera CPU's Korean deployment is not a routine customer win. The strategic signal: Nvidia is extending the Grace-Vera architecture from GPU-accelerated nodes into a full CPU-GPU heterogeneous platform, directly encroaching on AMD EPYC and Intel Xeon territory in data center compute. Korean adoption validates the Samsung HBM supply chain coupled with NVLink-C2C interconnect — a closed-loop verification of the entire compute-memory-interconnect stack, not a single-point breakthrough. The LG humanoid partnership reveals Nvidia's deeper play: Physical AI's endgame is not selling silicon, it is exporting an embodied-intelligence operating system. When inference compute migrates from cloud to robot bodies, Vera's real-time scheduling becomes the critical bottleneck breaker. Within 18 months, South Korea will likely become the second Physical AI industrial closed-loop after the US — chip design, memory, and end-product manufacturing all localized. Competitive windows are compressing. AMD's MI400 must deliver a unified CPU-GPU memory architecture by Q3 2026 or cede pricing power in enterprise AI inference. Tesla Optimus and Figure AI's vertical-integration paths will collide head-on with Nvidia's open-ecosystem model by 2027.