Google May Turn to Samsung for Next-Generation TPU Production - thelec.net

According to U.S. tech media The Information, Google is reportedly considering having Samsung Electronics manufacture certain components for its next-generation AI chip, codenamed 'Icefish.' The input... Read original →

Google’s decision to split its next-gen TPU production—assigning I/O dies to Samsung’s 2nm node while keeping compute cores at TSMC’s 1.4nm—is a tactical response to advanced-node bottlenecks. This accelerates chiplet adoption in AI accelerators, forcing co-evolution of HBM and interconnect standards toward heterogeneous integration. Geopolitically, Samsung’s Korean fabs offer a U.S.-compliant alternative amid tightening export controls, though yield risks persist. Strategically, TSMC retains dominance in core logic, but if NVIDIA or Apple follow suit in diversifying suppliers, its pricing power could erode. Over the next 18 months, foundry reliance will shift from monolithic to dual-sourced models, especially for non-critical dies. Samsung could capture 5–8% more market share in high-performance computing, redefining supply chain resilience beyond Taiwan, China.