Gelsinger calls HBM 'lousy', SK Hynix says it's not memory's final answer

Pat Gelsinger’s criticism of HBM highlights a critical bottleneck in AI chip performance, shifting focus from compute to memory. This critique undermines confidence among memory leaders like SK Hynix, especially in NAND and DRAM markets, where HBM’s high power consumption and limited density hinder AI scalability. Technologically, it accelerates R&D into alternatives like MRAM, ReRAM, and PCM, particularly in data center and edge computing. From a compliance standpoint, stricter export controls on AI chips may disrupt HBM supply chains, especially in Taiwan, China and Hong Kong, China. In response, GPU vendors like NVIDIA and AMD may develop in-house memory controllers to reduce reliance on HBM. Over the next 12–24 months, HBM’s dominance will erode as new architectures gain traction, reshaping the memory landscape.