Samsung, NVIDIA Discuss HBM and Foundry Collaboration - 조선일보

NVIDIA CEO Jensen Huang met with Samsung Electronics Vice Chairman Jun Young-hyun in Seoul to discuss collaboration on high-bandwidth memory (HBM) and semiconductor foundry services. The meeting under... Read original →

Samsung and NVIDIA’s strategic alignment is triggering a structural reshaping of the AI chip stack. Tight integration of HBM4E and SOCAMM demands earlier co-design between logic and memory, pressuring TSMC and Micron to accelerate CPO and near-memory computing roadmaps. Geopolitically, U.S. export controls on advanced packaging tools could inflate Samsung’s compliance costs in Korea, while NVIDIA’s reliance on a single DRAM source heightens supply chain fragility—especially as SK Hynix faces capacity constraints. In response, SK Hynix will likely fast-track HBM5 prototyping and deepen collaboration with Taiwan, China foundries on CoWoS-L to counterbalance Samsung. Within 18 months, HBM capability will become the decisive performance differentiator for AI accelerators; those integrating HBM5 with 4nm nodes first will dominate next-gen large model training, while laggards risk exclusion from the high-end AI ecosystem.