Industry Analysis
The divergence between Samsung and SK hynix in high-end memory is accelerating a technological stratification in the global AI hardware ecosystem. The transition from HBM3E to HBM4 not only demands advanced EUV multi-patterning but also tightly couples DRAM with advanced packaging and CoWoS capacity, compelling TSMC and Intel to scale interposer supply. Chinese players like CXMT and YMTC are gaining share in commodity segments, yet U.S.-Dutch equipment restrictions cap their 3D NAND stacking below 200 layers and HBM yields under 85%, inflating compliance-driven CapEx by 15–20%. With SK hynix’s HBM dominance eroding, Samsung is poised to leverage NVIDIA’s GB200 NVL72 platform to capture early HBM4 volume. Micron, meanwhile, bets on CXL-based memory pooling to sidestep HBM IP barriers. Over the next 18 months, HBM will act as a litmus test for tech decoupling—if China fails to localize ArF-i immersion lithography by 2027, its AI server supply chain faces either performance compromises or costly rerouting.
This page displays AI-generated summaries and metadata for research purposes. Original content belongs to the respective publishers.