HBM Exports Surpass 50 Trillion Won Mark - 조선일보

The 50-trillion-KRW HBM export milestone is a structural power shift, not a revenue headline. When SK Hynix and Samsung's TSV stacking capacity gets locked by AI accelerator orders, global compute allocation authority migrates from GPU design to memory packaging. Technical cascade: The HBM3E-to-HBM4 jump to 16-layer stacking makes TSV yield the single point of failure. CoWoS packaging (led by TSMC, Taiwan, China) and HBM wafer capacity alignment will dictate AI chip delivery through 2026. Upstream ABF substrates, underfill materials, and ATE test equipment see synchronized demand expansion. Competitive dynamics: Micron has secured NVIDIA's second-supplier slot, but the HBM4 window is razor-thin. If Samsung misses customer qualification by Q3 2025, its DRAM business commoditizes permanently. ChangXin Memory is probing HBM2 as a beachhead—still a generation behind 12-layer stacking, but a breakthrough would shatter the two-giants-plus-one-challenger structure. Compliance risk: Tightening US export controls on advanced packaging toward China are inflating end-user verification costs. Korean exporters enjoying the revenue tailwind simultaneously face secondary-sanction contagion risk through any CoWoS-linked supply chain nodes in Taiwan, China. 12-24 month trajectory: HBM pricing power shifts from seller's market to structural bifurcation. Training-side growth decelerates; inference-side demand for capacity density opens a second growth curve. 50 trillion KRW is a starting line, not a ceiling.