Samsung and SK invest trillions in AI infrastructure and semiconductor facilities - Global Sources

South Korea's capital deployment is not a capacity play—it is a bid to seize pricing power over global AI compute. With SK Hynix already commanding over half of NVIDIA's HBM3E supply, Samsung's follow-on investment reclassifies memory from a cyclical commodity to a strategic asset. Once HBM wafer output locks into a CoWoS advanced-packaging loop, Seoul will replicate the DRAM monopoly playbook of the 2010s. The ripple effects extend well beyond the fab floor. ASML's EUV order book gains another leg; photoresist and specialty-gas suppliers see their bargaining power repriced. Downstream, GPU vendors are forced to accelerate HBM4 interface standardization to avoid single-source lock-in. The hidden risk is regulatory. BIS export-control logic has already migrated from chips to equipment and materials. Any Korean firm retaining mainland-China production lines faces a binary compliance rupture—Samsung's Xi'an NAND line saw global logistics costs spike over 15% after the 2023 tightening. On the competitive front, TSMC will not cede its CoWoS high ground; advanced-packaging capacity is projected to expand another 40% within twelve months. Intel's 18A node, if it ships on schedule, will dismantle Korea's vertical-integration monopoly narrative. Core call: HBM supply-demand flips from scarcity to structural oversupply around Q2 2026. The resulting price war will erode most of this capex cycle's returns. First to build does not mean first to profit.