SK hynix’s $45 billion New York chip plan stalls over Washington's China restrictions - Korea JoongAng Daily

The structural collision here isn't about one fab—it's AI memory demand crashing into export-control architecture. SK hynix's HBM3E/HBM4 roadmap underpins every major AI accelerator. A stalled $45B US expansion fractures the HBM supply chain precisely when hyperscaler capex peaks. HBM4 stacking (12-Hi, 16-Hi) slips 18-24 months; NVIDIA's next-gen platforms and custom ASICs hit a hard memory wall. Upstream, ASML EUV and Lam TSV-deposition order books lose $10B+. The real trap is dual-jurisdiction exposure. Dalian fabs running US-origin equipment face forced bifurcation: separate tooling, EDA flows, talent pools. Industry benchmark: COGS +18-25%, time-to-market +12-18 months. Samsung's Taylor ramp and Micron's Idaho expansion are on schedule. If SK hynix loses 12-18 months of US capacity, its HBM share (currently 50%+) compresses to 35-40% by 2027. The window is closing. 12-24 month outlook: expect a pivot to Japan or Malaysia for neutral capacity, targeted HBM-specific export waivers from Washington, and a structural shift where memory—not logic—becomes the decisive chokepoint in US-China tech decoupling. HBM cannot be multi-sourced the way logic can; single-point failure means total chain paralysis.