SK Hynix bets on EUV to extend 6F² DRAM as CXMT pushes 4F²

This isn't a node war—it's a cost-structure divergence under lithography constraints. Hynix is squeezing the last density out of 6F² via EUV, a defensive play to avoid the yield cliff and capex spike of a 4F² transition while AI server demand devours capacity. ASML's EUV allocation has quietly become a strategic variable: each additional NXE:3800E locked in is an extra card for the 2026-27 1b nm race. CXMT, locked into DUV multi-patterning by export controls, faces 4F² with more mask layers and longer exposure times—likely 15-25% higher wafer cost than Hynix's EUV 6F². But CXMT's calculus isn't technological leadership; it's "good enough" density paired with domestic packaging to undercut on price at the low end. Samsung will likely accelerate full-EUV 1c nm to widen the generational gap; Micron will tilt further toward HBM and retreat from standard DRAM. The real 18-month variable isn't the node number—it's when 3D DRAM hits volume production. If any player ships commercial 3D DRAM by 2026, the planar scaling race becomes irrelevant overnight, and CXMT's 4F² investment risks being bypassed entirely. The moat is shifting from "who has the more advanced lithography tool" to "whose stacking process and TSV yield are more stable."