UMC utilization tops 90% in 2H26 as AI demand drives expansion

UMC's 90%+ utilization in 2H26 is not a "mature node recovery" story. The real signal: a single AI server's BOM contains PMICs, SRAM interface chips, RF front-ends, and analog blocks—all sitting in the 28-55nm band. While capital markets fixate on 2nm, mature processes are undergoing a structural repricing that the Street has systematically underpriced. Technical cascade: AI inference clusters demand extreme power efficiency, pushing PMICs toward multi-die modules and pulling BCD, SiGe BiCMOS, and SOI specialty volumes. At 90%+ utilization, UMC gains pricing leverage—but this also seeds the classic expand-then-glut cycle as 2027-2028 capacity (including overseas fabs) comes online simultaneously. Competitive dynamics: TSMC has explicitly deprioritized 28nm expansion, ceding mature-node share to UMC, GlobalFoundries, and Samsung. UMC's pricing window is roughly 18-24 months before mainland China wafer fabs ramp and compress margins industry-wide. Compliance exposure: Mature nodes sit outside the hardest export-control lines, but US scrutiny of "advanced packaging + mature logic" combinations is tightening. UMC's overseas fab yield-ramp costs remain a hidden P&L drag. Verdict: 2H26 marks the profit peak of this cycle. A mature-node price war is highly probable by 2028.