MU Q3 Deep Dive: AI Demand, HBM Momentum, and Supply Constraints Shape Outlook - The Globe and Mail

Micron's Q3 HBM acceleration is not a demand story—it is a structural realignment of the memory value chain. The bottleneck has migrated from wafer fab to advanced packaging: TSV stacking is pushing from 8 to 16 layers, and CoWoS-class capacity in Taiwan, China has become the true chokepoint. Supply constraints are therefore a systems-integration problem, not a capacity problem. Competitively, SK Hynix holds the NVIDIA anchor position in HBM3E, while Samsung is probing with a cost-attack strategy. Micron's structural edge—dual qualification with both AMD and NVIDIA—provides customer diversification neither rival matches. But that moat narrows at the HBM4 transition, where spec convergence will reignite price competition. On compliance, export controls have evolved from binary sell-or-not to architecture-level restrictions. Micron's mainland China data-center revenue faces sustained compression, and emerging AI infrastructure markets in the Middle East and Southeast Asia cannot yet absorb the gap. Twelve-to-twenty-four-month outlook: HBM enters a structural-premium regime. Inference clusters demand bandwidth far beyond training workloads, driving HBM capacity CAGR above 60%, while packaging expansion cycles run 18+ months. The supply-demand scissors gap means the market is systematically underpricing memory-sector earnings elasticity. Micron's re-rating window is open now.