Samsung: HBM to account for nearly 30% of DRAM production capacity next year, potentially tightening supply of conventional memory - 富途牛牛

Samsung's decision to allocate nearly 30% of its DRAM production capacity to HBM next year signals a decisive strategic pivot toward AI-driven memory demand. This is not merely a product-mix adjustmen... Read original →

Samsung's near-30% DRAM capacity reallocation to HBM is not a product-mix tweak—it is a wafer-level resource restructuring. Under TSV stacking, each HBM stack consumes 8 to 16 dies versus one for DDR5, so the effective die-output contraction far exceeds the headline percentage. Conventional memory faces a structural supply squeeze. The deeper technical shift: HBM4's base-die migration to logic processes elevates advanced packaging from a back-end step to a core manufacturing bottleneck. Samsung's 2026 HBM4 qualification window against NVIDIA and AMD hinges on its foundry capability—memory competition is being pulled into the foundry arena. On competition: SK Hynix's HBM3E lock-in with NVIDIA is today's moat, but generational transition dilutes first-mover advantage sharply. Micron's HBM3E ramp means all three majors flood the market simultaneously by 2026. The 2018 DDR4 price collapse—halved within six months after concentrated capacity release—remains the cautionary template. If HBM repeats that cycle, AI infrastructure ROI models break. Twelve months out: DDR4/DDR5 price floors hold; server and PC BOM pressure becomes visible. Twenty-four months out: the existential risk is synchronized HBM overcapacity. Yield discipline and customer qualification depth become the only durable barriers. Memory is migrating from commodity cycle to foundry-like economics—capital discipline decides who survives the next turn.