HBM prices surge as AI demand and HBM4 shift tighten supply - CHOSUNBIZ - Chosunbiz

The HBM market is entering a period of structural price escalation driven by the convergence of insatiable AI accelerator demand and the generational transition from HBM3E to HBM4. TrendForce's projec... Read original →

The HBM4 transition is not a pricing cycle—it is a structural repricing of the memory value chain. Escalating base-die process complexity and expanded die footprint physically compress wafer yield per unit, a constraint no capex wave can linearly offset. The more telling signal sits at the architecture layer: GPU and ASIC designers voluntarily stepping down from 12-high to 8-high stacks. This is not a yield compromise; it is the HBM bottleneck penetrating accelerator design definitions. The 8-high configuration carries a 10-20% per-gigabit premium due to fixed base-die cost amortization, yet packaging yield and system BOM savings make it the rational engineering trade-off. Over the next 12-24 months, the HBM question shifts from "how much to procure" to "how to architect around the constraint." For SK hynix, HBM4 delivers a structural margin tailwind, but 8-high volume ramp dilutes per-unit ASP—total revenue grows while unit-price elasticity is partially hedged. Samsung's 2H26 HBM4 qualification would reshape the supply landscape; Micron's process window keeps narrowing. The deepest risk is zero-sum wafer allocation: advanced-node capacity is contested simultaneously by commodity DRAM and HBM, and this structural tightness persists well into 2027.