Scenario
AI memory demand against a fixed silicon supply: HBM crowds out DRAM and a 2H-2027 capacity gate holds ASPs until a scenario breaks the chain silicon supply: fixed fab output supply lane wafers in HBM lane takes wafer area DRAM lane output squeezed capacity gate ~2H 2027 ASPs inflect 4 Hyperscaler AI capex GPU / accelerator server fleets logic memory stack ~8× memory/unit Pricing power holds until ~2H 2027 Cycle unwinds early supply · capex decel 2 1 7 wafer split NAND + HDD storage AI server / storage systems WFE demand fab orders sustains capex 3 6 5