The next AI shortage is memory, and one ratio explains it

The next AI shortage is not about the chips that think. It is about the memory that feeds them. Memory has quietly become a structural AI chokepoint on par with the GPUs themselves, and the cost is now flowing downstream into phones, PCs, and ordinary DRAM buyers who have no relationship with AI at all.

The load-bearing fact is one ratio: a gigabyte of high-bandwidth memory consumes roughly three times as much factory space as an ordinary gigabyte of DRAM. The Motley Fool's reporting on the AI memory shortage makes that capacity math legible, and the rest of the story follows. When the three suppliers that make HBM (SK Hynix, Micron, Samsung) retool fabs to feed AI demand, ordinary DRAM and flash get starved. Conventional memory prices jumped 55 to 60% in a single recent quarter, with some DRAM contracts more than doubling year over year. The memory market is on track to reach roughly $200 billion, about a quarter of all semiconductor sales.

The mechanism repeats: capacity math plus multi-year fab lead times plus a captive AI demand floor equals a widening gap, not a closing one. Analysts expect the squeeze through 2028; at least one major warns it could extend past 2030. The everyday proof is already in the share of a smartphone's bill of materials devoted to memory, and in expected unit declines in phones and PCs this year.

Watch the conventional DRAM and NAND price prints, not the AI capex headlines. They are the canary. Anyone selling a "memory ETF" as diversification is selling concentrated supply-chain risk in three names.