Micron says shortages in HBM, DRAM, and NAND could continue well beyond 2026. Demand for AI hardware is growing far faster than production capacity can expand.
Several technical factors are driving the bottleneck:
>Newer HBM generations use much larger die sizes, which reduces the number of usable chips per wafer.
>Performance gains between memory generations are slowing, so capacity scaling is becoming less efficient.
>Advanced DRAM manufacturing increasingly depends on EUV lithography, which is expensive and slower to ramp at scale.
>HBM production also consumes disproportionate wafer capacity compared with standard DDR memory, squeezing broader DRAM availability.
These pressures are spreading across the tech industry. AI companies and cloud providers keep locking in memory supplies early