Micron CFO: No ‘line of sight’ for when memory shortage will end - finance.yahoo.com

Micron's CFO conceding there is no visible endpoint to the memory shortage is a far more revealing signal than it reads. This is not a rerun of the 2017-18 smartphone-driven cycle. Current HBM demand is rigidly locked by exponential AI inference compute growth, and the binding constraint has shifted from wafer starts to advanced packaging yield. Technically, HBM3E TSV stacking yield remains the chokepoint; CoWoS capacity is contested simultaneously by GPU and ASIC programs. Micron's Idaho DRAM fab will not deliver meaningful output until 2026, leaving 2025 supply elasticity effectively at zero. Strategically, SK Hynix has already secured next-generation platform commitments from NVIDIA, while Samsung is probing Micron's enterprise SSD share through aggressive pricing. Micron's counter is to trade HBM3E qualification speed for market window—but the CFO's hedged language betrays internal uncertainty on ramp timelines. Over the next 12-24 months, memory is transitioning from a cyclical commodity to a strategic compute input. Once hyperscalers embed HBM into long-term agreements resembling GPU procurement frameworks, spot-market pricing logic collapses. The shortage will not end; it will migrate—spilling from HBM into DDR5, then into NAND—each migration reshaping the profit distribution across the stack.