Citi: AI Continuous Learning to Extend Memory Supply Shortage to 2031

Citi analysts noted that leading memory chip makers are poised to benefit from a structural shift in AI development. The firm expects continuous learning to drive significant growth in memory demand, extending market supply shortages into 2031. Continuous learning, which strengthens models by training on new tasks and knowledge, will generate ongoing demand for model updates and access to historical data, boosting memory usage in products such as HBM, server DDR5, and eSSD. Citi projects HBM bit demand to grow 62% year-over-year to 75.2 billion gigabits in 2027, and 69% year-over-year to 127 billion gigabits in 2028. Global DRAM demand is set to grow 30% and 35% year-over-year in 2027 and 2028, respectively, while supply grows only 19% and 22% over the same periods, resulting in supply-demand ratios of -8.7% and -9.7%. For NAND, demand will grow 29% and 33% year-over-year in 2027 and 2028, outpacing supply growth of 21% and 25%, with supply-demand ratios of -6.1% and -5.5%. Citi's top memory picks include Samsung Electronics, SK Hynix, Micron, Sandisk, and Kioxia, citing logic tied to memory shortages from continuous learning, HBM and server DDR5 demand, DRAM supply shortfalls, and tightening high-density eSSD and NAND supply, respectively.