Citi Forecasts Deepening AI Memory-Chip Shortages Through 2031

Citi forecasts that global memory-chip shortages will deepen through 2031 as AI-related demand for high-bandwidth memory, DRAM and NAND outstrips available supply. The bank ties its outlook to a potential shift toward AI systems that keep adding new information without losing access to what they already learned.

Citi calls that approach continual learning: updating models with new tasks and information while retaining prior knowledge. In its view, that would increase demand for memory used while systems operate and for storage that holds learned information.

The projected gaps in DRAM and NAND

For DRAM, Citi projects global demand growth of 30% in 2027 and 35% in 2028. It expects supply to grow 19% and 22% in those years, producing projected supply-demand ratios of negative 8.7% and negative 9.7%, respectively.

-9.7%DRAM supply-demand ratio

Citi projects a negative 9.7% DRAM supply-demand ratio in 2028.

-5.5%NAND supply-demand ratio

Citi projects a negative 5.5% NAND supply-demand ratio in 2028.

69%HBM bit-demand growth

Citi expects HBM bit demand to rise 69% year over year in 2028.

Citi forecasts NAND demand growth of 29% in 2027 and 33% in 2028, against supply growth of 21% and 25%. Its projected gap worsens from negative 0.8% in 2026 to negative 6.1% in 2027, then narrows slightly to negative 5.5% in 2028.

Pressure beyond a single memory type

High-bandwidth memory, or HBM, carries the sharpest growth forecast in the note. Citi projects HBM bit demand will rise 62% year over year to 75.2 billion gigabits in 2027, then another 69% to 127.0 billion gigabits in 2028. It also expects enterprise solid-state-drive demand to expand as AI systems need more room to retain learned information.

A long-range forecast rests on AI adoption

The figures are projections, not shortages already recorded or outcomes guaranteed to occur. Still, Citi’s call is broad: it expects demand growth to exceed supply growth for both DRAM and NAND in 2027 and 2028, while projecting pressure in HBM and enterprise storage as well. Whether continual learning is adopted at the scale built into those assumptions remains the central uncertainty.

Citi identified Samsung Electronics, SK Hynix, Micron, SanDisk and Kioxia as its preferred memory-semiconductor stocks. That is an investment-research view, not evidence that any of those companies will close the capacity gap Citi projects.