Morgan Stanley analyst Shawn Kim identifies memory as the next major AI bottleneck, shifting focus from compute to capacity. Kim explains that AI inference workloads depend on memory access for handling longer text, images, and video, creating a capacity-constrained cycle with long order visibility. The firm believes execution risk, not demand, is the key concern for 2026.

The analysis points to a rapid pricing upcycle for DRAM and NAND, with Morgan Stanley channel checks suggesting potential upside to aggressive quarterly price hikes. Inventories are falling across the supply chain, and text-only AI inference could consume a significant portion of global memory supply next year. This supply crunch is now the central debate for the industry.

Morgan Stanley advises buying into the bottleneck, recommending memory and semicap stocks like Micron, Samsung, SK Hynix, Western Digital, and ASML. Capital expenditure acceleration is considered inevitable, especially in DRAM, with meaningful greenfield expansions expected from 2027. The recommendation is to position for the winners in this constrained environment.