Nomura’s Assessment of AI‑Driven Memory Demand
Nomura analysts issued a note stating that worries about a slowdown in AI‑related memory demand are exaggerated. The commentary follows recent announcements by South Korean memory‑chip manufacturers and their affiliates of long‑term investment programmes amounting to roughly 4.8 quadrillion won (approximately US$3.5 trillion), of which about 3.7 quadrillion won is earmarked specifically for memory‑related projects.
Despite investor concerns that such massive capital outlays could lead to an oversupply, Nomura argues the market is instead confronting a severe shortage driven by robust AI demand. Producers are prioritising high‑margin, high‑bandwidth memory (HBM), while the supply of commodity DRAM and NAND remains constrained, reinforcing the shortage narrative.
The analysts highlight that the newly announced projects are unlikely to affect supply in the near term because semiconductor clusters require lengthy development cycles. For example, the Yongin Semiconductor Cluster, launched nine years ago, is projected to commence small‑scale production only in late 2027, indicating a lead time of more than a decade from the initial investment to meaningful output.
Nomura also addresses Meta’s (NASDAQ:META) decision to market surplus data‑center computing capacity. The brokerage dismisses the notion that this move signals weakening AI hardware demand, describing it as a natural monetisation step that can enhance returns on invested capital. Moreover, by potentially lowering computing costs, the action could stimulate additional usage, benefitting AI service providers such as OpenAI and Anthropic.
In summary, Nomura maintains that AI‑related memory demand will remain strong, the anticipated oversupply from South Korean investments is unlikely to materialise for several years, and Meta’s capacity‑monetisation strategy is expected to support rather than diminish AI hardware consumption.