Nomura analysts say market concerns that a fresh wave of capital spending in South Korea and Meta Platforms Inc.'s (NASDAQ:META) plan to sell off excess compute capacity will precipitate a memory-market downturn are exaggerated.
The brokerage noted that affiliates and Korean memory makers recently disclosed long-range investment programs amounting to about 4.8 quadrillion won, with roughly 3.7 quadrillion won earmarked for projects directly connected to memory production. Those totals have prompted investor worry that future supply could outstrip demand.
But Nomura's assessment paints a different picture. The bank argues that the industry is currently grappling with a significant shortage driven by AI workloads, and that production is being steered toward high-margin high-bandwidth memory - leaving supplies of commodity DRAM and NAND comparatively constrained. In other words, capacity additions tied to high-end AI memory and commodity memory are not interchangeable in the near term, according to Nomura.
Critically, the analysts emphasized that the bulk of the announced investments are unlikely to translate into immediate incremental supply. Semiconductor cluster projects typically involve lengthy development cycles, and Nomura highlighted the Yongin Semiconductor Cluster as an example: although it was launched nine years ago, Nomura estimates it will not reach small-scale production until late 2027. That timeline implies a gap of more than a decade between initial investment and meaningful output, suggesting a multi-year lag before the newly announced projects would affect overall supply.
Nomura also pushed back against the notion that Meta's decision to market spare data-center computing resources is an indicator of cooling demand for AI hardware. The brokerage framed such monetization as a rational step for large platform operators - a way to improve returns on invested capital. Nomura suggested that making excess capacity available could help downstream customers, naming OpenAI and Anthropic as examples of potential beneficiaries.
Far from damping hardware demand, the analysts argued, offering surplus compute to third parties could reduce the cost of compute and thereby stimulate additional usage. In Nomura's view, then, the actions of both Korean memory players and major cloud/AI platforms do not necessarily point to an imminent oversupply; instead, structural lead times and a pivot toward HBM imply continued tightness in key segments.
Summary: Nomura says recent investment announcements and Meta's steps to monetize spare compute are unlikely to cause a memory-cycle downturn, citing long project lead times and persistent AI-driven demand for HBM.