BlockBeats reported on July 18 that Citrini analyst Jukan shared a market analysis arguing that the recent pullback in memory-chip stocks was driven not only by deleveraging, but also by investors starting to price in a 2028 supply expansion.
The analysis said that while the industry broadly expects tight supply-demand conditions in high-end memory, including HBM, to persist through 2027, research houses and sell-side analysts generally believe the gap could begin to narrow in 2028 as Samsung Electronics, SK Hynix and other producers push ahead with large-scale capacity expansion.
Market may price in future supply growth earlier than usual
Under the traditional playbook, memory stocks tend to peak about two quarters before memory prices do. This analysis says the market may not be confined to that pattern and could start reflecting expectations for additional supply much earlier.
It added that the current bearish view rests on a single assumption: that newly added capacity will be released in a concentrated wave in 2028 and trigger another sharp drop in memory prices.
Past price collapses were mostly about supply, not weak demand
Looking back over history, the analysis said nearly every major collapse in memory pricing since the 1980s was caused by rapid supply expansion rather than falling demand. Demand kept growing through the PC, smartphone and cloud-computing eras, according to the note.
That is where the current AI cycle may differ. The report argues that demand for compute and memory in AI applications may carry much higher price elasticity than in prior cycles.
Paper cited in support of AI demand elasticity
The analysis cited Zhang and Zhang’s 2026 paper, The Economics of Digital Intelligence Capital, which estimated the price elasticity of demand for AI tokens at about 1.42. In other words, a 1% decline in price could lift demand by roughly 1.42%.
The paper said that when API prices fall sharply, developers do not just increase usage volume. They may also adopt more compute-intensive inference architectures, causing token consumption to rise in a convex pattern.
DRAM example points to a smaller profit decline
Using DRAM as an example, the analysis said a 30% fall in selling prices would usually lead to a steep drop in revenue and profit under a traditional cycle. Samsung Electronics posted a 52.8% year-over-year decline in operating profit in 2019.
In an AI-driven demand environment, however, if unit volume rises about 42% and process upgrades reduce costs by around 15%, industry revenue could remain broadly stable and profit declines could narrow to about 15%.
The analysis said that gap may shape whether memory makers should still be valued like traditional cyclical stocks at 5x to 6x earnings.