Citrini Analyst: AI demand elasticity or rewriting the storage cycle—expanding capacity doesn’t necessarily lead to a collapse in corporate profits

BlockBeats News, July 18. Analyst Jukan of Citrini posted that a latest market analysis believes that the recent pullback in storage chip stocks is not only influenced by deleveraging, but the market has also started pricing in expectations of a supply expansion in 2028 ahead of schedule. Although industry consensus generally expects the tight supply-demand situation for high-end storage such as HBM to persist through 2027, research institutions and sell-side analysts generally believe that with massive capacity expansions by manufacturers such as Samsung Electronics and SK Hynix, the supply-demand gap is likely to start easing in 2028. Traditional experience holds that storage stocks usually peak about two quarters ahead of storage prices, but the analysis suggests the market may not be limited to this pattern and could reflect expectations of increased future supply for an even longer period.

However, all bearish expectations are based on the same assumption—that new capacity will be released in a concentrated manner in 2028, leading storage prices to fall sharply again. Looking back at history, since the 1980s, almost every major storage industry price crash has been driven by rapid supply expansion rather than demand deterioration. Demand in the PC, smartphone, and cloud computing eras has actually continued to grow. This AI cycle may differ fundamentally from previous ones: AI applications have higher price elasticity for compute and storage demand. When prices fall, the pace of demand growth may exceed the impact caused by the price decline, weakening the damage of the traditional storage cycle.

The analysis cites the paper “The Economics of Digital Intelligence Capital” by Zhang and Zhang published in 2026, stating that the demand price elasticity for AI Token is about 1.42—meaning that for each 1% decrease in price, demand volume can increase by roughly 1.42%. The paper argues that when API prices are cut significantly, developers will not only increase call volume, but also adopt more compute-intensive inference architectures, leading Token consumption to grow on a convex curve. Taking DRAM as an example: if the selling price drops 30%, under the traditional cycle, revenue and profits typically shrink substantially—Samsung Electronics’ operating profit in 2019 fell 52.8% year over year. But in the AI-driven new demand environment, if sales grow by about 42%, while process upgrades reduce costs by about 15%, industry revenue could remain broadly stable, and the decline in profits could narrow to around 15%. The analysis believes this difference will determine whether storage manufacturers should continue to be valued at traditional cycle-based P/E multiples of 5 to 6.