In its latest research note, investment bank Jefferies issued a stark warning that the upward rally in global memory chip prices is far from over. Dramatic price increases well above market consensus are set to materialize in the second half of 2026, while substantial relief from supply tightness is unlikely to arrive until 2028.
The report, published following an expert conference call between Jefferies and a memory industry consultant, forecasts that DRAM chip prices will jump 40% to 50% quarter-on-quarter in Q3 2026, with an additional 30% to 40% sequential price increase projected for Q4. This outlook is far more aggressive than prevailing market expectations.
Prior consensus estimated a 15%–20% DRAM price rise in Q3 and a gain of less than 30% for Q4. Jefferies’ nearly doubled price projections reflect its grim assessment of the widening supply gap.
Forecasts from another research firm Aletheia Capital earlier this month corroborate the upward price trend, albeit with more modest figures: it expects average DRAM prices to climb 30% in Q3, followed by a further 10%–15% uptick in Q4.
Jefferies identifies explosive demand from the artificial intelligence sector as the primary driver of this price surge. Training and inference for large AI models consume massive volumes of High-Bandwidth Memory (HBM), fundamentally reshaping capacity allocation for DRAM production.
Roughly 50% of global memory production capacity is already locked in by major technology firms via long-term agreements (LTAs), a proportion projected to rise to 70% in the future. Leading manufacturers including Samsung, SK Hynix and Micron are prioritizing high-margin orders from AI server clients, severely squeezing memory supply for consumer electronics such as PCs, laptops and smartphones.
Excluding output from Chinese manufacturers, Jefferies estimates global memory chip supply will only expand 7% to 8% this year, resulting in a monthly wafer shortage of 150,000 to 200,000 units across DRAM and NAND markets.
Jefferies also anticipates prolonged price momentum for memory chips, with average selling prices (ASP) set to register a 40%–45% year-on-year increase throughout 2027.
Market participants have pinned hopes on capacity expansion by Chinese memory vendors as a potential supply remedy, yet Jefferies argues these expansions will deliver limited impact in the near term. The report explicitly states that Chinese DRAM and NAND makers will pose no meaningful competitive threat to global markets from 2026 through 2027, dismissing the notion of “cheap Chinese memory chips” as a misconception. While ChangXin Memory Technologies (CXMT) and Yangtze Memory Technologies (YMTC) are actively scaling production, they lag generations behind leading peers technologically, sell most output domestically, and carry negligible price advantages.
Jefferies projects tangible supply easing will not emerge until 2028. By then, gradual release of new global production capacity (set to lift supply by approximately 15%–20%) paired with potential cooling in AI demand may drive a 15%–20% year-on-year decline in memory chip ASPs.
Notably, Jefferies put forward a far-reaching viewpoint in its report: AI has permanently altered the underlying economics of the memory chip market. Traditionally defined by extreme cyclical boom-bust swings, the memory sector’s cyclicity is being superseded by structural demand growth fueled by AI. AI servers require several times more DRAM than conventional servers, while memory manufacturers have adopted unprecedented capital discipline in capacity expansion, prioritizing profit margins over market share. Accordingly, Jefferies believes DRAM manufacturers deserve higher valuation multiples than historical averages.
Overall, Jefferies’ report paints a picture of persistently elevated memory chip prices over the next two years. For consumers and enterprises alike, pricing pressure for memory modules, SSDs and all electronic devices incorporating memory components will continue trickling down to end markets across 2026 and 2027.
(Reprinted from https://news.eccn.com/）