The most likely answer is: the acute RAM squeeze does not really end in 2026. Prices and allocation pressure may stop accelerating first in late 2026, but broad easing is more likely in 2027, with HBM/AI-server memory remaining tight into 2027 and possibly 2028. For consumer RAM buyers, the pain should begin to moderate earlier than for hyperscale/server buyers, but a full return to pre-crisis pricing is unlikely unless AI capex slows or memory makers overbuild.
Most industry sources now describe this as a structural AI-driven memory upcycle rather than a normal short inventory cycle. TrendForce's 2026 forecast says AI server demand and profit-first supplier strategies are shifting capacity toward high-margin server/HBM applications, lifting DRAM and NAND prices even while consumer demand is weak. Its 2Q26 memory forecast says severe shortages should keep contract prices rising across memory sectors, with cloud providers using long-term agreements to secure supply.
The consensus timeline looks roughly like this:
1. AI has pulled memory capacity toward server/HBM, starving lower-margin products. TrendForce's 2Q26 report says prioritizing server memory heavily restricts consumer supply, cloud giants are securing long-term contracts despite rising costs, and shortages should drive continuous contract price hikes. Its 2026 forecast frames the price rise as structural: AI server dominance plus supplier profit prioritization causes capacity shifts into server DRAM/HBM and away from PC, mobile, and consumer applications.
2. Suppliers have unusually strong pricing power. TrendForce's Feb. 2026 DRAM bulletin says major suppliers entered 2026 with depleted inventory, were shifting capacity to high-margin HBM/server applications, and that the seller's market would persist with faster contract-price growth in the first half. Its Nov. 2025 server-DRAM report said cloud service provider demand and tight supply were driving server DRAM price increases above PC DRAM, while new cleanroom construction requires years.
3. HBM is not just another DRAM product; it consumes scarce advanced wafer, TSV, packaging, testing, and customer-qualification capacity. TrendForce's Dec. 2025 HBM bulletin says server DDR5 shortages and AI demand pushed DDR5 profitability close to HBM, narrowed expected 2026 HBM3e price declines, and gave suppliers enough pricing power to raise quotes. That matters because relief requires not only more wafers, but the right mix of leading-edge DRAM and advanced packaging.
4. Primary company evidence points to supply constraints lasting through 2026. Micron's fiscal Q2 2026 prepared remarks say AI increased memory demand and created structural supply constraints; data-center DRAM and NAND bit TAM were expected to exceed 50% of industry TAM for the first time in calendar 2026; both AI and traditional server demand were constrained by inadequate DRAM and NAND supply; and PC/smartphone units could fall low-double-digits partly because of DRAM and NAND supply constraints. Micron also said HBM4 volume shipments had started in early 2026, HBM4E would ramp in calendar 2027, and data-center NAND demand was significantly above available supply for the foreseeable future.
5. Market reporting is aligned with the analyst/company picture. Reuters reported in late 2025 that AI demand was driving a memory-chip supply crisis; later reporting from CNBC, Tom's Hardware, Data Center Dynamics, South China Morning Post, and others described rising memory prices, record supplier profits, hyperscaler long-term supply negotiations, and Samsung/SK hynix/Micron production increases that still may not immediately clear the shortage.
The main disagreement is not whether 2026 is tight; it is when tightness stops being painful. A moderate view says the consumer RAM market can ease in late 2026 as high prices destroy some PC/mobile demand and suppliers add bits through node transitions. A tighter view says the shortage persists through 2027 because AI server demand, HBM qualification cycles, and hyperscaler long-term agreements keep absorbing incremental supply. The bear case for buyers is an HBM-led shortage into 2028 if accelerator roadmaps, sovereign AI buildouts, and inference growth keep expanding faster than cleanroom and packaging capacity.
There is also uncertainty about the word "end." Spot DDR5 prices could flatten well before contract prices fall. Consumer DIMM availability can improve while server RDIMM, MRDIMM, LPDDR for AI servers, enterprise SSDs, and HBM remain constrained. Conversely, if suppliers over-allocate to server products and PC/mobile demand drops hard, commodity RAM could correct earlier without ending the AI-memory squeeze.
Watch for these indicators before calling an end to the crisis: