DRAM prices to double in Q1 as AI demand crowds out PC, mobile markets
DRAM contract prices will surge 90-95% quarter-on-quarter in Q1 2026, according to TrendForce's revised forecast this week. That's up sharply from the firm's early January estimate of 55-60%. NAND flash pricing is now expected to climb 55-60%, versus the earlier 33-38% projection.
The revision reflects two converging pressures. First, US cloud service providers are buying server DRAM and HBM at volumes that are straining fab capacity at Samsung, SK Hynix, and Micron. Second, Q4 2025 PC shipments exceeded expectations, depleting OEM inventories. Dell and HP typically purchase memory a year ahead - which is why prebuilt PCs held pricing while retail memory kits tripled - but that buffer is running out.
PC DRAM will see 100-110% increases this quarter. LPDDR4x and LPDDR5x memory, used in notebooks and smartphones, faces similar pressure with roughly 90% QoQ rises - "the steepest increases in their history," TrendForce noted. Server DRAM is up more than 60%.
The pattern is clear: fabs are prioritizing high-margin AI products over consumer markets. Nvidia's Vera Rubin NVL72 systems contain 54TB of LPDDR5x each. That's capacity not going into laptops.
NAND flash faces added strain as hyperscalers deploy SSDs for AI inference workloads. "The demand for high-performance storage has far surpassed initial expectations," TrendForce wrote, pointing to enterprise SSD order surges from North American CSPs since late 2025.
What this means in practice: System prices will climb as OEMs restock at new contract rates. For enterprise buyers, the Q4 2025 baseline already showed DRAM up 45-50% - Q1's revised numbers compound that. Organizations planning AI infrastructure expansions or hardware refreshes are facing a supply-constrained market where hyperscalers have first call on capacity.
Worth noting: Retail memory prices have stabilized in recent weeks after Q4 volatility, per PCPartPicker data. That reflects buyer resistance, not supply relief. Contract pricing - what enterprises actually pay - tells a different story.
The trade-off is stark. Fabs can sell to consumer markets or they can sell to AI customers paying premium rates. History suggests which they'll choose.