The root cause is straightforward: training and running large language models consumes enormous quantities of memory. Hyperscale data center operators have absorbed so much production that the “big three” suppliers — Samsung, SK Hynix, and Micron, which together control well over 90% of the global DRAM market — are effectively prioritizing AI chips at the expense of everything else .
This structural tilt means conventional DRAM and NAND Flash modules for laptops, desktops, and smartphones are no longer the industry’s focus. Instead, capacity is flowing into server-grade memory and HBM stacks that attach directly to AI accelerators. The consequence is a sustained undersupply of consumer-class memory that analysts now expect to last years, not months .
The magnitude of the price movement is without modern precedent. TrendForce originally forecast a 55% to 60% quarterly increase for Q1 2026, then revised it sharply upward to 90% to 95% in February as supply tightened beyond expectations . By June, the final data showed an even steeper 93% to 98% climb
.
PC DRAM turned in the most extreme moves. Contract prices for PC memory rose over 100% in Q1 2026 alone, setting a record for a single-quarter surge . Server DRAM tracked a roughly 90% increase over the same period
. In the second quarter, TrendForce expects conventional DRAM to rise another 58% to 63%, while NAND Flash accelerates to a 70% to 75% advance following a 55% to 60% increase in Q1
.
Retail spot markets have shown some signs of stabilization by mid-2026 as high prices suppress demand, but contract levels remain at record territory with no visible reversal . A 64 GB DDR5 kit now averages about $900, and it appears buyers are simply refusing to pay more, putting a temporary ceiling on further sharp climbs — though most forecasters do not expect meaningful price relief soon
.
Multiple industry players have now issued explicit public alerts.
Lexar: Chris Xia, Regional Manager for Australia and New Zealand at Lexar, told Tom’s Hardware in early June 2026 that RAM prices are expected to double by the end of the year, and warned that AI infrastructure build-out is consuming nearly all available memory chips from Samsung, SK Hynix, and Micron .
Samsung: During its Q1 2026 earnings call on April 30, memory chief Kim Jaejune said “significant shortages” across memory products will continue through at least 2027, with demand fulfillment rates at record lows. Customers, he said, are already pre-booking capacity for future years .
SK Hynix: Issued comparable warnings. An internal SK Hynix analysis leaked in December 2025 projected that standard DRAM shortages, excluding HBM and specialty modules, would persist until 2028 — more pessimistic than the consensus forecast at the time .
Apple: While Apple has not issued a public statement in the same terms, reports indicate it has faced significant challenges securing supply, with the shortage threatening product launches and pushing laptop prices up by approximately 40% .
On June 3, 2026, a cross-sector coalition of nine U.S. trade associations sent a joint letter to Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick warning of “an urgent imbalance in the market for memory chips.” The signatories represent broadband providers, medical device manufacturers, automobile manufacturers, and national retailers .
The groups argued that AI data center expansion is consuming an “enormous share of available memory chip capacity,” causing chip prices to surge and reducing supply for manufacturers well outside the tech sector. Without government intervention, they warned, the shortage threatens to drive up consumer costs and cause cascading supply chain disruptions across the American economy .
This letter is notable because it extends the memory crisis beyond semiconductors. It is no longer just a problem for PC builders; the automotive, medical, and telecommunications industries are now formally raising the alarm.
The shortage has spread well beyond consumer electronics:
The three dominant manufacturers are investing in additional capacity, but new fabrication plants for conventional memory will not be operational until at least 2027 to 2028 . In the near term, there is no supply-side solution. Samsung and SK Hynix continue to prioritize HBM and server DRAM allocations, leaving standard DRAM supply deeply constrained
.
There is also a potential political dimension. In January 2026, the White House announced a two-phase semiconductor tariff action, and Commerce Secretary Howard Lutnick separately warned unnamed memory chipmakers that they could face 100% tariffs unless they build production in the United States . Any escalation of trade measures could further tighten the market for U.S. buyers.
Meanwhile, U.S. lawmakers have separately called for memory chips to be added to export control lists, specifically targeting high-bandwidth memory used in Nvidia’s Blackwell and H200 platforms . The intersection of commercial shortage, government trade action, and technology export controls makes the supply picture unusually complex.
The consensus from every major forecaster and manufacturer is that there is no near-term fix. Samsung expects shortages through at least 2027. SK Hynix’s internal analysis reaches into 2028. Lexar expects consumer prices to double again by the end of 2026. TrendForce shows the pace of quarterly increases moderating slightly from the extreme Q1 spike, but prices are still rising fast and from an already elevated base .
For individual buyers, the calculus has shifted. Spot prices for some modules stabilized briefly in early 2026 as demand destruction set in, but contract-level pricing — which dictates what PC makers and enterprise buyers pay — continues to climb . The window for near-normal RAM pricing closed in late 2025, and there is broad agreement it will not reopen for at least 18 to 24 months.
Comments
0 comments