Industry executives say nearly all major memory chipmakers are running at or near full capacity, with 2026 production slots almost “sold out” due to AI demand
TAIPEI/TOKYO/SEOUL — Since October, SK Hynix's headquarters in Icheon, near Seoul, has seen a steady stream of visitors … X: @nikkeiasia and @nikkeiasia X: @nikkeiasia : The AI boom is driving huge demand for DRAM and NAND chips. Investment is rising, but supply is tight and prices keep climbing. https://asia.nikkei.com/... [image] @nikkeiasia : A looming shortage of memory chips is casting a shadow over upcoming product launches. https://asia.nikkei.com/... [image]
Context & Ripple Effects
AI demand had already shifted competition toward high-bandwidth memory, where SK Hynix led and Samsung and Micron were trying to narrow the gap in the earlier HBM race. This report extends that pressure beyond a single premium category to DRAM and NAND capacity more broadly.
The significance is not merely higher demand: buyers are competing for finite 2026 output before additional supply is available. Later coverage of shortages expected to persist through 2027 and sharply rising memory contract prices is consistent with a constrained supply response.
First-order effects
- Memory makers can allocate scarce 2026 DRAM and NAND output among customers from a position of limited available supply, while buyers face less flexibility in securing volume.
- AI infrastructure builders and other memory purchasers face tighter procurement conditions and higher memory costs as production slots approach exhaustion.
Second-order effects
- Data-center demand can crowd out other buyers of high-end memory; later reporting said data centers would absorb more than 70% of 2026 high-end output, reinforcing the allocation squeeze.
- Tight supply strengthens incentives for Samsung, SK Hynix and Micron to add capacity, but long lead times mean expansion cannot quickly relieve near-term shortages; subsequent contract-price increases show how rapidly that imbalance can transmit into purchasing costs.
Third-order effects
- Memory is becoming a binding constraint in AI infrastructure economics alongside compute, shifting competitive advantage toward firms that can secure supply and optimize memory use.
- If producers expand aggressively in response, the sector still faces the familiar risk that delayed capacity arrives after demand conditions change—an uncertainty highlighted by investor concerns about eventual oversupply.
The trend: AI infrastructure demand is turning memory from a cyclical component market into a near-term capacity bottleneck, while the durability of that shift depends on how supply expansion catches up.