Morgan Stanley: Memory Cycle Not Over, DRAM Preferred Over NAND, OEMs Over Module Makers
Miles Bennett
Morgan Stanley's July 2 report says the memory upcycle is intact, but the trade has shifted from broad-based rallying to structural divergence — DRAM over NAND, original manufacturers over module assemblers, and investors now need to pick sides.
How strong is the AI demand pull?
AI-related NAND demand is projected to rise from 205 EB in 2025 to 609 EB in 2027 — tripling in two years. AI's share of total NAND demand climbs from 18% to 41%.
Global NAND supply sufficiency is forecast at negative 15% in 2026 and negative 9% in 2027. This means → the market stays in shortage, but the tightness eases at the margin — not a blanket squeeze.
In plain terms = AI is the engine behind memory price hikes, but it doesn't pull every car on the train — only the ones closest to it.
Prices are already diverging — who's rising, who's stuck?
DRAM: server-grade products see ~20% quarter-on-quarter price gains in 3Q26. Legacy DDR3 and DDR4 jump 30–40%, driven by supply tightening plus AI demand spillover.
NAND: enterprise SSD-related NAND prices rise ~30% QoQ, but consumer-grade NAND posts only modest gains. This means → within "memory," the pricing power gap between enterprise and consumer has widened dramatically.
Long-term agreements — LTAs, contracts with price floors and ceilings negotiated between OEMs and major buyers — are reshaping pricing dynamics. Customers will pay more for DRAM to lock in supply; NAND price hikes already face pushback.
Why does Morgan Stanley favor DRAM tactically?
Four reasons: more favorable LTA terms, higher demand visibility from AI and server workloads, EUV and capacity constraints that limit rapid expansion and enforce supply discipline, and potential HBM4E — the next generation of high-bandwidth memory — capacity squeeze that could tighten DRAM supply further.
In plain terms = DRAM holds three strong cards — buyers competing to sign contracts, demand that's visible, and supply that can't ramp fast.
Morgan Stanley flags one caveat: if year-on-year price gains plateau around 4Q26 and 2028 supply-demand remains unclear, near-term catalysts may fade — but LTA-driven earnings visibility could still support a valuation re-rating.
Why OEMs over module makers?
In past cycles, module makers — middlemen who buy memory chips from OEMs and assemble them into modules or SSDs — hoarded cheap inventory at the bottom and released it on the upswing for outsized profit swings. This cycle, they face three constraints: low-cost inventory runs out by year-end, consumer price hikes narrow in 2H26, and OEMs are allocating more supply directly to cloud customers.
This means → the module makers' traditional "buy low, sell high" arbitrage is being squeezed by OEM direct-supply strategies.
Hyperscale buyers prefer purchasing directly from NAND OEMs or signing LTAs. Enterprise SSD revenue accounts for only about 10–20% of most Asian SSD module makers' sales — too little to offset weakening consumer demand.
Three risks that can't be ignored
AI capex slowdown is the biggest macro risk — the engine behind memory pricing could itself decelerate.
Consumer weakness: smartphone and PC buyers are pushing back on further price hikes, and order cuts have already appeared.
2028 supply reversal: if new wafer capacity ramps faster than expected and supply discipline loosens, NAND could tip into oversupply. This reflects the memory cycle's perennial problem — price hikes incentivize expansion, and expansion kills the rally.
What is Morgan Stanley's bottom line?
The memory cycle is not over, but the "buy anything" phase is ending.
The next leg is not about whether memory is still rising — it's about distinguishing who holds stronger pricing power, more secure supply access, and clearer demand visibility.
In plain terms = buying memory stocks used to mean "buy the sector." Now it means "pick your seat" — and those sitting on the DRAM OEM side hold the better hand.
Content is for reference only, not financial advice.