Storage no longer bought blindly? Morgan Stanley: Original manufacturers > Module manufacturers, DRAM > NAND

AI is still extending the storage upcycle, but industry trading logic is shifting from “broad price hikes” to “structural differentiation.” Morgan Stanley’s latest core judgment is that storage remains in a favorable cycle, but prefers OEMs to module makers, and DRAM to NAND.

According to Chase Trading Desk, Morgan Stanley updated its NAND supply-demand model in its Global Tech report released on July 2.Their estimates show that AI-related NAND demand will grow by 60% year-on-year in 2027, driving a roughly 9% supply-demand gap in the global NAND market that year. The shortage persists, but the market is no longer experiencing undifferentiated tightness.

The most crucial change lies in the two ends of demand. Server and AI-related demand remains strong, and long-term supply agreements (LTA) provide price downside protection; but the consumer side is showing signals of a price ceiling. Module makers and distributors’ inventories are rising, smartphone and PC clients are struggling between sales volume and profit margins, and actual order cuts have begun after price increases in 2Q26.

This means investors need to reassess the relative positions within the storage chain.DRAM is tactically superior to NAND due to better LTA terms, higher demand visibility, supply discipline limited by EUV, and potential HBM4E capacity squeezing; within NAND, OEMs are preferred over module makers due to stronger profit resilience and supply control.

AI demand remains the main theme, NAND shortage continues until 2027

AI is becoming the core source of incremental NAND demand. AI NAND demand is expected to rise from 205 EB in 2025 to 400 EB in 2026, and further to 609 EB in 2027. AI’s share in total NAND demand will increase from 18% in 2025 to 32% in 2026, and then to 41% in 2027.

On the aggregate level, global NAND demand is projected to rise from 1111 EB in 2025 to 1250 EB in 2026, and reach 1484 EB in 2027. Supply for the same period is expected at 1128 EB, 1058 EB, and 1347 EB respectively. The corresponding supply adequacy rates are 2% in 2025, negative 15% in 2026, and negative 9% in 2027.

These figures show that the industry shortage has not ended quickly. Even if supply recovers to 27% year-on-year growth in 2027, AI servers, enterprise SSDs, QLC storage, and CSP inventory buffer will still be enough to absorb a large amount of new supply.

However, this does not mean NAND pricing can rise indefinitely. AI demand and consumer demand have clearly diverged. Shortages are mainly concentrated in server and AI-related products, while consumer-grade products' ability to bear price hikes is declining.

Price signals diverge: strong server, consumer segment peaking

Channel research shows that 3Q26 TLC enterprise SSD-related NAND prices rose about 30% quarter-on-quarter, while consumer-grade NAND products only saw a small increase. On the DRAM side, server-grade product prices rose about 20% quarter-on-quarter in 3Q26, and traditional DDR3 and DDR4, due to tighter supply and increased AI-related demand, saw increases of 30% to 40%.

LTAs are changing price fluctuation patterns. Memory suppliers and major customers are still negotiating long-term agreements, which typically include price caps and floors. Floors help protect OEM profitability and valuation, while caps limit further price surges.

Customer attitudes have also diverged. Customers are more willing to pay higher prices for DRAM to secure supply, whereas NAND price hikes are meeting resistance. This correlates with rising margin pressure for consumer electronics customers.

Inventory is also sending warning signals. Supplier inventory remains at historical lows, but module maker inventories have increased significantly, and distributor consumer memory inventory is also relatively high. Distributors believe demand hasn't fundamentally deteriorated, but aggressive price hikes over the past three quarters have pushed up costs, deterred small buyers, and shrinking transaction volumes have increased inventory holding pressure.

Why DRAM is preferred over NAND

We maintain a constructive view on the overall storage cycle, but tactically lean toward DRAM.

There are four reasons. First, DRAM’s LTA terms are more favorable, and clients are more willing to pay to secure supply. Second, demand visibility is higher; AI computing and related server demand remain key pillars. Third, supply discipline is clearer: EUV and other technologies, as well as capacity constraints, limit rapid expansion. Fourth, potential HBM4E capacity squeeze may further tighten DRAM supply-demand.

However, storage stocks are still influenced by “rate of change.” If price rises year-on-year plateau around 4Q26 and supply-demand remain unclear in 2028, near-term cycle catalysts may weaken. But profitability visibility brought by LTAs could still support valuation repricing.

Why OEMs are preferred over module makers

Within NAND, preference is toward OEMs over module makers, mainly due to profit resilience and stronger supply control.

In a traditional cycle, module makers typically stockpile low-cost inventory at cycle bottoms, then release inventory in upcycles to achieve higher profit elasticity. However, this model also brings clear cyclicality: once low-cost inventory is exhausted and profits peak, stock prices often come under pressure.

This round is different. If the AI-driven shortage is sustained by LTAs for three to five years, module makers' profit margins may be more stable than in the past. However, module makers still face three constraints: their low-cost inventory will be gradually depleted within the year, consumer price increases will narrow in 2H26, and OEMs are allocating more supply to CSP clients, limiting module makers' shipment growth in 2026 and 2027.

Furthermore, large-scale customers tend to purchase directly from NAND OEMs or sign LTAs, which may restrict module makers’ long-term addressable market within enterprise SSDs and AI storage. Most Asian SSD module makers’ enterprise SSD revenue contribution is still only 10% to 20%, making it hard to offset weakness in consumer SSD.

Three types of risks investors should focus on

First is a slowdown in AI capital expenditure. This is the biggest macro risk facing storage stocks. As long as AI capex does not peak in the near term, storage profitability can likely continue beyond 2027.

Second is consumer segment pressure. Smartphone and PC clients' ability to tolerate continued price hikes is decreasing, order cuts have emerged, and module maker and distributor inventories are rising. If consumer volume continues to be weak, the logic of broad NAND price increases will be diminished.

Third is supply reversal in 2028. If new wafer capacity is released quickly and supply discipline relaxes, NAND may face oversupply risk. Conversely, if new AI inference SSD products move into mass production, they could consume about three times more capacity than regular SSDs, further tightening industry supply.

In conclusion, the storage cycle is not over, but the “blind buy” phase is ending. AI demand continues to support high-end storage and server products, LTAs improve profitability visibility; but consumer prices are peaking, module maker inventories are rising, and supply variables increase in 2028. For investors, the key for the next stage is not to judge whether storage is still in an upcycle, but to distinguish who has stronger pricing power, steadier supply access, and clearer demand visibility.

Risk Disclosure and DisclaimerThe market carries risks; investing requires caution. This article does not constitute personal investment advice, nor does it consider individual users’ specific investment goals, financial situations, or needs. Users should determine whether any opinions, views, or conclusions in this article are suitable for their particular situations. Investments based on this content are at users’ own risk.