Is memory no longer a blind buy? Morgan Stanley: OEMs > module makers, DRAM > NAND
In its global technology research report released on July 2, Morgan Stanley revised its NAND supply-demand outlook. The firm projects a 60% year-over-year surge in AI-driven NAND demand by 2027, which a 9% supply deficit in the worldwide market. While supply shortages will continue, the market's tightness will become more selective rather than widespread.
Crucially, demand trends are diverging. While server and AI sectors remain strong鈥攕hielded from price drops by long-term agreements (LTAs)鈥攖he consumer segment is hitting a price ceiling. Inventories are climbing for distributors and module manufacturers, and the smartphone and PC sectors are experiencing squeezed margins and lower sales volumes. In fact, actual order reductions have already started following price increases in the second quarter of 2026.
Consequently, investors should re-evaluate their positioning across the memory supply chain. From a tactical standpoint, DRAM outperforms NAND because of stronger long-term agreement (LTA) terms, clearer demand visibility, and strict supply limits tied to EUV capacity constraints. Additionally, potential HBM4E production could crowd out conventional DRAM supply. Within the NAND space, Integrated Device Manufacturers (IDMs) are a safer bet than module makers due to their superior profit resilience and tighter control over supply.
AI demand remains the dominant driver, with NAND shortages extending into 2027
AI-specific NAND demand is forecast to jump from 205 EB in 2025 to 400 EB in 2026, reaching 609 EB by 2027. Consequently, AI's share of the overall NAND market will expand from 18% to 32%, and ultimately hit 41% over those three years.
On a macro level, total global NAND demand is projected to scale from 1,111 EB (2025) to 1,250 EB (2026), and top out at 1,484 EB in 2027. With projected supply at 1,128 EB, 1,058 EB, and 1,347 EB for those respective years, the market will shift from a 2% surplus in 2025 to severe deficits of -15% in 2026 and -9% in 2027.
These projections demonstrate that the industry's supply crunch is far from over. Even with an anticipated 27% year-over-year recovery in production by 2027, the influx of new capacity will be largely consumed by enterprise SSDs, AI servers, QLC storage, and CSP inventory cushions. Nevertheless, this does not mean NAND prices will climb indefinitely. A sharp divide has formed between AI and consumer demand; while shortages remain tightly focused on the server and AI sectors, consumer-grade products are hitting a wall regarding further price hikes.
Diverging price signals: strong in servers, topping out in consumer segment
Supply chain feedback reveals that third-quarter 2026 NAND prices for enterprise TLC SSDs jumped roughly 30% quarter-over-quarter, contrastingly sharply with flat or minimal growth in consumer lines. Concurrently, server DRAM climbed by about 20% QoQ, while legacy DDR3 and DDR4 prices spiked 30% to 40% as shrinking supply collided with mounting AI demand.
Meanwhile, the adoption of long-term agreements (LTAs) is flattening the industry鈥檚 typical price volatility. Current negotiations between memory makers and major clients feature price bands; the established price floors secure vendor profit margins and valuations, while the price ceilings cap any aggressive future price hikes.
Buyer behavior has split by memory type. Customers are willing to accept higher DRAM prices to guarantee supply, but they are resisting NAND hikes due to shrinking margins in consumer electronics.
Inventory metrics also signal caution. While manufacturers maintain historically low stockpiles, inventories have surged among module makers and consumer-grade distributors. According to distributors, underlying demand remains stable, but three consecutive quarters of aggressive price hikes have priced out smaller buyers and slowed down overall transactions, increasing inventory carrying burdens.
The Case for DRAM Superiority over NAND
> Superior Contract Terms: DRAM benefits from stronger long-term agreement (LTA) structures and a higher buyer willingness to pay for guaranteed supply.
> Robust Demand Visibility: AI computing and enterprise server needs provide a highly predictable demand runway.
> Disciplined Supply Expansion: Extreme ultraviolet (EUV) lithography requirements and physical capacity bottlenecks act as a natural barrier to sudden oversupply.
> HBM4E Production Crowding: Shifting manufacturing lines toward high-bandwidth memory (HBM4E) is expected to further restrict standard DRAM output.
Preferring Integrated Device Manufacturers (IDMs) Over Module Makers
Within the NAND ecosystem, IDMs offer significantly more resilient profit profiles and tighter control over physical supply than independent module makers.
> The Flaw in the Traditional Module Maker Model:
Historically, module makers thrived by buying cheap inventory at the cycle's trough and selling it during upswings. However, once this low-cost stock is depleted and margins peak, their stock prices usually face severe pressure.
> Structural Headwinds for Module Makers: Even if a multi-year, AI-driven LTA cycle stabilizes margins more than usual, module makers face distinct limits: their cheap inventory will vanish within the year, consumer price hikes will decelerate in H2 2026, and primary foundries are prioritizing shipments directly to Cloud Service Providers (CSPs).
> Enterprise Market Limitations: Hyperscale clients prefer buying directly from foundries. Because most Asian module makers still rely on consumer SSDs for 80% to 90% of their revenue, they cannot easily pivot to offset consumer electronics weakness.
Three Macro Risks to Monitor
> AI Capex Declines: This is the primary risk for memory valuations. If AI-related capital expenditure continues its upward trajectory, memory profitability could easily stretch past 2027.
> Consumer Sector Resistance: Smartphone and PC manufacturers are pushing back against further price hikes, prompting immediate order cuts and accumulating inventory at the distributor level. This could derail a generalized NAND recovery.
> The 2028 Supply Pivot: If wafer capacity expands prematurely, NAND could face an oversupply crisis by 2028. Conversely, if high-capacity AI inference SSDs (which consume three times the wafer volume of standard SSDs) hit mass production, the supply crunch could intensify instead.
In its global technology research report released on July 2, Morgan Stanley revised its NAND supply-demand outlook. The firm projects a 60% year-over-year surge in AI-driven NAND demand by 2027, which a 9% supply deficit in the worldwide market. While supply shortages will continue, the market's tightness will become more selective rather than widespread.
Crucially, demand trends are diverging. While server and AI sectors remain strong鈥攕hielded from price drops by long-term agreements (LTAs)鈥攖he consumer segment is hitting a price ceiling. Inventories are climbing for distributors and module manufacturers, and the smartphone and PC sectors are experiencing squeezed margins and lower sales volumes. In fact, actual order reductions have already started following price increases in the second quarter of 2026.
Consequently, investors should re-evaluate their positioning across the memory supply chain. From a tactical standpoint, DRAM outperforms NAND because of stronger long-term agreement (LTA) terms, clearer demand visibility, and strict supply limits tied to EUV capacity constraints. Additionally, potential HBM4E production could crowd out conventional DRAM supply. Within the NAND space, Integrated Device Manufacturers (IDMs) are a safer bet than module makers due to their superior profit resilience and tighter control over supply.
AI demand remains the dominant driver, with NAND shortages extending into 2027
AI-specific NAND demand is forecast to jump from 205 EB in 2025 to 400 EB in 2026, reaching 609 EB by 2027. Consequently, AI's share of the overall NAND market will expand from 18% to 32%, and ultimately hit 41% over those three years.
On a macro level, total global NAND demand is projected to scale from 1,111 EB (2025) to 1,250 EB (2026), and top out at 1,484 EB in 2027. With projected supply at 1,128 EB, 1,058 EB, and 1,347 EB for those respective years, the market will shift from a 2% surplus in 2025 to severe deficits of -15% in 2026 and -9% in 2027.
These projections demonstrate that the industry's supply crunch is far from over. Even with an anticipated 27% year-over-year recovery in production by 2027, the influx of new capacity will be largely consumed by enterprise SSDs, AI servers, QLC storage, and CSP inventory cushions. Nevertheless, this does not mean NAND prices will climb indefinitely. A sharp divide has formed between AI and consumer demand; while shortages remain tightly focused on the server and AI sectors, consumer-grade products are hitting a wall regarding further price hikes.
Diverging price signals: strong in servers, topping out in consumer segment
Supply chain feedback reveals that third-quarter 2026 NAND prices for enterprise TLC SSDs jumped roughly 30% quarter-over-quarter, contrastingly sharply with flat or minimal growth in consumer lines. Concurrently, server DRAM climbed by about 20% QoQ, while legacy DDR3 and DDR4 prices spiked 30% to 40% as shrinking supply collided with mounting AI demand.
Meanwhile, the adoption of long-term agreements (LTAs) is flattening the industry鈥檚 typical price volatility. Current negotiations between memory makers and major clients feature price bands; the established price floors secure vendor profit margins and valuations, while the price ceilings cap any aggressive future price hikes.
Buyer behavior has split by memory type. Customers are willing to accept higher DRAM prices to guarantee supply, but they are resisting NAND hikes due to shrinking margins in consumer electronics.
Inventory metrics also signal caution. While manufacturers maintain historically low stockpiles, inventories have surged among module makers and consumer-grade distributors. According to distributors, underlying demand remains stable, but three consecutive quarters of aggressive price hikes have priced out smaller buyers and slowed down overall transactions, increasing inventory carrying burdens.
The Case for DRAM Superiority over NAND
> Superior Contract Terms: DRAM benefits from stronger long-term agreement (LTA) structures and a higher buyer willingness to pay for guaranteed supply.
> Robust Demand Visibility: AI computing and enterprise server needs provide a highly predictable demand runway.
> Disciplined Supply Expansion: Extreme ultraviolet (EUV) lithography requirements and physical capacity bottlenecks act as a natural barrier to sudden oversupply.
> HBM4E Production Crowding: Shifting manufacturing lines toward high-bandwidth memory (HBM4E) is expected to further restrict standard DRAM output.
Preferring Integrated Device Manufacturers (IDMs) Over Module Makers
Within the NAND ecosystem, IDMs offer significantly more resilient profit profiles and tighter control over physical supply than independent module makers.
> The Flaw in the Traditional Module Maker Model:
Historically, module makers thrived by buying cheap inventory at the cycle's trough and selling it during upswings. However, once this low-cost stock is depleted and margins peak, their stock prices usually face severe pressure.
> Structural Headwinds for Module Makers: Even if a multi-year, AI-driven LTA cycle stabilizes margins more than usual, module makers face distinct limits: their cheap inventory will vanish within the year, consumer price hikes will decelerate in H2 2026, and primary foundries are prioritizing shipments directly to Cloud Service Providers (CSPs).
> Enterprise Market Limitations: Hyperscale clients prefer buying directly from foundries. Because most Asian module makers still rely on consumer SSDs for 80% to 90% of their revenue, they cannot easily pivot to offset consumer electronics weakness.
Three Macro Risks to Monitor
> AI Capex Declines: This is the primary risk for memory valuations. If AI-related capital expenditure continues its upward trajectory, memory profitability could easily stretch past 2027.
> Consumer Sector Resistance: Smartphone and PC manufacturers are pushing back against further price hikes, prompting immediate order cuts and accumulating inventory at the distributor level. This could derail a generalized NAND recovery.
> The 2028 Supply Pivot: If wafer capacity expands prematurely, NAND could face an oversupply crisis by 2028. Conversely, if high-capacity AI inference SSDs (which consume three times the wafer volume of standard SSDs) hit mass production, the supply crunch could intensify instead.
Jul 2, 2026 路 7:13 PM UTC
5
23
169
35,445
