Even as AI computing power demand continues to be unleashed, memory manufacturers are seeing structural divergence.
Morgan Stanley noted in its latest report that with the advancement of KV Cache compression technology, server system optimization, and Long-Term Agreement (LTA) negotiations, the legacy memory industry faces inevitable valuation downward revision pressure.
While market enthusiasm for AI chips remains high, customer affordability regarding high DRAM prices has become a material constraint. In this context, Morgan Stanley has further lowered target prices for DRAM-related manufacturers and reiterated its preference for legacy NAND Flash over DDR4.
<h2>AI Compression Technology Impact: DRAM Demand Logic Shifts</h2>
The market may have overreacted to the impact of model compression technologies like TurboQuant, but the influence of software optimization on hardware demand cannot be ignored. At the Computex exhibition in Taiwan, companies like Skymizer showcased their weight compression technologies, which can significantly speed up inference. Although these technologies do not currently target KV Cache as directly as TurboQuant, their development path is clear.
More importantly, at the CFM Summit in Shenzhen, multiple flash memory suppliers have introduced solutions aimed at reducing system DRAM usage. For instance, Longsys's HLC technology, Phison's AI PC solution combined with iGPU, and Tencent Cloud's DRAM optimization solution all point to the same trend: system manufacturers are reducing their reliance on DRAM capacity through architectural innovation. Considering that the current cost per GB of DRAM is 50 times that of NAND, this optimization incentive is very strong.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/b787d368-73b7-440d-b67d-735bd6bb47e1.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/b787d368-73b7-440d-b67d-735bd6bb47e1.png"/>
<h2>Enterprise Demand Robust, but Price and Capacity Concerns Emerge</h2>
Enterprise-level token demand continues to rise, with edge AI becoming a new growth engine for the memory market. However, alongside the surge in demand, DRAM prices remain high, with current spot prices for DDR4 gradually approaching a cyclical peak. Simultaneously, capacity expansion by major domestic manufacturers, coupled with new industry capacity slated for release in the second half of 2027, adds uncertainty to future price trends.
While long-term agreements can provide some price support for the coming years and solidify stock price bottoms to a certain extent, their lock-in &#34;ceiling&#34; effect may also limit near-term price upside elasticity. Based on this, research institutions expect the average selling price of related products to enter a quarter-on-quarter downward trend starting from the third quarter of 2027 and have accordingly significantly lowered earnings forecasts for 2028.
<h2>Pure NAND Preferred Over DDR4, Focusing on AI New Opportunities</h2>
Morgan Stanley broadly lowered target prices for DRAM-linked legacy memory manufacturers. The report pointed out that valuation multiple compression is the primary adjustment factor, meaning that even if short-term prices do not see a significant decline, the valuation center for relevant targets has already experienced a systematic downward shift.
From a configuration logic perspective, pure NAND manufacturers are relatively less impacted and may even show stronger earnings elasticity during cyclical upswings; meanwhile, companies with AI-related technological capabilities have received overweight ratings due to their positioning in fields such as advanced packaging.
Although DDR4-related manufacturers can still benefit from supply shortages in 2026, entering 2027, with intensifying competition and rising downside risks for prices, their earnings expectations will face greater pressure. This is the core logic behind maintaining an &#34;equal-weight&#34; rating while simultaneously lowering target prices.

FTXL

Morgan Stanley states that while demand for AI computing power continues to be unleashed, customer affordability regarding high DRAM prices has become a material constraint, putting DRAM-related manufacturers under pressure for valuation downward revisions. Technologies such as KV Cache compression and server system optimization are reducing reliance on DRAM capacity, shifting the investment logic from being cycle-driven to cost-constraint-led, with pure NAND manufacturers becoming relative beneficiaries

- Despite the ongoing demand for AI computing power, significant structural differentiation is emerging among memory manufacturers.  
- Morgan Stanley's latest report highlights the pressure on Legacy Memory valuations due to KV Cache compression technologies and customer constraints on DRAM prices, leading to revised target prices.  
- As enterprises drive demand for memory, the combination of high DRAM prices and increasing production capacity by major domestic manufacturers adds uncertainty to future pricing trends.

Wallstreetcn

First Trust Nasdaq Food & Semicon

Direxion Daily MU Bull 2X Shares

VanEck Semiconductor ETF

Micron Tech

iShares Semiconductor ETF

SPDR S&P Semicon

GraniteShares 2x Long MU Daily ETF

Direxion Semicon Bull 3X

Invesco Semiconductors ETF

Samsung Electronics 

XL2CSOPHYNIX

Morgan Stanley: Valuation Downshifts for Legacy Memory Inevitable; Prohibitive Costs Become Core Constraint in AI Era