Morgan Stanley: Downgrading the valuation of old memory is inevitable; in the AI era, “unaffordable” has become the core constraint.

```

Although the demand for AI computing power continues to be released, memory manufacturers are experiencing structural differentiation.

Morgan Stanley pointed out in its latest report that with the advancement of KV Cache compression technology, server system optimizations, and long-term agreement (LTA) negotiations, the traditional memory (Legacy Memory) industry is facing unavoidable valuation downgrade pressure.

Although the market’s enthusiasm for AI chips has not subsided, customers’ tolerance for high DRAM prices has become a substantial constraint. Against this backdrop, Morgan Stanley further lowered the target prices of DRAM-related manufacturers and reiterated a preference for traditional NAND Flash over DDR4.

Impact of AI Compression Technology: Changing DRAM Demand Logic

The market may have overreacted to the impact of model compression technologies such as TurboQuant, but the influence of software optimization on hardware demand can no longer be ignored. At the Computex exhibition in Taiwan, companies like Skymizer showcased weight compression technologies that can significantly accelerate inference speeds. Although these technologies have not yet directly applied to KV Cache like TurboQuant, their development path is clear.

More importantly, at the CFM summit in Shenzhen, several flash memory suppliers launched solutions aimed at reducing system DRAM usage. For example, Longsys’s HLC technology, Phison’s AI PC solutions combined with iGPU, and Tencent Cloud’s DRAM optimization all point to the same trend: system manufacturers are using architectural innovations to reduce dependence on DRAM capacity. Considering that the current per GB cost of DRAM is 50 times that of NAND, the motivation for such optimization is very strong.

Strong Enterprise Demand, But Concerns Over Price and Capacity

Enterprise-level Token demand continues to climb, and edge AI is becoming a new growth engine for the memory market. However, alongside surging demand are high DRAM prices, with current DDR4 spot prices approaching a stage high. Meanwhile, expansion by major domestic manufacturers, together with new industry capacities to be gradually released from the second half of 2027, adds uncertainty to future price trends.

Long-term agreements can provide some price support for the next few years and help solidify stock price floors, but their price locking “ceiling” effect may also restrict short-term price elasticity. Based on this, research institutions expect the average selling price of related products to enter a quarter-on-quarter decline channel starting from Q3 2027 and have therefore sharply lowered their profit forecasts for 2028.

Pure NAND Outperforms DDR4, Focus on New AI Opportunities

Morgan Stanley has broadly lowered target prices for traditional memory manufacturers related to DRAM. The report points out that a downgrade in valuation multiples is the main adjustment, which means that even if prices do not fall sharply in the short term, the valuation center of related targets has already systematically moved down.

From a configuration logic perspective, pure NAND manufacturers are less impacted and may even demonstrate stronger profit elasticity during upcycles; companies with AI-related technological capabilities have been given overweight ratings due to their advanced packaging and other layout advantages.

Although DDR4-related manufacturers can still benefit from supply shortages in 2026, after entering 2027, as competition intensifies and price decline risks rise, their profit expectations will face greater pressure. This is the core logic behind maintaining a “neutral weight” rating and simultaneously lowering target prices.

Risk Disclosure and DisclaimerThe market has risks; investment requires caution. This article does not constitute individual investment advice and does not take into account the special investment goals, financial status, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article fit their particular circumstances. Investing accordingly is at one’s own risk. ```