US memory stocks have recently faced a selloff, but Morgan Stanley believes the several headwinds worrying the market were already foreseeable a month ago and do not constitute new risks, with the supply-demand tightness on the data center side showing no signs of easing whatsoever. Morgan Stanley maintains its positive view on the memory sector, characterizing this pullback as an attractive buying window.
According to a research report published by Morgan Stanley on July 20, analysts led by Joseph Moore confirmed after intensive visits with data center procurement channels last week that there are no signs of any weakening in the intensity of the memory shortage. Morgan Stanley estimates that third-quarter data center memory prices have risen at least 25% sequentially compared to similar products in the second quarter, exceeding both its own and third-party institutions' previous forecasts.
Morgan Stanley points out that the core logic of this memory cycle lies in the fact that memory is increasingly becoming one of the primary bottlenecks in AI infrastructure construction, a structural constraint expected to persist for several years. Against this backdrop, Morgan Stanley believes the risk-reward profile for memory stocks is rapidly catching up to the previously more favored NVIDIA and Broadcom, with the current selloff having created a strong entry opportunity.
Headwinds Are Not New, Market Reaction is Overdone
Morgan Stanley clearly states that the major concerns recently weighing on memory stocks—a slowdown in the second derivative of growth, rising capital expenditures, and customer de-specification—are all "known cards" that could have been predicted a month ago and do not represent new fundamental changes.
Regarding price growth rates, Morgan Stanley acknowledges that the slowdown in the second derivative is an objective fact but emphasizes that this was inevitable. According to SIA data, DRAM prices rose approximately 70% sequentially in the first quarter and over 40% in the second quarter. Morgan Stanley notes that with the industry's quarterly revenue having climbed from around $46 billion a year ago to over $200 billion, maintaining the above growth rates is both impossible and would be destructively impactful on demand. "Everyone knew this weeks before the stock prices peaked," Morgan Stanley writes.
Regarding long-term agreements (LTAs), Morgan Stanley believes their significance lies more in confirming the supply-demand tightness revealed by channel checks rather than imposing rigid constraints on pricing. Morgan Stanley also points out that Micron's statement on its earnings call about "Q2 prices potentially being the ceiling for some newly signed agreements" falls within a relatively conservative framing—according to industry channel information, these agreements are most likely older deals that were already agreed in principle but had a lengthy legal approval cycle, while new agreements currently under negotiation will have higher price ceilings.
Data Center Shortage Continues to Worsen, AI Demand is the Core Driver
Morgan Stanley emphasizes that this memory cycle differs fundamentally from historical cycles: the demand side is almost entirely driven by data centers, with mixed signals from the consumer, PC, and smartphone markets being mere "false signals" that should not be taken as evidence of a cyclical turn.
Morgan Stanley states that cloud computing customers are paying premiums above expected Q2 prices for six-week futures to expedite memory acquisition—"Do we think these customers are paying premiums to stockpile in warehouses?" Morgan Stanley retorts. This phenomenon directly confirms that the supply-demand tightness is not inventory-driven but a genuine capacity bottleneck.
On the demand side, Morgan Stanley notes that AI compute spending is growing at over 50%, far exceeding the 3% to 5% annual growth in the PC and smartphone markets, and as AI's share of overall demand continues to expand, this gap will become increasingly pronounced. The manufacturing complexity of HBM4 will consume a significant amount of capacity, while HBM memory capacity will double with the launch of the Rubin Ultra platform next year; meanwhile, demand for rack-based low-power DDR5 and enterprise storage is also robust. Regarding NAND, Morgan Stanley points out that industry capital expenditures continue to remain exceptionally restrained, with an expected increase next year insufficient to substantially expand supply.
Cycle Duration More Important Than Peak Magnitude
Morgan Stanley believes the current market debate should shift from "how high will peak earnings be" to "how long can high earnings last," with the latter being more significant for valuation support.
Morgan Stanley notes that measures such as long-term agreements and customer engineering optimization are, to some extent, compressing the amplitude of the cycle but simultaneously extending its duration. "Several years of earnings climbing from current operating levels are likely to support high valuations more strongly than a single exceptionally strong year," Morgan Stanley writes.
Regarding de-specification risks, Morgan Stanley acknowledges that NVIDIA has implemented significant cuts to LPDDR5 memory usage in racks and is promoting broader compute, working memory, and storage architecture reorganization to optimize memory constraints. However, Morgan Stanley believes the underlying logic of these moves is precisely that "the memory shortage will persist for several years"—this is a signal, not a headwind. As supply gradually releases, memory usage will inevitably expand in tandem.
Comments