Is the sharp decline in storage stocks a 'fake dip'? Morgan Stanley: All the so-called bearish factors are already well-known and priced in; the market has overreacted.

U.S. memory stocks have recently faced selling pressure, but Morgan Stanley believes that the key bearish factors worrying the market were already foreseeable a month ago and do not represent new risks. The supply-demand tightness on the data center side shows no sign of easing. Morgan Stanley maintains its positive outlook on the memory sector and views the current pullback as an attractive buying opportunity.

According to a research report published by Morgan Stanley on July 20, analysts Joseph Moore and his team confirmed after intensive visits last week to data center procurement channels that there is no indication of any weakening in the intensity of the memory shortage. Morgan Stanley estimates that data center memory prices in the third quarter will rise by at least 25% quarter-over-quarter compared to second-quarter equivalents—higher than both its own prior forecast and those of third-party institutions.

Morgan Stanley notes that the core thesis of this memory cycle is that memory is increasingly becoming one of the primary bottlenecks in AI infrastructure development—a structural constraint expected to persist for several years. Against this backdrop, Morgan Stanley believes the risk-reward profile of memory stocks is rapidly catching up with previously favored names like NVIDIA and Broadcom, and the current sell-off has created a compelling entry point.

Bearish factors are not new; the market overreacted

Morgan Stanley explicitly states that the recent concerns weighing on memory stocks—slowing second-order growth momentum, rising capital expenditures, and customer de-speccing—are all 'open secrets' that were predictable a month ago and do not constitute new fundamental changes.

On price momentum, Morgan Stanley acknowledges that the slowdown in second-order growth is an objective reality but emphasizes that it was inevitable. According to SIA data, DRAM prices rose approximately 70% quarter-over-quarter in Q1 and surged over 40% in Q2. Morgan Stanley points out that, given the memory industry’s quarterly revenue has climbed from roughly $46 billion a year ago to over $200 billion, sustaining such high growth rates would be impossible and would inflict destructive demand-side impacts. 'Everyone knew this several weeks before the stock peaked,' Morgan Stanley wrote.

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 hard price constraints. Morgan Stanley also notes that Micron’s comment during its earnings call—that 'Q2 pricing may represent the ceiling for some newly signed agreements'—falls within a conservative framing. Based on industry channel information, these agreements are likely older deals that had already been agreed upon in principle but were delayed due to lengthy legal approval processes, whereas new agreements currently under negotiation are expected to carry higher price ceilings.

Data center shortages continue to intensify, driven primarily by AI demand

Morgan Stanley stresses that this memory cycle differs fundamentally from historical cycles: demand is now almost entirely driven by data centers, while mixed signals from consumer, PC, and smartphone markets are merely 'false signals' and should not be interpreted as indicators of a cyclical turn.

Morgan Stanley notes that cloud customers are paying premiums above expected Q2 prices to secure memory on a six-week forward basis—'Do we really think these customers are paying premiums just to stockpile inventory in warehouses?' Morgan Stanley rhetorically asks. This phenomenon directly confirms that the supply-demand tightness stems not from inventory dynamics but from genuine capacity constraints.

On the demand side, Morgan Stanley notes that AI computing power expenditure is growing at a rate exceeding 50%, significantly outpacing the annual growth rates of 3% to 5% seen in the PC and smartphone markets. As AI’s share of total demand continues to expand, this divergence will become increasingly pronounced. The manufacturing complexity of HBM4 will consume substantial production capacity, and with the launch of the Rubin Ultra platform next year, HBM memory capacity is expected to double. Meanwhile, demand for low-power DDR5 for server racks and enterprise storage remains robust. Regarding NAND, Morgan Stanley observes that industry capital expenditures have remained unusually restrained; although they are expected to rise somewhat next year, the increase will likely be insufficient to meaningfully expand supply.

The duration of the cycle matters more than the magnitude of its peak.

Morgan Stanley believes the market debate should shift from 'how high peak earnings can go' to 'how long elevated earnings can last,' as the latter has far greater implications for valuation support.

Morgan Stanley points out that initiatives such as long-term agreements and customer engineering optimizations have somewhat dampened the amplitude of the cycle while simultaneously extending its duration. 'Several consecutive years of rising profitability from current levels are likely to provide stronger support for high valuations than a single exceptionally strong year,' Morgan Stanley wrote.

Regarding the risk of specification downgrades, Morgan Stanley acknowledges that NVIDIA has already significantly reduced its usage of LPDDR5 memory in server racks and is actively driving a broader reconfiguration of compute, working memory, and storage architectures to alleviate memory constraints. However, Morgan Stanley argues that the underlying rationale for these actions is precisely that 'memory shortages will persist for several years'—a signal, not a bearish indicator. As supply gradually comes online, memory consumption will inevitably expand in tandem.

Editor/KOKO