Has storage topped out?

Morgan Stanley believes that the global memory chip industry is approaching the "peak rate of change," but this does not signify the end of the cycle.

According to a note from the ZhuiFeng trading desk, Shawn Kim and Ryan Kim, Asia-Pacific tech analysts at Morgan Stanley, stated in a July 6 research report that the three core controversies in the current memory chip market are: whether the price hike cycle has peaked, why long-term agreements (LTAs) have failed to drive valuation revaluation, and whether this is a cyclical top or a mid-bull market correction. The core conclusion of the report can be summed up in one sentence: The rate of change in price hikes is peaking, but the cycle itself is not yet over.

Morgan Stanley believes that with the largest AI computing buyer reportedly starting to sell idle computing power and increasing enterprise demand for "token minimization," the upward momentum in the memory sector is waning.

This means that ahead of the upcoming earnings season, related stocks will face short-term price weakness and extremely high volatility. The market is currently extremely crowded, and capital is preparing to rotate into lagging sectors. Morgan Stanley's bottom-line recommendation is: Long-term outlook remains bullish (with earnings expected to grow 35-40% by 2027), but near-term corrections are to be wary of.

Three Core Controversies: What the Market Is Debating

Morgan Stanley points out that three core controversies have repeatedly emerged in investor conversations over the past week, forming a key framework for understanding the current trajectory of the memory sector.

Controversy 1: Is computing power truly in oversupply?

An unverified rumor is circulating that one of the largest capital spenders in the AI space allegedly has excess computing power available for sale. The bearish interpretation is that if hyperscalers have excess computing power, the entire AI infrastructure buildout could be oversupplied. However, Morgan Stanley offers an alternative interpretation: this is merely companies optimizing capital returns by monetizing idle infrastructure, not equivalent to a genuine oversupply of computing power.

The true verification moment will be the Q2 2026 earnings season — if hyperscalers maintain or raise their capex guidance, it will be a good buying opportunity for memory stocks; if they lower it, the oversupply narrative will continue to fester.

Controversy 2: The battle between "maximizing" and "minimizing" token consumption

A new phenomenon has emerged in AI application deployment: many companies previously encouraged employees to use AI to generate as many tokens as possible ("token maxing"), but this led to IT budget overruns, prompting enterprises to seek cheaper alternatives.

Specific manifestations:

• Enterprises are increasingly adopting open-source large models (with notable performance from open-source LLMs from China) for handling basic queries;

• Adding an "orchestration layer" on top of frontier models, routing simple tasks to open-source models and only calling frontier models for complex tasks;

• Market focus is shifting to: how token providers will present this trend in their earnings reports and guidance for the second half of 2026.

The firm's conclusion is that Q2 2026 (June quarter) is not a major issue for the AI supply chain, but market concerns are shifting to the impact of cheaper tokens on second-half guidance.

Controversy 3: Why haven't stock prices revalued after LTA signings?

The signing of long-term procurement agreements (LTAs) should be a catalyst for valuation revaluation of memory stocks, but the market response has been muted. Morgan Stanley's explanation is: The market remembers — past LTAs were either renegotiated or ultimately forced customers to take inventory they did not need (analogous to the experience of analog semiconductor companies during COVID).

Of course, there is also a view that current memory LTAs are structural (rather than cyclical), provided AI demand remains strong. However, whether earnings expectations can continue to be revised upward remains the biggest uncertainty for investors — especially regarding when and by how much memory prices will continue to exceed expectations, thereby pushing up 2028 EPS, with the timeline being highly unclear.

Peak Rate of Change: Peaking Across Three Dimensions

Morgan Stanley clearly states that the memory industry is approaching the "peak rate of change," reflected in three dimensions:

Year-over-year pricing growth: DRAM year-over-year price increases have significantly fallen from Q1 highs and are expected to continue narrowing in subsequent quarters;

Inventory changes: The improvement in the inventory cycle is leveling off;

EPS revision breadth: The breadth of earnings expectation revisions in the DRAM sector has hit historical highs (currently around 89%), with limited room for further upward revisions.

This 'peak rate of change' signal is the core reason why memory stocks need a phased consolidation.

Notably, since the rise of generative AI in November 2022, the memory sector has experienced three cyclical pullbacks (corresponding to the US-Iran conflict -15%, profit-taking after a sharp rally -32%, the so-called "reciprocal tariff day" -20%, and the current pullback of about -17%). Morgan Stanley characterizes these pullbacks as normal corrections within a structural bull market, not the start of a bear market.

At the same time, Morgan Stanley points out that the most immediate pressure on the memory sector currently comes from positioning levels, rather than a fundamental breakdown.

Memory stocks are among the most heavily concentrated holdings in the market. Recent volatility has made it increasingly difficult to maintain historically high net exposure — even against a backdrop of rising spot prices and increasing volatility, this dynamic has become more pronounced. Over the past week, multiple investors expressed high sensitivity to this dynamic in discussions with Morgan Stanley and showed strong interest in exploring "broadening laggard opportunities."

The recent weakness in hyperscaler stocks could be a leading indicator that memory stocks (as core beneficiaries of AI spending) are about to underperform the broader market. From a seasonal perspective, the current window is also a relatively difficult period for the overall market.

Finally, Morgan Stanley explicitly states that at this stage, the earnings commentary of hyperscalers will have a greater impact on stock prices than the management commentary of memory companies themselves — because memory companies at this point in the cycle are likely to maintain a relatively optimistic tone.

For AI spenders, the "token maximization" effect is expected to support Q2 2026 results, but whether Q3 2026 guidance falls short of market expectations will be the next major point of contention — token usage optimization, competition from low-cost open-source LLMs, and the impact of "chipflation" on margins are all potential downside risks.