Morgan Stanley's research report on July 6 indicated that the global memory chip industry is approaching the peak of its rate of change, though the cycle has not yet ended. The report highlights three key controversies: whether computing power is truly in excess, the debate between "maximization" and "minimization" of token consumption, and why long-term agreements (LTAs) have failed to drive a valuation re-rating. From the perspectives of year-over-year price growth, inventory changes, and the breadth of EPS revisions, the industry's upward momentum is waning. Morgan Stanley believes that the largest AI computing buyers are beginning to sell idle computing capacity, and corporate demand for "token minimization" is intensifying. Although the long-term outlook remains bullish (projected 35-40% profit growth by 2027), the short term will face weak pricing and high volatility, as market capital shifts toward stagnant sectors. The earnings guidance from hyperscale cloud providers will have a greater impact on stock prices than statements from memory chip companies themselves.

Article author and source: Zhui Feng Trading Desk

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

According to Zhui Feng Trading Desk, Morgan Stanley Asia-Pacific technology analysts Shawn Kim, Ryan Kim, and others stated in their July 6 report that the three most critical debates in the current memory chip market are: whether the price increase cycle has peaked, why long-term agreements (LTAs) have failed to drive a valuation re-rating, and whether this represents a cycle top or a bull market correction. The report’s core conclusion can be summarized in one sentence: the rate of change in price increases is nearing its peak, but the cycle itself has not yet ended.

Morgan Stanley believes that the upward momentum in the storage sector is waning, as the largest AI computing buyer is reportedly selling idle computing power, and enterprise demand for "token minimization" intensifies.

This means that, ahead of the upcoming earnings season, related stocks are likely to face short-term price weakness and heightened volatility. The market is currently extremely crowded, with capital preparing to rotate into lagging sectors. Morgan Stanley’s bottom-line recommendation is: still bullish in the long term (with earnings expected to grow 35-40% by 2027), but cautious of near-term pullbacks.

Three core controversies: What is the market debating?

Morgan Stanley noted that over the past week, investor discussions have repeatedly centered on three key controversies, which serve as the essential framework for understanding the current performance of the storage sector.

Dispute One: Is Hashrate Really in Excess?

An unverified rumor is circulating in the market that one of the largest capital spenders in the AI sector allegedly has excess computing power available for sale. Bearish interpretations suggest that if hyperscalers have surplus capacity, AI infrastructure may be oversupplied. However, Morgan Stanley offers an alternative view: this simply reflects companies optimizing capital returns and monetizing idle infrastructure—not the same as genuine computing power oversupply.

The true test will come during the Q2 2026 earnings season—if hyperscale cloud providers maintain or raise their capital expenditure guidance, it will be a strong buying opportunity for storage stocks; if they lower it, the oversupply narrative will continue to gain traction.

Dispute Two: The Debate Between Maximization and Minimization of Token Consumption

A new phenomenon has emerged during the implementation of AI applications: many companies previously encouraged employees to use AI-generated tokens as much as possible ("token maxing"), but this led to overspending on IT budgets, prompting companies to seek more cost-effective alternatives.

Specifically manifested as:

• An increasing number of enterprises are adopting open-source large models—among which open-source LLMs from China stand out—to handle basic queries;
• Layer an orchestration layer on top of frontier models, routing simple tasks to open-source models and only invoking frontier models for complex tasks;
• Market focus shifts to how token providers will present this trend in their earnings reports and their guidance for the second half of 2026.

The conclusion of this line is: The second quarter of 2026 (June quarter) poses little issue for AI supply chains, but market concerns have shifted toward the impact of cheaper tokens on guidance for the second half of the year.

Issue three: Why hasn't the stock price been revalued after the LTA was signed?

The signing of a long-term agreement (LTA) was expected to be a catalyst for a revaluation of inventory stock, but the market response has been muted. Morgan Stanley’s explanation is that the market still remembers: past LTAs were either renegotiated or ultimately forced clients to accept inventory they didn’t need (analogous to the experience of simulation semiconductor companies during COVID).

There is also the view that current LTA storage holds structural (rather than cyclical) significance, assuming strong AI demand persists. However, whether earnings expectations can continue to be raised remains investors’ greatest uncertainty—particularly regarding when and to what extent storage prices will continue to exceed expectations, thereby pushing up 2028 EPS, a timeline that remains highly unclear.

Rate of change peaks: Three dimensions peak simultaneously

Morgan Stanley explicitly states that the storage industry is approaching a "peak rate of change," evident in three dimensions:

Year-over-year pricing: DRAM price increases have significantly cooled from their first-quarter peak, and further quarterly gains are expected to continue narrowing;

Inventory Changes: The rate of improvement in inventory cycles is stabilizing;

Earnings Per Share Revision Breadth: The revision breadth for earnings estimates in the DRAM sector has reached a historical high (currently around 89%), leaving limited room for further upward revisions.

This signal of "the rate of change peaking" is the primary reason storage stocks need a阶段性休整.

Notably, since the generative AI surge began in November 2022, the storage sector has experienced three cyclical corrections (corresponding to the Iran-Israel conflict at -15%, profit-taking after a sharp rally at -32%, the so-called “equal tariff day” at -20%, and the current correction of approximately -17%). Morgan Stanley characterizes these corrections as normal adjustments within a structural bull market, rather than the beginning of a bear market.

Meanwhile, Morgan Stanley noted that the most immediate pressure on the storage sector currently comes from positioning, not a fundamental collapse.

Storage stocks are among the sectors with the highest concentration of market positions. Recent increases in volatility have made it increasingly difficult to maintain net exposures at historical highs—a dynamic that has become even more pronounced even amid the backdrop of rising spot prices accompanied by rising volatility. Over the past week, several investors expressed heightened sensitivity to this dynamic and showed strong interest in expanding opportunities in laggard sectors during discussions with Morgan Stanley.

The recent weakness in large-scale cloud provider stocks may be a leading signal that storage stocks (as core beneficiaries of AI spending) are about to underperform the broader market. From a seasonal perspective, this current time window is also a relatively challenging period for the overall market.

Finally, Morgan Stanley explicitly stated that, at this stage, the earnings guidance from hyperscale cloud providers has a greater impact on stock prices than comments from storage companies' own management—because storage companies are likely to maintain an optimistic stance given their position in the cycle.

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