
Cloud Capex Isn't Slowing Down, But the Timeline Has Changed | Analysis of Morgan Stanley's Latest Research Report
On Nov 15, Morgan Stanley downgraded a batch of ratings. The targets were cloud-side semiconductors and legacy memory. The report openly admitted at the outset that these downgrades provoked investor backlash.
Its response was straightforward. Demand remains strong. But if you wait until the cloud capex slowdown becomes visible before acting, the price adjustment will already be complete.
This is a Q&A-style report addressing the four most concentrated investor concerns. The sector view remains "Attractive." Bullish on the industry, but reducing positions in individual names first—that’s the overall stance.
The report is authored by four analysts.
Daniel Yen, Charlie Chan.
Daisy Dai, Ethan Jia.
What Are Investors Actually Asking?
Four questions, each sharper than the last.
Question one: Is the cut in Montage Electronics' earnings forecast justified? Question two: What is the basis for multiple compression, and where are the catalysts?
Question three: Have shipment forecasts for servers and CPUs been revised, and why act now when demand is so strong? Question four: Why retain CXMT and Nanya Technology?
The report answers each point systematically. Taken together, the four answers form the core thesis: the demand outlook is correct, but the timing is wrong.
A Downgrade Before the Exit
Two prior reports existed. One addressed the peak in cloud-side semiconductor growth. The other discussed the fading pricing power in legacy storage. This piece serves as the follow-up clarification.
The downgraded companies fall into two categories: those heavily exposed to the cloud side, and those in legacy storage. The former's revenue tracks server deployments, while the latter follows storage prices.
Within the cloud-exposed group, Aspeed provides server management chips, and Montage handles memory interface chips. The report adopted a more conservative assumption for RCD shipments in 2027.
An RCD (Registered Clock Driver) sits on the memory module. It represents the "content" within a single server, i.e., value per unit. The report states this opportunity remains intact.
In the legacy storage category, Phison received a "Underweight" rating. Winbond and Macronix were adjusted in the same batch. The changes were concentrated on Nov 10 and 11.
The report did not frame the downgrade as bearish toward the sector. It repeatedly emphasized that demand remains robust, supported by Q3 earnings confirming current spending. What has changed is the assessment of timing.
To summarize its stance in one sentence: By the time the slowdown is confirmed by data, the valuation adjustment will have already occurred. Therefore, reduce positions ahead of visibility, not after.
Demand Unchanged, Shipping Timing Shifted
The two hardest figures on the page both relate to timing.
The first is 5GW. The U.S. team shifted approximately 5GW of new capacity installations from 2028 to 2029. Per the report, this equals about 20% of the originally expected 2028 additions.
The second is the capex forecast. Forecasts for hyperscaler spending remain unchanged: $1.6 trillion in 2027 and $1.8 trillion in 2028.
Together, the message is clear. Installation intent remains, but the timing for equipment procurement and deployment is harder to pinpoint. Orders will arrive late, but they won't disappear.
The report attributes the cuts to three uncertainties: procurement timing, platform switching, and supply chain execution. It simultaneously acknowledges that the 2027 RCD shipment assumption may be overly conservative.
This point deserves highlighting. An overly conservative assumption leaves upside room. If customer orders and deployments proceed more smoothly than currently modeled, this upside could materialize.
The Q3 earnings season has just wrapped up. The report suggests this timing warrants incorporating slightly more risk, citing post-midterm election execution uncertainty.
It also highlights a distinction: capex intent and capex realization are two different things. The former depends on guidance, while the latter hinges on equipment and deployment.
Pricing Power in Legacy Storage is Weakening
The issue with legacy storage lies not in demand, but in whether pricing and returns can be sustained.
During upcycles, long-term agreements at fixed prices allow vendors to raise rates quickly. Once demand weakens, the protection these contracts offer diminishes.
The report explicitly states that as shortages narrow, the sustainability of pricing and returns must be reexamined. Its updated supply-demand model points to normalization by 2028.
The conclusion impacts valuation multiples. Lower multiples should be applied ahead of normalization, rather than waiting for peak earnings to roll over.
Another observation supports this view. The breadth of earnings upgrades is peaking. Visibility into 2028 server production is fading.
Combined, these factors undermine confidence in extrapolating growth rates. This explains why the firm adjusts valuations before adjusting earnings.
Two Names Retained, Differing Logic
The report maintains "Overweight" ratings on CXMT and Nanya Technology. The reasons for retaining each differ.
CXMT benefits from three additional drivers: capacity expansion, localization, and process migration. Management noted customers have already booked 2027 capacity for related enterprise products.
DDR5 supply remains tight. This serves as the second rationale for retaining CXMT.
Nanya Technology relies on process and DDR5 transition. DDR5 already accounts for roughly 10% of its revenue. This trajectory differentiates it from other DDR4 suppliers.
The report makes a structural call on niche memory: a smaller supply base means prices are more sensitive to minor supply-demand shifts, and direct exposure to AI demand is lower.
This dynamic works both ways. Elasticity comes from here, and so does fragility.
Both still face the memory cycle. The difference is that CXMT carries extra growth drivers, while Nanya has a clearer transition path.
Who Remains Under "Overweight" on the Coverage List?
The report concludes with a coverage list, providing ratings and Nov 13 closing prices. Target prices are absent.
Several remain rated "Overweight": TSMC, MediaTek, Alchip.
Plus SMIC, ASMPT, and GlobalUnichip.
Several were downgraded to "Underweight": MetaX, Vanguard, Silergy, Phison.
Additionally, a power semiconductor vendor, AWSC.
These names align more closely with legacy storage and mature nodes.
Cross-referencing the list with the body clarifies the distribution of downgrades. Cloud-side exposure and legacy storage pricing are the two threads driving this adjustment. All else remains unchanged.
Conditions That Would Invalidate the Stance
Bulls can monitor three signals: faster deployment, upward revisions to earnings forecasts, and renewed capex acceleration after Q1 2027.
If any of these materialize, this caution would no longer hold. The report itself identifies faster deployment as a factor that would challenge its cautious view.
Conversely, weakening order momentum and slowing price hikes are the catalysts recognized by the report. It admits the timing for such developments is uncertain.
The entire conclusion rests on one assumption: that the 5GW delay will genuinely translate into postponed equipment orders. If this assumption fails, the downgrade came too early.
The report also leaves two caveats. First, niche memory exhibits high price elasticity; it rises fast and falls fast. Second, its direct exposure to AI demand is low. When the AI narrative cools, the impact is actually limited.
These two points explain why the same report advocates trimming some positions while adding others. They are not facing the same cycle.
Another monitoring window is worth noting: Q1 2027 capex guidance will serve as the initial test for this thesis. An upgrade to guidance would signal that caution should step aside.
How to Interpret This Rating Table
Ratings correspond to risk-adjusted total return over the next 12 to 18 months. The target price timeframe matches. Thus, this is a cross-year assessment, not a quarterly view.
The meaning of the three tiers must be clarified. "Overweight" implies outperforming the average return of the covered universe, "Neutral" denotes market parity, and "Underweight" indicates underperformance.
Ratings describe relative performance, not absolute gains or losses. A sector view rated "Attractive" does not imply every name is a hold.
The prices in the table reflect Nov 13 closing levels. They serve as valuation baselines, not price targets. The report notes that historical prices have not been adjusted for stock splits.
Final reminder: The value of this report lies not in its conclusions, but in its timing. It simultaneously inserts "strong demand" and "exit now" into the narrative. Readers must determine which catalyst plays out first.
Institutional Ledger | Morgan Stanley
Source: Morgan Stanley Research Report · Compiled by TechFlow Research | Report Release Date: November 15, 2026 | Valuation Baseline Date: Closing on November 13, 2026

This article constitutes TechFlow Research's compilation and interpretation of a third-party brokerage research report. The ratings, target prices, earnings forecasts, and related judgments cited herein represent the views of that brokerage's analysts, reflect only their institution's stance, do not represent TechFlow Research's position, and do not constitute investment advice of any kind.
Markets carry risks, decisions require independence. This article should not be used as a basis for buying or selling any securities.













