Morgan Stanley said in a Sept. 7 report that it is maintaining an equal-weight rating on the U.S. semiconductor process equipment sector, arguing that rising wafer fab equipment forecasts are no longer enough on their own to drive a broader re-rating in the group.
The bank said the market is facing a clear split: WFE estimates keep moving up, but investors are becoming less willing to pay for additional upside unless they see stronger evidence that the cycle can extend. It highlighted AEIS, MKS and ONTO as names the market is pricing as if earnings peak in 2027.
Why SPE stocks are not moving higher
Morgan Stanley said hesitation around SPE shares can be explained through two related but distinct frameworks.
The first is the AI return-on-capital framework. The bank said SPE sits inside the broader AI infrastructure buildout, where semiconductor capacity expansion maps directly to gigawatt-scale deployments by hyperscalers. If investors lose confidence in the pace of those deployments, whether because data center construction is delayed or because stress shows up in hyperscaler bonds or credit default swaps, AI infrastructure-linked equities become harder to own.
The bank said its view on the fundamentals has not changed, but the way investors are looking at those fundamentals has. Because it is not comfortable with pushing WFE estimates materially higher from here, Morgan Stanley said it is applying trough-cycle valuation multiples to 2027 or 2028 earnings per share.
The second framework is that expectations are already full. Morgan Stanley said buy-side expectations for 2027 and 2028 WFE are running at about $230 billion and more than $300 billion, versus its own forecasts of $223 billion and $254 billion. In that setup, the market sees limited room for further upward revisions and is again leaning on trough-cycle multiples.
How WFE translates into compute buildouts
Morgan Stanley updated its estimate for how much WFE demand is tied to each gigawatt of compute. While Rubin Ultra specifications are not yet final, the bank estimates roughly $3.4 billion of WFE demand per GW. It also said each $100 billion of AI capital spending would correspond to about $7 billion of WFE demand.
That estimate is below Lam Research’s $9 billion to $10 billion figure. Morgan Stanley said the gap comes from the scope of its work: the bank focused only on Nvidia and did not include Google TPU or Amazon Trainium.
Using that framework, Morgan Stanley’s internet team forecasts hyperscaler capital spending at $1.2 trillion in 2027, equivalent to 29GW of compute. That translates into about $87 billion to $125 billion of incremental WFE demand starting in 2025. If non-AI end markets do not add incremental capacity, 2027 WFE would land around $204 billion to $242 billion. Morgan Stanley’s own forecast is $223 billion, near the middle of that range.
Earlier bull-case drivers have largely played out
The report said several of Morgan Stanley’s earlier bull-case drivers have already materialized.
Intel, in its latest earnings report, guided that 2027 capital spending would be meaningfully above 2026 levels and raised $20 billion in equity. Kioxia’s K3 announcement and Solidigm’s Dalian expansion suggest that NAND greenfield projects have gained market acceptance. Morgan Stanley also said memory makers are pulling capacity forward where possible, and it expects DRAM shipments in the December quarter at Lam Research, Applied Materials and Tokyo Electron to exceed $6 billion.
The bank pointed to three possible sources of incremental upside. On Terafab, it estimates about $2 billion of annual WFE contribution in 2027 and 2028, rising to $6 billion in 2029. On mature logic, Morgan Stanley expects a 2% decline in 2026 followed by 18% growth in 2027. It said the market is underestimating the elasticity of a recovery in mature-node manufacturing.
DRAM supply and the HBM de-spec debate
On DRAM, Morgan Stanley said bit supply is up 36% year to date in 2026, while TrendForce forecasts 31% growth for the full year. Morgan Stanley’s own model points to bit supply growth of 32% in 2026 and 38% in 2027, with HBM bit growth at 55% and 57%.
The bank also addressed market concern over a shift in HBM from 12-layer to 8-layer stacks. It said moving from 12 layers to 8 does not mean every process step falls in direct proportion. Morgan Stanley cited comments from SK hynix saying 12-layer stacked chips are 40% thinner than 8-layer versions and have gaps that are 13% narrower, making control materially harder.
In Morgan Stanley’s view, if HBM roadmaps cannot move to 16 layers and beyond, demand for hybrid bonding would weaken and process control intensity would become harder to push higher.
Intel supplier share could be reshuffled
The report said Intel has historically been an important customer for Applied Materials and Tokyo Electron, with the two companies holding similar shares of Intel capital spending.
Morgan Stanley said that may be changing. Since Lip-Bu Tan became Intel chief executive and Naga Chandrasekaran took charge of foundry operations in June 2024, Intel’s manufacturing strategy has shifted, and supplier share could be redistributed.
The clearest example, the bank said, is process control intensity. It estimates that process control equipment has risen from about 6% of Intel WFE in 2023 to about 10% in 2026, with KLA and Onto Innovation already benefiting.
Valuation mismatch between subsystem names and OEMs
Morgan Stanley also argued that subsystem suppliers are being valued out of line with OEMs. Inventory behavior in this cycle looks different from prior ones. Absolute inventory at Applied Materials and Lam Research rose 8% over the past two quarters, but inventory days fell by 19 days because both companies used stock on hand to meet urgent customer orders. Inventory days are now at their lowest level since the December 2021 quarter.
Against that backdrop, the bank said AEIS and MKS look mispriced. It expects subsystem suppliers to perform better in this cycle because OEMs do not have a backlog of subsystem inventory left to burn off. When the cycle slows, Applied Materials and Lam Research are unlikely to face the same inventory drag from subsystems that hit in the previous cycle.
Even so, AEIS and MKS are currently being valued more like early-cycle stocks, while OEMs are valued on 2028 earnings. Morgan Stanley said that disconnect amounts to mispricing, because the inventory correction that hurt subsystem suppliers last cycle is unlikely to repeat to the same degree this time.
Report note
This article is a summary and interpretation of a Morgan Stanley research report dated Sept. 7, 2026, combined with publicly available market information. Any ratings, target prices, earnings forecasts and related judgments cited here reflect the views of the brokerage analysts and their institution only, and do not constitute investment advice.
Markets carry risk, and investment decisions should be made independently. This article should not be used as the basis for buying or selling any security.
