Morgan Stanley sees the 2026–27 WFE bull case becoming the base case and favors discounted subsystem names AEIS and MKS, plus ONTO.

Overview

Morgan Stanley examines seven debates surrounding semiconductor capital equipment (SPE). Its central conclusion is that the WFE upcycle has become stronger than previously expected, but stock performance remains constrained by concerns about the durability of AI infrastructure returns and by expectations that 2027–28 upside is already largely anticipated.

Core views

Morgan Stanley says the 2026–27 WFE bull case is now its base case because industry spending has continued to exceed physical and capacity constraints. Since December 1, it has raised its 2026 and 2027 WFE forecasts by 26% and 54%, respectively, and increased its 2026 DRAM WFE forecast by $17bn, or 47%. It forecasts more than $120bn of DRAM WFE in 2026–27: $53bn in 2026, up 70% year on year, and $70bn in 2027, up 33%. Leading-logic WFE is forecast to grow 72% in 2026 and 50% in 2027, while the firm expects NAND growth to exceed leading logic, which in turn exceeds DRAM growth, in percentage terms in 2027. The report argues that weak relative performance of SPE stocks reflects an AI-return-on-investment debate more than a deterioration in equipment fundamentals. The market increasingly maps semiconductor capacity requirements to hyperscaler compute deployment measured in gigawatts. If investors lose confidence in deployment—because of a delayed data-center buildout or concerns reflected in hyperscaler bond or CDS markets—AI infrastructure stocks become harder to own even if near-term fundamentals remain intact. Morgan Stanley notes that its coverage trades on trough multiples applied to CY27 or CY28 EPS because it is not comfortable assuming further WFE estimate revisions. A competing view is that buy-side WFE estimates are already full: market expectations are about $230bn for 2027 and $300bn plus or minus for 2028, versus Morgan Stanley estimates of $223bn and $254bn. To test the AI-to-WFE link, Morgan Stanley updates its NVIDIA-based WFE-per-gigawatt calculation. It estimates roughly $3.4bn of WFE per GW for Rubin Ultra, broadly consistent with Vera Rubin, or about $7bn of WFE per $100bn of AI capex. This is below Lam Research's $9–10bn estimate and is described as a lower bound because the calculation focuses only on NVIDIA and excludes TPU and Trainium. Morgan Stanley's Internet team forecasts hyperscaler capex of $1.2tr in 2027, equivalent to 29GW of compute. That translates into $87–125bn of incremental WFE demand from 2025 and, assuming no incremental capacity from non-AI end markets, implies 2027 WFE of $204–242bn versus the $223bn Morgan Stanley estimate. Several earlier upside cases have already materialized. Intel guided 2027 capex to be significantly above 2026 and raised $20bn of equity; Kioxia's K3 announcement and Solidigm's Dalian expansion have made NAND greenfield spending more visible; and memory makers are pulling forward capacity. Morgan Stanley models combined LAM, AMAT and TEL DRAM shipments above $6bn in the December quarter as major fab projects begin tool installation. It nonetheless identifies possible further additions of more than 1%, or $2–3bn, to WFE estimates from faster TeraFab activity and a mature-logic recovery. Its model includes about $2bn of TeraFab WFE in 2027–28 and $6bn in 2029, while mature logic declines 2% in 2026 before growing 18% in 2027. On DRAM, the firm does not make a supply-demand call but emphasizes that usage roadmaps are moving higher. DRAM bits grew 36% year to date through June, and Morgan Stanley models bit-supply growth of 32% in 2026 and 38% in 2027, including HBM growth of 55% and 57% and non-HBM growth of 30% and 36%. Assuming 10% ASP growth and 38% supply growth, it estimates 2027 DRAM TAM at $979bn; 41% bit growth would take the figure to $1tr. Suppliers expecting DRAM to outgrow total WFE in 2027 would imply more than $75bn of DRAM WFE, up 40% year on year, versus Morgan Stanley's $70bn estimate. The report shifts the Intel discussion from whether Intel spends to which suppliers capture the spending. Morgan Stanley models Intel WFE growth of about 70% year on year, contributing 13% of the WFE increase into 2027. Intel historically represented more than 10% of AMAT and TEL revenue, with CY22 Intel-capex shares of 10.4% for AMAT and 10.3% for TEL. The firm does not rule out a supplier reshuffle as Intel's manufacturing strategy changes; it estimates process-control intensity has risen from roughly 6% of Intel WFE in 2023 to about 10% this year, with KLAC and ONTO already benefiting. Morgan Stanley also assesses HBM specification risk. A shift in NVIDIA Rubin Ultra configurations from 12-high to 8-high HBM does not, in its view, reduce every process step proportionately because thinner chips and narrower gaps complicate packaging. It assumes $2bn of equipment per 100kwpm of HBM, but sees the greatest downside in process-control tools and backside grinding/polishing. If the roadmap does not progress to 16-high and beyond, hybrid bonding may not be needed and HBM process-control intensity may cease rising. For stock selection, Morgan Stanley prefers AEIS and MKS over LAM and KLA among its Overweight names and calls ONTO an idiosyncratic share-gain pick. In a CY27 upside case with WFE above roughly $230bn, it sees AEIS earning closer to $20 and MKS closer to $21, implying about 13x and 12x earnings, respectively. AMAT, LAM and KLA trade at a 64% premium to AEIS and MKS, compared with an average 8% premium since 2019. Morgan Stanley argues the discount is excessive because, although subsystem companies should undergrow OEMs in 2026, OEM inventory has not been built at AEIS and MKS to the extent seen in the prior cycle. Therefore, a future slowdown should bring less subsystem inventory whiplash than before.

Analysis framework

Morgan Stanley first updates its WFE forecasts and breaks demand into DRAM, logic and NAND. It then links AI capex and NVIDIA compute deployment to equipment demand using a WFE-per-gigawatt calculation, compares its estimates with market expectations, and tests upside scenarios from Intel, NAND, DRAM, TeraFab and mature logic. Finally, it evaluates supplier exposure, HBM process intensity, inventory dynamics and relative valuation to identify preferred stocks.

Methodology notes

• Industry AnalysisSupply-demand framework

  WFE demand is assessed through end-market capacity expansion, DRAM bit supply, logic demand, NAND projects and fab tooling activity.

  The report uses expected capacity additions and equipment spending by memory and logic customers to build its WFE outlook and identify possible upside drivers.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  Hyperscaler AI capex and compute deployment are translated into semiconductor capacity and then WFE demand.

  Morgan Stanley estimates WFE per NVIDIA gigawatt and applies that relationship to projected hyperscaler capex and compute deployment.

• Valuation methodsP/E and PEG Valuation

  Forward earnings multiples and relative valuation premiums are compared across SPE OEMs and subsystem suppliers.

  The report applies trough multiples to CY27 or CY28 EPS and contrasts AEIS/MKS valuations with AMAT/LAM/KLA to support its relative preference.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Advanced Energy Industries (AEIS)

  Preferred subsystem supplier in Morgan Stanley's WFE-upside scenario.

  Strengths

  Could earn closer to $20 if CY27 WFE exceeds roughly $230bn; trades below its historic forward-P/E range.

  Weaknesses

  Expected to undergrow OEMs in 2026.

  Comparison

  AMAT/LAM/KLA trade at a 64% premium to AEIS/MKS versus an 8% average premium since 2019.

  Risks

  AI deployment uncertainty and a weaker-than-expected WFE cycle.

• MKS Inc. (MKS)

  Preferred subsystem supplier in Morgan Stanley's WFE-upside scenario.

  Strengths

  Could earn closer to $21 if CY27 WFE exceeds roughly $230bn; Morgan Stanley sees reduced risk of prior-cycle inventory whiplash.

  Weaknesses

  Expected to undergrow OEMs in 2026.

  Comparison

  Trades at a historic discount to SPE OEMs alongside AEIS.

  Risks

  AI deployment uncertainty and a weaker-than-expected WFE cycle.

• ONTO Innovation (ONTO)

  Morgan Stanley's idiosyncratic share-gain pick.

  Strengths

  Morgan Stanley sees early benefits from higher Intel process-control intensity.

  Risks

  HBM specification changes could pressure process-control demand.

• Applied Materials (AMAT), Lam Research (LAM) and KLA (KLA)

  SPE OEM and process-control comparables.

  Strengths

  Beneficiaries of broad WFE spending; AMAT and LAM inventory days are near cycle lows.

  Weaknesses

  Trade at a substantial premium to AEIS/MKS.

  Comparison

  AMAT/LAM/KLA trade at a 64% premium to AEIS/MKS versus an 8% historical average since 2019.

  Risks

  A slowdown in AI deployment or a change in HBM process intensity could affect demand.

• Tokyo Electron (TEL)

  Potential beneficiary of Intel WFE growth.

  Strengths

  History suggests TEL is the largest beneficiary of Intel WFE; it held 10.3% of Intel capex in CY22.

  Weaknesses

  Supplier share may be reshuffled as Intel's manufacturing strategy changes.

  Comparison

  TEL's CY22 Intel-capex share was close to AMAT's 10.4%.

  Risks

  Intel supplier-share changes could alter spending capture.

Key data

• 2026/27 WFE forecast revisions+26% / +54% since December 1Morgan Stanley's upward revisions to its WFE forecast.
• 2026/27 DRAM WFE$53bn / $70bnForecasts for 2026 and 2027; up 70% and 33% year on year, respectively.
• Leading-logic WFE growth72% in 2026; 50% in 2027Morgan Stanley forecast.
• WFE per AI capex~$7bn per $100bn of AI capexNVIDIA-based estimate; described as a lower bound.
• Hyperscaler capex and compute$1.2tr and 29GW in 2027Morgan Stanley Internet-team forecast used in the WFE translation.
• 2027 AI-implied WFE range$204–242bnCompared with Morgan Stanley's $223bn 2027 estimate.
• DRAM bit-supply growth32% in 2026; 38% in 2027Morgan Stanley forecast; HBM bits are modeled at 55% and 57% growth.
• OEM premium to AEIS/MKS64%Compared with an 8% average premium since 2019.

Impact & implications

Morgan Stanley views continued WFE strength as supportive for semiconductor-equipment demand, but says broad stock upside depends on clearer evidence that AI compute deployment and associated returns can persist. Its relative preference is for AEIS and MKS because their valuation discount to OEMs appears inconsistent with lower expected inventory-unwind risk, while ONTO is supported by a distinct share-gain thesis.

Risks

• If confidence in hyperscaler gigawatt deployment weakens because of data-center delays or credit concerns, AI infrastructure stocks may remain difficult to support.
• The report sees the greatest HBM-related downside risk in process control and backside grinding/polishing if the roadmap does not advance to 16-high stacks and beyond.
• Morgan Stanley is not comfortable assuming further WFE forecast revisions because market expectations for 2027–28 may already be full.
• A supplier reshuffle at Intel could change which equipment companies capture Intel's higher spending.

What to watch

• Evidence that hyperscaler AI capex and compute deployment can reach the modeled 29GW in 2027.
• Whether 2027 WFE tracks Morgan Stanley's $223bn estimate, the $204–242bn AI-implied range, or higher market expectations.
• Intel's 2027 capex increase and the distribution of orders among equipment suppliers.
• DRAM capacity pull-forwards, bit-supply growth and whether DRAM WFE moves above Morgan Stanley's $70bn 2027 estimate.
• The HBM roadmap, including whether it progresses beyond 12-high or 8-high configurations toward 16-high stacks.
• Potential TeraFab acceleration and a mature-logic recovery as sources of additional WFE demand.