AI data center expansion could push annualized WFE demand toward $300B+, leaving further upside for semiconductor equipment
AI summary card
AI data center expansion could push annualized WFE demand toward $300B+, leaving further upside for semiconductor equipment
Bernstein translates future AI data center GW addition scenarios into wafer demand and WFE investment, arguing that a 50GW/year incremental scenario alone is enough to support more than $700B of cumulative WFE in 2027-2029, and it remains positive on the semiconductor capital equipment sector, with a particular preference for AMAT.
- Although major U.S. semiconductor equipment stocks have pulled back about 25%-30% from their highs, the sector is still up more than 80% year to date, and valuations have expanded significantly.
- The report estimates that every additional 1GW/year of AI data center compute growth requires close to 50K WSPM of new advanced logic, DRAM/HBM, and NAND capacity.
- The 50GW/year incremental scenario implies cumulative WFE of about $736B for 2027-2029, with annualized WFE of about $291B in 2029.
- Under stronger 75GW/year and 100GW/year scenarios, forward P/FE for U.S. semiconductor equipment stocks could fall to the low- to mid-teens, or even 10x or below, while also implying significant EPS upgrades.
- Bernstein remains bullish on the semiconductor equipment sector and believes AMAT has the highest exposure to incremental DRAM/HBM wafer capacity, making it particularly attractive.
Report interpretation
Overview
This report discusses the potential boost to global wafer fab equipment (WFE) demand from AI data center expansion. Bernstein notes that semiconductor equipment stocks have recently been volatile, with covered U.S. names pulling back about 25%-30% from their highs, but the sector is still up more than 80% year to date. The core question is: after the current valuation expansion, how much additional WFE will future AI data center construction actually require, and will it be enough to continue supporting further upside in equipment stocks? By converting 2030 incremental data center GW scenarios into wafer demand, equipment capital intensity, and company earnings impact, the report concludes that even if the previously discussed $200B WFE market was seen as somewhat idealized, the industry may approach that level as soon as next year, while reasonable AI buildout scenarios could push annualized WFE demand toward $300B+.
Core views
The core view of the report is that the AI data center construction pipeline already far exceeds the currently installed base, and that if future incremental compute deployments of tens of GW or more materialize, they will require substantial new advanced logic, DRAM/HBM, and NAND capacity. Bernstein estimates that each additional 1GW/year of compute capacity growth requires close to 50K WSPM of new capacity, of which about half comes from DRAM, about 20% from NAND, about 15% from HBM, and about 10% from advanced logic. Using a 50GW/year incremental scenario as the base case, cumulative WFE demand in 2027-2029 would be about $736B, with 2027/2028/2029 WFE at about $200B, $245B, and $291B, respectively. If investors continue to believe in the multi-year AI data center expansion trend, the report argues that semiconductor equipment stocks still have further upside even at current levels.
Analysis framework
The report uses bottom-up scenario modeling: first, it takes the annual incremental GW in 2030 versus 2026 as the input and estimates incremental wafer demand for logic, DRAM, HBM, and NAND; second, it applies the equipment capital intensity of different technology steps to derive AI-driven WFE demand; finally, it adds non-AI baseline WFE and combines market share, incremental margins, and company assumptions to assess the impact on revenue, EPS, and valuation for the covered semiconductor equipment companies.
Methodology notes
Convert annual incremental data center compute GW in 2030 relative to 2026 into wafer capacity and equipment demand.
The model does not count cumulative new capacity added from 2026-2030; instead, it measures how much single-year incremental compute capability in 2030 increases relative to the 2026 baseline. For example, a 50GW scenario means that by 2030, single-year incremental capability is about the 2026 baseline of 20GW plus 50GW.
Estimate wafers required per GW based on demand from the Vera Rubin architecture, CPUs, HBM, DRAM, and NAND.
The report uses the current Vera Rubin architecture to estimate in-rack GPU, CPU, HBM, DRAM, and NAND wafer demand, and adds out-of-rack CPU, DDR DRAM, and data center NAND demand.
Estimate WFE based on equipment cost per 10K WSPM for advanced logic, HBM/DRAM, and NAND.
The report assumes equipment costs per 10K WSPM of about $3.4B for advanced logic/foundry, about $1.4B for HBM/DRAM, and about $1.3B for NAND, concluding that each additional 1GW/year requires about $8B of incremental WFE.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- AMATOne of the core beneficiaries of AI-driven WFE upside, and the report is especially positive on it.
- Strengths
- It has the highest exposure to incremental DRAM/HBM wafer capacity, along with strong exposure to key technology inflection points, and its valuation remains attractive relative to peers.
- Weaknesses
- The stock price has already risen significantly, and the target price in the table is below the 2026-07-17 closing price.
- Comparison
- Compared with other U.S. equipment stocks, AMAT benefits more in scenarios with a high contribution from incremental DRAM/HBM.
- Risks
- A slowdown in AI data center construction, lower-than-expected memory capex, and weaker-than-expected equipment order conversion.
- LRCXBenefits from key technology inflection points including GAA, packaging, HBM, and NAND upgrades.
- Strengths
- The report says the company benefits from key technology inflection points and maintains supportive commentary on CY26/27.
- Weaknesses
- Valuation has expanded, and it is sensitive to the memory and NAND upgrade cycle.
- Comparison
- Like AMAT, it is a core covered U.S. semiconductor equipment name, with similar benefit direction but different subsegment exposure.
- Risks
- Uncertainty around the pace of NAND upgrades, customer capex cycles, and the pace of AI buildout.
- KLACHas structural growth drivers under the positive WFE trend.
- Strengths
- Strong and durable competitive position, lower China substitution risk, and stronger capital allocation discipline.
- Weaknesses
- Its quality characteristics may already be reflected in a premium valuation.
- Comparison
- Relative to some equipment peers, KLAC has a more prominent competitive advantage in inspection/metrology.
- Risks
- A slowdown in advanced-node investment, contraction of valuation premium, and demand upgrades falling short of expectations.
- ASMLA key beneficiary of rising advanced logic and DRAM capex.
- Strengths
- The report expects solid growth over the next five years and maintains Outperform and top pick, with a target price of EUR 2,500 / USD 2,859.
- Weaknesses
- Valuation depends on a relatively high target PE, and exposure is concentrated in EUV/advanced-node investment.
- Comparison
- It holds a unique lithography position in the global semiconductor equipment chain.
- Risks
- Advanced logic demand, export restrictions, and customer capex volatility.
- Tokyo ElectronThe world's fourth-largest SPE supplier and Japan's largest SPE supplier, benefiting from expanding equipment demand.
- Strengths
- It covers six major product areas, and the report expects it to benefit from share gains and more competitive pricing after yen depreciation.
- Weaknesses
- The target price is below the 2026-07-17 closing price, implying limited short-term upside.
- Comparison
- A leading Japanese equipment company with broad product coverage.
- Risks
- Yen volatility, intensifying competition, and a downturn in the equipment cycle.
- KokusaiA beneficiary of rising batch ALD adoption in advanced nodes and NAND.
- Strengths
- Leader in batch ALD, with improving adoption prospects in GAA and NAND.
- Weaknesses
- The target price is below the 2026-07-17 closing price, and valuation appears relatively fully digested.
- Comparison
- Compared with diversified equipment vendors, Kokusai is more focused on the deposition subsegment.
- Risks
- A slower-than-expected recovery in NAND investment and slower-than-expected adoption in advanced nodes.
- ScreenA cleaning equipment supplier rated Market-Perform.
- Strengths
- Leading in cleaning equipment, valuation is relatively low within coverage, and potential upside from panel-level packaging is worth watching.
- Weaknesses
- Lacks clear specific growth drivers, cleaning intensity has not increased, competition is intense, and declining China revenue could create downside margin risk.
- Comparison
- Compared with Outperform names, the report is more cautious on Screen's growth visibility.
- Risks
- A decline in China revenue share, pressure from global and Chinese competitors, and margin compression.
- LasertecA beneficiary related to mask inspection and actinic inspection.
- Strengths
- The leading mask inspection supplier with about 50% share, and the sole supplier of actinic inspection; revenue growth could re-accelerate after the launch of ACTIS 200HiT.
- Weaknesses
- Highly dependent on specific inspection technologies and customer adoption timing.
- Comparison
- Compared with general equipment vendors, Lasertec is more concentrated in the inspection niche.
- Risks
- KLA potentially launching actinic inspection would pose a major threat.
- SandiskA direct beneficiary of AI-driven data explosion and rising NAND demand.
- Strengths
- As a pure-play NAND name, it benefits from long-context demand driven by agentic and inference workloads; new storage LTAs could reduce earnings cyclicality and drive rerating.
- Weaknesses
- NAND pricing and supply-demand cycles may still cause significant volatility.
- Comparison
- Compared with equipment vendors, Sandisk is a more direct expression of demand-side elasticity in memory.
- Risks
- Deterioration in NAND supply-demand, lower-than-expected AI storage demand, and failure of valuation rerating to materialize.
Key data
- Active U.S. data center capacityAbout 51GWAs of the end of June, up about 11GW year over year.
- Data center capacity pipelineAbout 338GWUp about 217GW over the past 12 months, growing much faster than active capacity.
- Wafer consumption per Vera Rubin rackAbout 65 wafersCovers logic, HBM, DRAM, and NAND.
- Capacity required per additional 1GW/yearAbout 46K-50K WSPMIncludes in-rack and out-of-rack DRAM, logic, HBM, and NAND.
- WFE per additional 1GW/yearAbout $8BEstimated based on equipment capital intensity across technology steps.
- Cumulative WFE in 2027-2029 under the 50GW scenarioAbout $736BIncludes about $376B of AI incremental WFE and about $120B per year of non-AI baseline WFE.
- Annualized WFE under the 50GW scenario2027E $200B; 2028E $245B; 2029E $291BThe report believes annualized WFE approaching or exceeding $300B is achievable under reasonable assumptions.
- Annualized WFE under the 100GW scenario2027E $200B; 2028E $371B; 2029E $542BDemand could be significantly above the base-case scenario under higher-GW deployment scenarios.
- Valuation implication of the 50GW scenarioForward P/FE for U.S. semiconductor equipment stocks at roughly mid-teens to low 20xImplies significant upside versus current consensus EPS.
- Valuation implication of the 75GW and 100GW scenarios75GW implies low- to mid-teens P/FE; 100GW implies 10x or belowThe 100GW scenario implies more than 150% upside versus current consensus EPS.
Impact & implications
The investment implication of the report is that AI data center expansion could generate WFE demand far above traditional cycle assumptions, making current seemingly elevated equipment stock valuations more reasonable under a higher earnings framework. If AI buildout continues to be delivered, revenue and EPS for equipment stocks may continue to be revised upward, leaving further room for sector gains. Structurally, DRAM/HBM contributes the most to incremental wafer demand, so companies with greater exposure to related process steps and equipment benefit more; Bernstein specifically highlights AMAT as attractive because it has the highest DRAM/HBM exposure.
Risks
- The AI data center construction pipeline may fail to materialize as planned, making the incremental GW scenarios too high.
- Semiconductor equipment industry capacity or supply chains may be unable to keep up with higher WFE demand scenarios.
- The capital intensity assumptions used in the report may be viewed as too low or too conservative, and actual costs and timing may differ.
- Current equipment stock valuations have already expanded significantly; if earnings upgrades fail to materialize, valuation pullback risk will be high.
- China substitution, export restrictions, customer capex cycles, and memory pricing cycles may all affect the degree of benefit for different equipment companies.
- The report does not include replacement demand for legacy computing infrastructure before 2030; actual demand could be higher, but that also increases model uncertainty.
What to watch
- Whether active capacity and construction pipelines for U.S. and global data centers continue to expand.
- Whether annual WFE spending in 2027-2029 progresses toward $245B, $291B, or even higher levels.
- Whether capex allocation across DRAM/HBM, NAND, and advanced logic matches the report's assumptions.
- Whether orders, revenue, and EPS for U.S. equipment stocks such as AMAT, LRCX, and KLAC continue to be revised upward versus market consensus.
- Changes in the share of non-U.S. equipment vendors such as ASML, Tokyo Electron, Kokusai, and Lasertec in advanced-node and memory investment.
- The actual demand pull on CPU, DRAM, HBM, and NAND from AI inference, agentic workloads, and long-context applications.