AI datacenter expansion could push WFE demand to a new level
AI summary card
AI datacenter expansion could push WFE demand to a new level
Bernstein believes that if AI datacenter construction continues to advance, global semiconductor equipment demand could rise further from about $200 billion to close to or above $300 billion, leaving room for upward earnings revisions in semiconductor equipment stocks.
- For every additional 1GW/year of AI compute growth, the report estimates that nearly 50K wafers/month of incremental wafer capacity is needed, spanning advanced logic, DRAM/HBM, and NAND.
- Under the 50GW/year incremental scenario, cumulative WFE demand in 2027-2029 is estimated to exceed $700 billion, with annualized spending potentially approaching $300 billion.
- Higher 75GW/year and 100GW/year scenarios would bring greater room for upward earnings revisions and make forward valuations of U.S. semiconductor equipment stocks appear lower.
- The report maintains a positive view on the semiconductor equipment sector and especially favors AMAT, because DRAM/HBM accounts for a relatively high share of incremental wafer capacity and AMAT has greater related exposure.
Report interpretation
Overview
This report discusses how AI datacenter expansion could drive global semiconductor capital equipment demand. Although semiconductor equipment stocks have been highly volatile recently, with some U.S. covered names down about 25%-30% from their highs, they are still up sharply year to date and valuations have expanded significantly. The report's core question is: if AI datacenter construction continues to scale up, how much additional WFE investment will be needed in the future, and do current share prices already fully reflect these expectations?
Core views
The report argues that AI datacenter pipeline capacity is growing rapidly and far exceeds existing deployed scale, which could continue to drive capacity expansion in advanced logic, DRAM/HBM, and NAND. Bernstein estimates that each additional 1GW/year of compute growth requires roughly nearly 50K WSPM of incremental capacity and generates about $8 billion of incremental WFE demand. Under the base case of 50GW/year incremental growth, cumulative WFE demand in 2027-2029 exceeds $700 billion; under higher GW deployment scenarios, equipment demand and the room for upward earnings revisions in semiconductor equipment stocks could be even greater.
Analysis framework
The report uses a bottom-up scenario analysis: it first estimates incremental wafer demand for logic, HBM, DRAM, and NAND based on assumptions for added AI datacenter GW/year capacity by 2030; it then applies capital intensity by equipment type to derive AI-driven WFE demand; finally, combining non-AI baseline WFE spending, company market share, and margin assumptions, it evaluates the impact on revenue, EPS, and valuation for covered semiconductor equipment companies.
Methodology notes
Convert added datacenter GW/year capacity into logic, HBM, DRAM, and NAND wafer demand
The report is based on the Vera Rubin architecture and also incorporates in-rack GPU, CPU, HBM, NAND, as well as ex-rack CPU, DRAM, and NAND demand to estimate the wafer capacity required for each GW of additional compute.
Estimate equipment cost required per 10K WSPM separately for advanced logic, HBM/DRAM, and NAND
The report uses equipment cost intensity assumptions to convert incremental wafer capacity demand into incremental WFE spending; advanced logic is about $3.4 billion per 10K WSPM, HBM/DRAM about $1.4 billion per 10K WSPM, and NAND about $1.3 billion per 10K WSPM.
Use single-year added compute growth in 2030 versus the 2026 baseline as the scenario input
The 50GW scenario means building enough semiconductor capacity by the end of 2029 so that roughly 70GW of compute can be added in 2030 alone, i.e. the roughly 20GW baseline in 2026 plus a 50GW increment, rather than cumulative added capacity from 2026 to 2030.
Assess earnings and valuation of equipment stocks based on assumptions such as market share and incremental margins
The report compares the EPS upside versus current consensus expectations and forward P/FE levels for U.S. semiconductor equipment stocks under the 50GW, 75GW, and 100GW scenarios.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- AMATCore beneficiary; Outperform, target price $525.00
- Strengths
- Strong exposure to key technology inflection points, with valuation still attractive relative to peers; high exposure to incremental DRAM/HBM capacity.
- Weaknesses
- The current price is already close to the report target price, and the sector as a whole has experienced significant valuation expansion.
- Comparison
- The report especially favors AMAT among U.S. semiconductor equipment coverage.
- Risks
- A slowdown in AI buildout, weaker-than-expected memory capex, and valuation pullback.
- LRCXBeneficiary; Outperform, target price $340.00
- Strengths
- Benefits from key inflection points such as GAA, advanced packaging, HBM, and NAND upgrades.
- Weaknesses
- Equipment-cycle and memory-capex volatility may still affect earnings.
- Comparison
- Along with AMAT and KLAC, it is one of the core U.S. semiconductor equipment coverage names.
- Risks
- NAND or HBM investment pacing falls short of expectations, or industry valuation multiples compress.
- KLACBeneficiary; Outperform, target price $197.50
- Strengths
- Has structural growth drivers, a strong competitive position, lower China substitution risk, and disciplined capital allocation.
- Weaknesses
- The current price is above the report target price, and the valuation premium requires continued growth delivery.
- Comparison
- The report believes its high-quality competitive position supports a premium valuation.
- Risks
- A decline in the WFE cycle, lower-than-expected inspection and metrology demand, and contraction in the valuation premium.
- ASMLBeneficiary; Outperform, target price EUR 2,500.00; U.S.-listed target price USD 2,859.00
- Strengths
- Higher DRAM and advanced logic capex supports growth over the next five years; the report maintains it as a top pick.
- Weaknesses
- Valuation is sensitive to long-term growth and high target PE assumptions.
- Comparison
- A core global supplier of high-end lithography, driven by demand at advanced nodes.
- Risks
- Slower advanced logic capacity expansion, delayed customer capex, and FX and geopolitical restrictions.
- Tokyo ElectronBeneficiary; Outperform, target price ¥59,200
- Strengths
- The world's fourth-largest SPE supplier and Japan's largest SPE supplier, with coverage across multiple product lines.
- Weaknesses
- The target price is below the current price shown in the table, limiting near-term return potential.
- Comparison
- A representative Japanese semiconductor equipment company.
- Risks
- Yen movements, competitive pricing, and capex cycle volatility.
- KokusaiBeneficiary; Outperform, target price ¥8,240.00
- Strengths
- A leader in batch ALD; higher adoption at advanced nodes, especially GAA, is beneficial to the company; NAND demand is showing positive signals.
- Weaknesses
- The business is relatively sensitive to specific process technologies and memory cycles.
- Comparison
- Has advantages in the ALD niche segment.
- Risks
- A weaker-than-expected NAND recovery and slower adoption of advanced nodes.
- ScreenNeutral name; Market-Perform, target price ¥12,600
- Strengths
- A leading supplier of cleaning equipment, with one of the lower valuations in coverage.
- Weaknesses
- Fewer specific growth drivers, no increase in cleaning intensity, and intense competition.
- Comparison
- Compared with other Outperform-rated names, the report is more cautious on it.
- Risks
- Margin pressure from a decline in China revenue mix, and competition from TEL, Lam, ACMR, Naura, and others.
- LasertecBeneficiary; Outperform, target price ¥50,000
- Strengths
- An important mask inspection supplier with about 50% share, and the sole supplier of actinic inspection; rollout of ACTIS 200HiT could reaccelerate growth.
- Weaknesses
- After past high growth, it is more dependent on ramp-up of new products.
- Comparison
- Uniquely positioned in the mask inspection niche market.
- Risks
- A potential launch of actinic inspection by KLA would pose a major threat.
- SandiskBeneficiary; Outperform, target price $3,000
- Strengths
- As a pure-play NAND name, it directly benefits from AI-driven data explosion and demand from long-context inference and agent workloads.
- Weaknesses
- NAND cyclicality may still cause earnings volatility.
- Comparison
- Compared with diversified equipment companies, Sandisk has more direct exposure to NAND demand.
- Risks
- Downturns in NAND pricing and demand cycles, and long-term agreements failing to sufficiently reduce earnings volatility.
Key data
- Existing U.S. datacenter capacityabout 51GWAs of the end of June, up about 11GW year over year.
- Datacenter capacity pipelineabout 338GWUp about 217GW over the past 12 months, growing significantly faster than existing deployed capacity.
- Incremental capacity needed per GWabout 46K-50K WSPMCovers advanced logic, DRAM, HBM, and NAND; about half is DRAM, about 20% is NAND, about 15% is HBM, and about 10% is advanced logic.
- Incremental WFE demand per GWabout $8 billionBased on tool-cost metrics and capital-intensity assumptions.
- AI incremental WFE demand under the 50GW scenarioabout $376 billion in 2027-2029Average AI-related incremental WFE demand of about $125 billion per year.
- Total WFE spending under the 50GW scenario$200 billion in 2027, $245 billion in 2028, and $291 billion in 2029, totaling $736 billion over three yearsIncludes roughly $120 billion/year of non-AI baseline WFE spending.
- Higher GW scenariosUnder the 100GW scenario, WFE could reach about $542 billion in 2029Indicating that if AI deployment is more aggressive, WFE demand could be significantly higher than in the base scenario.
- Valuation sensitivityUnder the 50GW scenario, P/FE is in the mid-teens to low-20x range; under the 75GW scenario, low- to mid-teens; under the 100GW scenario, around 10x or lowerCorresponding to scenario valuations for covered U.S. semiconductor equipment companies over the next several years.
- EPS upside revision potentialThe 75GW scenario implies more than 100% upside to current consensus EPS; the 100GW scenario implies more than 150% upsideBased on the report's scenario comparison against CY28/CY29 consensus expectations.
Impact & implications
If AI datacenter buildout remains strong, the current high valuations of semiconductor equipment stocks may be partly absorbed through future earnings upgrades. The report implies that investors should not focus only on recent price gains and valuation expansion, but should also assess how AI compute buildout may raise the upper bound of the WFE cycle. The most direct beneficiaries include equipment suppliers tied to DRAM/HBM, advanced logic, and NAND, with AMAT particularly favored due to its higher DRAM/HBM exposure.
Risks
- AI datacenter buildout may progress more slowly than expected, preventing the projected WFE demand upside from materializing.
- The semiconductor equipment industry may be unable to expand capacity in time under high-GW deployment scenarios, limiting revenue realization.
- Current valuations of semiconductor equipment stocks have already expanded significantly; if earnings upgrades fall short of expectations, valuation compression may occur.
- If the capex mix across DRAM/HBM, NAND, or advanced logic differs from the report's assumptions, the ranking of beneficiaries across companies may change.
- Geopolitics, export controls, China substitution, and customer capex cycles may affect equipment orders.
- The report does not include replacement demand for legacy compute infrastructure before 2030, so actual demand may be higher, but this also means the model results are sensitive to assumption changes.
What to watch
- Whether newly deployed datacenter capacity and pipeline capacity in the U.S. and globally continue to expand.
- Whether annual WFE spending in 2027-2029 moves toward the path of $200 billion, $245 billion, and $291 billion.
- The intensity of DRAM/HBM capex and its share within incremental wafer capacity.
- Orders, backlog, service revenue, and management guidance for CY26/CY27 at companies such as AMAT, LRCX, and KLAC.
- Changes in AI server architectures, including the impact of different architectures such as Vera Rubin, AMD Helios, and TPU on wafer demand.
- Whether NAND datacenter demand, long-term agreements, and long-context inference workloads support storage-related names such as Sandisk.