Hyperscaler capex and custom silicon investment continue to be revised upward, benefiting the European semiconductor equipment supply chain

Overview

This report draws read-across from the results and guidance of 2QCY26 hyperscalers and semiconductor value-chain companies. Its central conclusion is that AI computing demand, enterprise AI adoption, cloud backlog, custom accelerator expansion, and capacity additions in advanced nodes and memory are collectively driving further capex revisions upward. Goldman Sachs believes this has constructive implications for European semiconductor capital equipment and Neocloud-related companies.

Core views

The report's core view is that higher hyperscaler capex is not an isolated phenomenon, but is supported by AI monetization, cloud demand, expanding enterprise workloads, and internal chip-development projects. At the same time, commentary from TSMC, Samsung, Lam Research, and KLA indicates that spending on advanced logic, memory, advanced packaging, and fab equipment continues to strengthen. Accordingly, ASMI benefits from ALD and epitaxy demand, ASML from EUV and advanced DUV demand, BESI from advanced packaging and the proliferation of custom chips, Nebius from tight AI compute supply and expanding enterprise demand, and Technoprobe from probe card demand and increasing testing complexity.

Analysis framework

The report uses a cross-company earnings read-across approach, comparing AI revenue, cloud backlog, capex, and data center construction plans at hyperscalers including Microsoft, Amazon, Alphabet, and Meta with investment guidance and WFE forecasts from semiconductor value-chain companies such as TSMC, Samsung, Lam Research, and KLA, and then mapping these trends to the demand sensitivity of European semiconductor equipment and AI infrastructure companies.

Methodology notes

• Equity research disclosure frameworkGS Factor Profile

  Compares stock characteristics using growth, financial returns, valuation multiples, and composite metrics.

  The report discloses that Goldman Sachs Factor Profile uses standardized rankings and percentiles to assess a stock's growth, financial returns, valuation multiples, and composite characteristics relative to the market and industry peers.

• M&A probability assessment frameworkM&A Rank

  Uses grades from 1 to 3 to assess the probability of a company becoming an acquisition target.

  The disclosure section states that M&A Rank 1 represents high probability, 2 represents medium probability, and 3 represents low probability; when the rank is 1 or 2, M&A factors are typically incorporated into the target price.

• Financial databaseQuantum

  Goldman Sachs' proprietary database for financial history, forecasts, and ratio analysis.

  The disclosure states that Quantum can be used for in-depth analysis of individual companies as well as comparisons among companies across industries and markets.

Asset mapping & comparison

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

• ASMI

  Benefits from expanding investment in advanced logic and DRAM.

  Strengths

  Broader GAA adoption, increased memory production, and higher process intensity support ALD and epitaxy demand.

  Weaknesses

  The report does not provide specific order, revenue, or valuation sensitivity data.

  Comparison

  Compared with equipment companies more focused on lithography or packaging, ASMI has a more concentrated read-across to process deposition intensity.

  Risks

  Demand visibility for related equipment could decline if advanced-node capacity expansion or DRAM investment slows.

• ASML

  Benefits from higher TSMC investment and increased memory capacity demand.

  Strengths

  EUV and advanced DUV demand is supported by advanced-node progression, additional customer layers, and capacity expansion at leading-edge nodes.

  Weaknesses

  The report mentions market questions regarding new lithography entrants from non-Western markets but believes the technological barriers are very high.

  Comparison

  ASML's barriers in lithography are higher than those in ordinary equipment segments, particularly because EUV is more difficult to replace.

  Risks

  Potential competition, geopolitical restrictions, or changes in customer capex timing could affect demand realization.

• BESI

  Benefits from increased investment in custom silicon and advanced packaging.

  Strengths

  Demand for flip-chip, TCB, and HB is expected to increase as internally developed accelerators such as TPU, Trainium, and Maia expand.

  Weaknesses

  The report does not quantify the conversion of advanced packaging investment into BESI revenue.

  Comparison

  Compared with front-end equipment, BESI has a stronger read-across to packaging complexity and heterogeneous integration trends.

  Risks

  The pace of advanced packaging capacity expansion, customer project transitions, and equipment procurement cycles may create volatility.

• Nebius

  Benefits from AI compute demand, capacity constraints, and growing contract backlog.

  Strengths

  Demand is expanding from frontier-model developers to enterprise workloads, supporting the deployment of new capacity and greater enterprise customer exposure.

  Weaknesses

  The report does not disclose Nebius's own capacity, pricing, or profitability data.

  Comparison

  Nebius represents a Neocloud read-across and has more direct exposure to AI compute supply and demand than semiconductor equipment companies.

  Risks

  Data center construction, power supply, GPU or accelerator availability, and the realization of enterprise demand remain uncertain.

• Technoprobe

  Benefits from probe card demand driven by accelerator shipments and capacity expansion at foundries and memory manufacturers.

  Strengths

  Differences in custom chip architectures, chiplets, and advanced packaging increase testing intensity and the importance of identifying defective bare dies before assembly.

  Weaknesses

  The report does not provide information on probe card pricing, orders, or customer concentration.

  Comparison

  Technoprobe has a more concentrated read-across to testing complexity rather than simply wafer capacity expansion.

  Risks

  Demand could fall below expectations if custom chip projects decline, testing intensity is lower than expected, or customer capex is delayed.

Key data

• Microsoft Azure AIOver $100 billion, up 41% yoyThe report states that paid Microsoft 365 Copilot seats exceeded 30 million, with quarterly net additions of more than 10 million.
• Amazon 2026 cash capex outlookApproximately $220 billion, versus approximately $200 billion previouslyThe company expects available power capacity to double by the end of 2027 versus 2025, while AWS backlog reached $496 billion and grew at a triple-digit rate year over year.
• Alphabet 2026 capex outlook$195 billion-$205 billion, versus $180 billion-$190 billion previouslyGoogle Cloud revenue grew 82% yoy, with backlog of approximately $514 billion.
• Meta 2026 capex guidance$130 billion-$145 billion, versus $125 billion-$145 billion previouslyMeta announced a partnership with BlackRock to develop a new 1GW data center. The Goldman Sachs US team raised its cumulative 2027-2028 capex estimate by approximately 14%.
• TSMC FY26 capex outlook$60 billion-$64 billion, versus $52 billion-$56 billion previouslyInvestment priorities include N2 and subsequent nodes, advanced packaging, and the ramp-up of A14 ahead of high-volume production in 2028.
• Lam Research 2026 WFE outlookLow-$150 billion range, versus $140 billion previouslyThe increase is driven by stronger spending on logic, memory, and advanced packaging.
• KLA 2027 reasonable WFE levelApproximately $190 billionThe Goldman Sachs US team forecasts WFE spending of $141 billion/$186 billion/$208 billion in 2026/2027/2028, respectively.

Impact & implications

If cloud capex and AI custom silicon investment continue to expand, equipment intensity in advanced nodes, advanced packaging, and testing could increase further. The report believes this will strengthen the medium- to long-term demand visibility of the European semiconductor equipment supply chain and could improve the demand foundation for Neocloud assets as new AI compute capacity is brought online.

Risks

• Higher cloud provider capex could slow if returns from AI monetization fall short of expectations.
• Delays in advanced-node, advanced-packaging, or memory capacity expansion would weaken the positive read-across for equipment companies.
• Upward WFE revisions depend on simultaneous strength across logic, memory, and packaging; weakness in any one area could affect the cycle.
• Lithography competition from non-Western markets, export restrictions, or geopolitical factors could affect demand and supply for equipment companies such as ASML.
• AI compute infrastructure construction is constrained by power, data centers, supply chains, and the pace at which customer contracts are finalized and realized.

What to watch

• Whether Amazon, Alphabet, Meta, and Microsoft continue to raise their subsequent capex guidance.
• Whether growth in Google Cloud, AWS, and enterprise AI workloads can support higher AI infrastructure investment.
• The pace of capex for TSMC's N2, subsequent nodes, advanced packaging, and A14 mass-production preparations.
• Updates from Samsung on tight memory supply and demand and long-term agreement coverage for 2027-2028.
• Further revisions by Lam Research, KLA, and the Goldman Sachs US team to 2027-2028 WFE spending.
• Subsequent order trends, revenue guidance, and customer demand validation for ASMI, ASML, BESI, Nebius, and Technoprobe.