Upward AI capex revisions continue to support the semiconductor equipment chain, with ASML seen as a core beneficiary
AI summary card
Upward AI capex revisions continue to support the semiconductor equipment chain, with ASML seen as a core beneficiary
Goldman Sachs believes that higher hyperscaler capex, AI accelerator growth, and data center capacity expansion will lift advanced logic and memory demand, with ASML benefiting most due to its EUV monopoly and dual exposure to logic and memory.
- A 1GW data center capacity build is expected to correspond to roughly $50-60 billion in capex and generate about $500-600 million of lithography equipment content.
- The U.S. semiconductor team has revised WFE forecasts upward: about 132 for 2026E, 160 for 2027E, and 174 for 2028E, significantly above prior estimates.
- TSMC's AI accelerator revenue is expected to grow at a mid-to-high-50% CAGR in 2024-2029, above the prior mid-40% expectation, and lift overall USD revenue CAGR to nearly 25%.
- ASML is viewed as the biggest beneficiary of incremental advanced logic and memory demand because it is the sole supplier of EUV tools.
- AI applications are expanding from chatbots and big tech/enterprise use cases to Agentic AI and Physical AI, broadening medium- to long-term compute demand scenarios.
Report interpretation
Overview
This report discusses the drivers and outlook for AI capex, with a focus on hyperscaler capex upward revisions after 1QCY26, AI data center capacity expansion, growth in AI accelerator demand, and the impact on semiconductor equipment, memory, and power semiconductor supply chains. While the report identifies ASML HOLDING NV as a relevant name, the content is more thematic in its outlook on AI infrastructure and the semiconductor equipment ecosystem.
Core views
The core view is that the AI capex cycle still has structural support: on one hand, cloud vendors and large technology companies continue to increase investment in AI infrastructure, lifting data center construction and advanced-node demand; on the other, memory supply may become constrained by cleanroom space and related bottlenecks, which could limit AI expansion and support WFE demand. ASML, as the sole supplier of EUV lithography tools and a beneficiary of both advanced logic and memory expansion, is seen as the most direct beneficiary in the equipment chain; Infineon, with its silicon, silicon carbide, and gallium nitride power semiconductor portfolio, benefits from more complex power-delivery architectures in data centers.
Analysis framework
The report evaluates the beneficiary segments through a top-down sizing of AI capex and data center capacity, a bottom-up estimate of semiconductor equipment content, and supply-chain mapping. Key evidence includes hyperscaler capex upward revisions, changes in WFE forecasts, assumptions for TSMC AI accelerator revenue growth, capex and lithography content corresponding to 1GW of data center capacity, and the evolution path of AI applications.
Methodology notes
1GW data center capacity corresponds to roughly $50-60 billion in capex
The report uses data center capacity as the core unit of AI infrastructure investment and further estimates its pull-through to the semiconductor equipment chain.
1GW data center capacity corresponds to roughly $500-600 million of lithography equipment content
The report maps AI data center expansion to advanced logic and memory demand, and then to lithography equipment content, in order to identify ASML's degree of benefit.
Four-factor comparison across Growth, Financial Returns, Multiple, and Integrated
Goldman Sachs uses its proprietary factor framework to compare stocks' growth, financial returns, valuation multiples, and composite scores versus the market and sector peers.
Assessing a company's likelihood of becoming an acquisition target on a scale of 1 to 3
Goldman Sachs's M&A Rank divides covered companies into high, medium, and low tiers based on their potential probability of being acquired, but the input content does not show ASML's specific score under this framework.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- ASML HOLDING NVPrimary beneficiary
- Strengths
- Sole supplier of EUV tools, with significant exposure to both advanced logic and memory demand; the lithography content corresponding to 1GW of data center capacity provides a quantifiable incremental tailwind.
- Weaknesses
- Highly dependent on advanced-node capex and the semiconductor cycle; if AI capex slows or customer expansion is delayed, order elasticity could weaken.
- Comparison
- Compared with the broader semiconductor supply chain, ASML's beneficiary path is more concentrated in advanced lithography equipment, with greater scarcity value.
- Risks
- Capex downgrades after overheating in AI investment, changes in the pace of memory or logic expansion, export controls, and geopolitical restrictions.
- InfineonBeneficiary of data center power-delivery solutions
- Strengths
- A silicon, silicon carbide, and gallium nitride power semiconductor portfolio that can benefit from the shift to more complex power-delivery solutions in data centers.
- Weaknesses
- Affected by power semiconductor cycle dynamics and volatility in automotive and industrial demand; AI data center contributions may take time to show up.
- Comparison
- Compared with ASML, Infineon's benefit chain is more tilted toward power infrastructure and power management; the direct exposure is slightly weaker, but the application scope is broader.
- Risks
- Price competition in power semiconductors, slowing end-market demand, and changes in technology direction.
- TSMCAI accelerator demand validation indicator
- Strengths
- AI accelerator revenue is expected to grow at a mid-to-high-50% CAGR from 2024 to 2029 and drive overall revenue CAGR close to 25%.
- Weaknesses
- Capital intensive, with high requirements for advanced-node yield, customer concentration, and capacity planning.
- Comparison
- TSMC is an important validation point for AI compute-chip manufacturing demand, and its growth expectations provide indirect support for equipment makers such as ASML.
- Risks
- AI chip demand falls short of expectations, changes in customers' in-house design pace, advanced-node expansion costs, and geopolitical risks.
Key data
- Report date2026-04-02The cover date is 2 April 2026.
- 1GW data center capacity capex$50-60bnThe report estimates that 1GW of DC capacity corresponds to about $50-60bn of capex.
- 1GW data center capacity lithography content$0.5-0.6bnThe report estimates that 1GW of DC capacity can translate into about $500-600 million of lithography equipment content.
- WFE forecast: 2026E132The chart shows the revised 2026E WFE forecast at 132, about 20% above 2025's 110.
- WFE forecast: 2027E160The chart shows the revised 2027E WFE forecast at 160, about 21% above 2026E.
- WFE forecast: 2028E174The chart shows the revised 2028E WFE forecast at 174, about 9% above 2027E.
- TSMC AI accelerator revenue CAGRmid-to-high 50% CAGR in 2024-2029The report says TSMC's expected revenue growth from AI accelerators was revised up from the prior mid-40% range to the mid-to-high-50% range.
- TSMC overall revenue CAGR guidanceapproach 25% in USD termsThe report notes that TSMC's overall revenue CAGR guidance in USD terms was raised to nearly 25%, from about 20% previously.
- Goldman Sachs global equity coverage count3,055 equity securitiesThe disclosure appendix shows that as of 2026-01-01, Goldman Sachs Global Investment Research covered 3,055 stocks.
Impact & implications
If AI capex continues to be revised upward, the investment impact will concentrate in advanced nodes, memory, lithography, power semiconductors, and data center infrastructure. ASML's EUV monopoly makes it highly sensitive to advanced logic and high-end memory demand; power semiconductor companies such as Infineon benefit from the upgrade of data center power architectures from traditional solutions to more complex and efficient ones. For investors, the report emphasizes that AI capex is not only a chip-demand issue, but also transmits upstream to capital equipment through data center capacity, cleanroom bottlenecks, and equipment content intensity.
Risks
- Hyperscaler capex may slow or be re-examined after a high base, leading to downward revisions in AI infrastructure orders.
- AI data center investment could become overheated in phases; if returns on compute monetization are insufficient, demand for related equipment and chips could pull back.
- Memory may become a bottleneck due to cleanroom space and supply-demand tightness, which both supports equipment demand and may constrain AI system delivery pace.
- Demand for advanced semiconductor equipment is heavily affected by global trade restrictions, export controls, and geopolitical factors.
- The report does not disclose ASML's specific rating and target price, so a complete valuation view on the stock cannot be derived directly from it.
What to watch
- Whether hyperscalers' subsequent quarterly capex guidance continues to be revised upward.
- Whether WFE forecasts maintain the upward revision trend from 2026E to 2028E.
- Whether TSMC's AI accelerator revenue growth reaches the mid-to-high-50% CAGR path.
- The pace of new cleanroom and capacity construction at memory makers.
- Changes in ASML EUV orders, shipments, and advanced logic/memory customer demand.
- Infineon's content increase and customer validation progress in data center power semiconductors.