BofA warns of fragilities in the AI capex boom and expects continued pressure on European equities and cyclical sectors

Overview

Centered on whether the AI capex boom can persist, the report analyzes current earnings expectations, risk premia, model competition, financing conditions, power constraints, and US macro and energy risks. It concludes that markets are pricing in excessive certainty around the realization of AI profits and that European equities do not offer sufficient risk compensation. It therefore maintains a negative view and adopts a primarily defensive portfolio across sectors, styles, and countries.

Core views

AI capex has become a dominant force in global equity markets and macroeconomic performance. Consensus expectations for US hyperscaler capex over the next 12 months have risen from less than $300 billion at the start of 2025 to $940 billion, equivalent to 3% of US GDP. This spending has helped global growth and corporate earnings remain resilient despite trade conflicts and the energy shock caused by the US-Iran war. It has also produced the typical effects of a large-scale capex cycle: equipment and service suppliers recognize revenue immediately, while spenders capitalize capex on their balance sheets, with the related costs only entering the income statement later through depreciation. As a result, aggregate market margins receive a near-term boost from revenues being recognized earlier and costs later. Capital flows have also made the market highly momentum-driven, with industries close to the AI spending chain significantly outperforming over the past three years. The report argues that current prices already embed exceptionally optimistic assumptions about AI returns. Consensus margins for global equities over the next 12 months have reached an all-time high of 14.5%. Technology is expected to contribute 25% of global profits over the next 12 months, above its 15% average over the past decade and the 13% peak during the technology boom cited in the report. Markets also view this profitability as a persistent trend, with consensus compound EPS growth expectations of 18% over the next three years and 22% over the next five years, close to levels normally achieved only during post-recession recoveries. Meanwhile, the global equity risk premium has fallen to 3%, its lowest since 2002, while the European risk premium has declined to 4.5%, its lowest since the beginning of this century. BofA believes these high earnings expectations are not accompanied by risk compensation commensurate with the uncertainty surrounding AI commercialization. Whether AI investment can ultimately persist depends on whether hyperscaler investment returns can cover the cost of capital, which in turn depends primarily on competitive intensity and pricing power. The report compares three historical paths: the cloud computing market of the 2010s was able to absorb massive depreciation without collapsing because it had only a few dominant players and strong pricing power and margins; fiber investment in the 1990s reversed sharply because of excess capacity, deteriorating pricing power, and falling profitability; and although the airline industry created enormous value for consumers, intense competition converted most economic value into consumer surplus, leaving corporate margins persistently thin. The report accordingly emphasizes that the usefulness of AI technology does not necessarily mean AI suppliers will generate sufficient profits to support current valuations. Recent market performance indicates that this fragility persists. The MSCI Europe AI Capex Select Index underperformed the broader market by 22% between June 22 and July 29. After hyperscalers' second-quarter results and AI revenues proved better than feared, the index rebounded by 14% relative to the market over the following two weeks, but AI capex beneficiaries have recently underperformed again by 6%. Given the extremely optimistic current pricing, the report expects the concerns that triggered the July reversal to intensify again over the coming months. The first source of pressure is competition at the AI model layer. Several Chinese open-weight models offer capabilities close to frontier models at a lower per-task price, encouraging customers to switch to lower-cost alternatives. The Silicon Data Token Expenditure Index has fallen 50% since May, reflecting both price cuts by frontier laboratories such as Anthropic and OpenAI and increased use of low-cost Chinese models. Because frontier model providers contribute a significant portion of hyperscalers' AI revenue, declines in model prices and pricing power could propagate through the entire AI ecosystem, undermining the market's elevated expectations for technology-sector and aggregate margins. The second source of pressure is the cost of capital. As enormous investment commitments turn hyperscalers' previously abundant free cash flow negative, AI capex is becoming increasingly dependent on debt financing. Rising bond supply has more than doubled these companies' credit spreads from last year's low of 55 basis points to 120 basis points, while risk-free rates have risen to a two-decade high. The report notes that historical capital cycles typically end when expected investment returns fall below the cost of capital, and AI investment is now coinciding with greater reliance on financing and higher debt costs. The third source of pressure comes from electricity, infrastructure, and political constraints. Power generation and transmission bottlenecks could prevent planned data centers from obtaining sufficient electricity, while public opposition to data-center construction is also increasing in the US. The report cites Texas as an example: after data-center development became an issue in the gubernatorial race, Governor Greg Abbott announced a temporary halt to approvals for AI-related construction projects. Even if AI business models ultimately prove sufficiently profitable, these physical and policy constraints could still obstruct the sustained expansion in spending anticipated by the market. Beyond AI risks, the US labor market and energy supply could also weigh on risk assets. Three-month US employment growth is again close to zero, which has typically warned of macro weakness in past cycles. Weak retail sales, declining consumer confidence, and slowing inflation momentum have also prompted BofA's US rates strategy team to turn bullish on duration following the recent bond selloff. Meanwhile, the US-Iran conflict remains unresolved, and energy shipments through the Strait of Hormuz are close to zero. The US Strategic Petroleum Reserve has fallen from 415 million barrels to 296 million barrels, while Chinese crude imports have declined from 12 million barrels per day in the prior year to approximately 7 million barrels per day, indicating that calm in the oil market partly depends on temporary measures. Refined-product crack spreads remain high, and US gasoline prices have rebounded to a two-month high of $4.8 per gallon, placing additional pressure on consumers. Prediction markets assign only a 19% probability that Strait of Hormuz shipments will return to normal before early December. Eurozone macro surprises have turned positive again, but the report believes the improvement may be difficult to sustain. Tight credit conditions, the fading World Cup effect, and rising natural gas prices could cause PMI to reach the peak of this recovery in July before falling to 49 by year-end. BofA's Stoxx 600 framework indicates that every three-point decline in eurozone PMI reduces the index's macro-implied fair value by approximately 5%; every 100-basis-point increase in real yields likewise reduces fair value by approximately 5%. The report currently places greater emphasis on downside from rising risk premia and weakening EPS expectations, while also arguing that if US employment and inflation continue to slow, falling bond yields will weigh on the value-oriented structure of European indices. Taking together the risks from AI commercialization, US employment, and energy supply, BofA expects the Stoxx 600 to fall to 610 by end-2026 and to 580 in early Q2 2027. The related charts summarize this as nearly 10% downside by year-end and more than 10% downside by Q2 next year. European equities remain underweight relative to global equities because their earnings structure remains at a long-term disadvantage. Cyclicals have already underperformed defensives by 6% over the past two months, reversing approximately one-third of their gains since March, but their prior relative performance had reached a 30-year high. If AI momentum continues to reverse, risk premia rise, and bond yields fall, BofA expects cyclicals to have approximately 6% further downside relative to defensives by early 2027. In style allocation, the report is underweight cyclicals relative to defensives and value relative to growth, while overweight quality stocks and small caps relative to large caps. Food and beverages and telecommunications are the preferred defensive overweights because they have the strongest negative correlations with AI proxy indicators and typically outperform when risk premia rise. Pharmaceuticals also remain overweight, as declining bond yields should ease previous pressure. Software remains overweight due to its defensive growth characteristics and better-than-feared resilience against AI competition. The report expects approximately 5% further relative upside even after the sector outperformed by 30% over the past month. Real estate is overweight because of downside risks to bond yields, while chemicals are overweight because real rates may decline. Luxury goods were upgraded after underperforming by 50% over the past three years, as current pricing is viewed as implying an excessively pessimistic global growth scenario and the sector could benefit from Chinese fiscal stimulus. Among cyclical industries, BofA remains underweight banks, capital goods, and semiconductors, and has downgraded financial services from neutral to underweight. Banks are at a 15-year high relative to the market, while both banks and financial services face pressure from rising risk premia, falling bond yields, and slowing eurozone growth. Semiconductors have underperformed by 15% since late June, but the AI capex narrative could deteriorate further. Airlines, autos, utilities, insurance, energy, mining, and construction materials remain neutral. The US-Iran standoff has undermined the baseline scenario of falling oil prices on which airlines depend; structural problems in autos limit the potential for a cyclical catch-up; utilities have become attractive again after their recent correction; risks for insurance are skewed to the downside; energy and mining have experienced a sharp reversal and recent pullback, respectively; and macro-implied downside for construction materials is now relatively limited. In country allocation, BofA is overweight Switzerland to benefit from its defensive performance when risk premia widen, and overweight Germany because markets are pricing in a pessimistic growth outlook even as fiscal stimulus approaches. The UK was recently upgraded from neutral to overweight. France remains neutral because of policy and macro uncertainty. Spain and Italy are both underweight, primarily because their banking sectors may underperform. Small caps and German equities are viewed as undervalued cyclical hedges within an overall defensive portfolio and could benefit from the recent improvement in eurozone macro data. Overall, the report does not deny that AI capex can persist; rather, it argues that current prices leave insufficient margin of safety for uncertainty surrounding competition, financing, infrastructure, and the macroeconomic path.

Analysis framework

The report first measures the scale of AI capex and its effects on growth, profits, and market momentum, then compares current earnings expectations and equity risk premia with historical levels. It subsequently uses capital-cycle cases such as cloud computing, fiber optics, and airlines to identify how competition and investment returns determine whether a boom can persist, and tests current fragilities using model prices, credit spreads, electricity, and approval data. Finally, it incorporates AI risks alongside macro variables such as employment, inflation, energy supply, PMI, real yields, and credit spreads into models based on long-term historical relationships, using the results to formulate forecasts for the Stoxx 600 and style, sector, and country allocations.

Methodology notes

• Industry/Sector Analysis FrameworkSupply-demand framework

  Analysis of supply, competition, and pricing power in AI computing and model markets

  By examining the increase in model suppliers, penetration of low-priced open-weight models, declining token prices, and data-center power constraints, the report assesses whether expanding AI supply can generate prices and profits commensurate with capex.

• Cycle and Business Conditions FrameworkCapacity/Equipment Cycle (Juglar)

  Comparison of large-scale technology capex cycles

  The report compares AI investment with historical buildout cycles in cloud computing, fiber optics, radio, and railways, focusing on how overinvestment, capacity, depreciation, pricing power, investment returns, and the cost of capital determine whether a boom continues or reverses.

• Cycle and Business Conditions FrameworkBusiness-Cycle Inflection-Point Analysis

  Identification of inflection points in PMI, employment, and consumer data

  The report uses three-month US employment growth, retail sales, consumer confidence, and eurozone PMI trends to determine whether macro momentum is weakening and maps these signals to equity risk premia and relative sector performance.

• (Out-of-vocabulary Method)

  Long-term historical regression and fitting of macro drivers

  BofA conducts weighted regressions using historical relationships between asset performance and variables such as PMI, real bond yields, policy uncertainty, the euro, and oil prices. It then combines actual data, analyst forecasts, and outlier adjustments to generate forecasts for the Stoxx 600 and relative sector performance. The report also cautions that historical correlations and forecasts may not persist or materialize.

Asset mapping & comparison

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

• Stoxx 600/European Equities

  The report maintains a negative view and is underweight relative to global equities.

  Strengths

  Eurozone macro surprises have recently turned positive again, while some areas benefit from low valuations and fiscal stimulus support.

  Weaknesses

  The earnings structure is at a long-term disadvantage relative to global markets, and current pricing does not adequately compensate for uncertainty surrounding AI profits, risk premia, and the macro outlook.

  Comparison

  Underweight relative to global equities; targets of 610 at end-2026 and 580 in early Q2 2027.

  Risks

  A reversal in AI momentum, rising risk premia, downward revisions to EPS expectations, US economic weakness, and energy supply shocks.

• Food and Beverages, Telecommunications, Pharmaceuticals

  The report identifies them as its principal defensive overweight positions.

  Strengths

  Food and beverages and telecommunications have relatively high negative correlations with AI proxy indicators and typically outperform when risk premia widen; pharmaceuticals could benefit from falling bond yields.

  Weaknesses

  Pharmaceuticals are relatively sensitive to a weaker US dollar because of their high share of US revenue.

  Comparison

  Food and beverages and pharmaceuticals are the report's preferred defensive overweights, while the recent correction in telecommunications is considered excessive.

  Risks

  If risk premia decline rather than rise or bond yields remain elevated, the relative advantage of defensive allocations may be weaker than the report expects.

• Software

  Remains overweight as a defensive-growth allocation.

  Strengths

  Resilience against AI competition has been better than feared, and the sector has defensive growth characteristics.

  Weaknesses

  It has already outperformed by 30% over the past month and still faces risks from AI competition.

  Comparison

  The report expects approximately 5% further upside relative to the market.

  Risks

  If AI competition materially erodes software business models, current resilience may not persist.

• Banks, Financial Services, Capital Goods, Semiconductors

  Banks, capital goods, and semiconductors remain underweight, while financial services were downgraded from neutral to underweight.

  Weaknesses

  Banks are at a 15-year high relative to the market; semiconductors and capital goods are highly exposed to the AI capex narrative; and financials are vulnerable to falling yields and rising risk premia.

  Comparison

  Banks and capital goods are the report's preferred cyclical underweights; semiconductors have underperformed by 15% since late June.

  Risks

  AI capex falling short of expectations, widening credit risk premia, declining bond yields, and slowing eurozone growth.

• Airlines, Autos, Utilities, Insurance, Energy, Mining, Construction Materials

  The report maintains a neutral view on all of these sectors.

  Strengths

  Utilities have become attractive again after their recent correction; macro-implied downside for construction materials is limited.

  Weaknesses

  Airlines are affected by oil-price uncertainty, autos face structural problems, risks for insurance are skewed to the downside, and momentum in energy and mining has recently weakened.

  Comparison

  None are included among the principal overweight or underweight positions.

  Risks

  The US-Iran standoff, oil prices and refined-product costs, structural competitive problems, and changes in macro momentum.

• Small Caps, Quality Stocks, Growth Stocks

  Overweight small caps relative to large caps and overweight quality stocks; underweight value relative to growth.

  Strengths

  Quality stocks could benefit from the rate and risk-premium environment; small-cap pricing is considered excessively pessimistic.

  Weaknesses

  Small caps remain cyclically sensitive.

  Comparison

  As a style allocation within an overall defensive portfolio, small caps provide catch-up exposure to improving eurozone data.

  Risks

  If eurozone PMI falls to 49 as forecast in the report, the cyclical catch-up in small caps may be limited.

• Swiss, German, UK, French, Spanish, and Italian Equities

  Overweight Switzerland, Germany, and the UK; neutral France; underweight Spain and Italy.

  Strengths

  Switzerland is defensive; Germany has pessimistic pricing and impending fiscal stimulus; the UK was recently upgraded from neutral to overweight.

  Weaknesses

  France faces policy and macro uncertainty; Spain and Italy are affected by the potential underperformance of their banking sectors.

  Comparison

  Germany and the UK are overweight, France is neutral, and Spain and Italy are underweight.

  Risks

  Weak bank performance, policy uncertainty, and renewed slowing in eurozone growth momentum.

• Luxury Goods, Real Estate, Chemicals

  The report is overweight all three, with luxury goods recently upgraded.

  Strengths

  Luxury goods are pessimistically priced after underperforming by 50% over three years and could benefit from Chinese fiscal stimulus; real estate benefits from falling bond yields; chemicals benefit from declining real rates.

  Weaknesses

  Luxury goods still depend on global growth and the realization of Chinese stimulus, while real estate and chemicals are sensitive to interest rates and the macro path.

  Comparison

  These positions represent valuation-recovery or rate-sensitive overweights within the defensive portfolio.

  Risks

  Chinese stimulus falling short of expectations, further weakening in global growth, or real rates failing to decline.

Key data

• US hyperscaler capex over the next 12 months$940 billionEquivalent to 3% of US GDP, up from less than $300 billion at the start of 2025
• Consensus margin for global equities over the next 12 months14.5%All-time high
• Technology sector's share of global profits over the next 12 months25%All-time high; the average over the past decade was 15%, while the technology-boom peak cited in the report was 13%
• Consensus compound EPS growth18% over the next three years and 22% over the next five yearsThe report argues that such growth rates normally occur only during post-recession recoveries
• Global equity risk premium3%Lowest since 2002
• European equity risk premium4.5%Lowest since the beginning of this century
• Relative performance of European AI capex beneficiaries-22%, +14%, -6%Underperformed by 22% from June 22 to July 29, outperformed by 14% over the following two weeks, and recently declined by 6% relative to the market again
• Silicon Data Token Expenditure IndexDown 50% since MayReflects price cuts for frontier models and customers switching to lower-cost models
• Hyperscaler credit spreads120 basis pointsMore than doubled from last year's low of 55 basis points
• US Strategic Petroleum ReserveDown from 415 million barrels to 296 million barrelsThe report therefore argues that the current calm in energy supply partly depends on temporary measures
• Chinese crude oil importsDown from 12 million barrels per day to approximately 7 million barrels per dayThe chart separately shows 7.2 million barrels per day in June, the lowest since late 2016, with imports beginning to recover in July
• US gasoline prices$4.8 per gallonRebounded to a two-month high
• Probability of Strait of Hormuz shipping recovery19%Prediction-market probability of a return to normal before early December
• Stoxx 600 target610 at end-2026; 580 in early Q2 2027The report expects nearly 10% downside by year-end and more than 10% downside by Q2 next year
• Stoxx 600 sensitivity to PMIApproximately 5% decline in fair value for every three-point decline in PMIEstimate from BofA's macro-driver model
• Stoxx 600 sensitivity to real yieldsApproximately 5% decline in fair value for every 100-basis-point increase in real yieldsEstimate from BofA's macro-driver model
• Forecast for cyclicals relative to defensivesApproximately 6% further downside by early 2027Already underperformed by 6% over the past two months
• Eurozone PMI forecast49 at end-2026The report believes July may mark the peak of the current recovery

Impact & implications

The report argues that if uncertainty over AI commercialization drives risk premia back upward, European equities, cyclicals, and semiconductors and capital goods with high exposure to AI capex will face further pressure. If weaker US employment and inflation lead to falling bond yields, banks, financial services, and the value-oriented structure of European indices will also be constrained. By contrast, food and beverages, telecommunications, pharmaceuticals, quality stocks, and defensive-growth software should provide better hedges against a reversal in AI momentum, while undervalued small caps, German equities, and luxury goods are used to retain limited exposure to cyclical improvement.

Risks

• Intensifying competition at the AI model layer and falling token prices could weaken the pricing power and margins of frontier models, hyperscalers, and the entire AI ecosystem.
• Negative hyperscaler free cash flow and the rise in credit spreads from 55 basis points to 120 basis points could cause investment returns to fall below the rising cost of capital.
• Constraints on power generation, transmission, and data-center approvals could obstruct the sustained expansion in AI capex expected by the market.
• Near-zero US employment growth, weak retail sales, and declining consumer confidence could signal weakening macro momentum.
• The US-Iran conflict and disruption to shipping through the Strait of Hormuz remain unresolved, while high refined-product prices could place further pressure on consumers.
• Market earnings expectations and valuations are excessively optimistic, while global and European equity risk premia are at multi-year lows, increasing the potential correction following disappointment.
• Tight eurozone credit, the fading World Cup effect, and rising natural gas prices could cause PMI to fall to 49 by end-2026.

What to watch

• Monitor whether momentum in European AI capex beneficiaries continues to reverse after their recent 6% relative decline.
• Track the adoption of Chinese open-weight models, token prices, and changes in the Silicon Data Token Expenditure Index.
• Watch hyperscaler free cash flow, bond issuance, credit spreads, and whether investment returns can cover the cost of capital.
• Monitor electricity supply, transmission bottlenecks, and data-center approval policies in regions such as Texas.
• Track three-month US employment growth, retail sales, consumer confidence, and inflation momentum.
• Monitor Strait of Hormuz shipping, the US Strategic Petroleum Reserve, refined-product crack spreads, and US gasoline prices.
• Watch whether eurozone PMI peaks in July and declines toward 49 by year-end.
• Track whether US high-yield credit spreads, the European equity risk premium, and bond yields move as the report expects.