Goldman Sachs: 2027 AI Capital Expenditure Could Reach $1.1 Trillion, Volatility Intensifies

Overview

This report explores the upside potential of AI capital expenditure and its impact on market volatility. Goldman Sachs believes the consensus expectation for 2027 hyperscaler capital expenditure is overly conservative; actual spending could far exceed current forecasts, continuing to support AI infrastructure stock earnings and share prices in the short term. However, as valuations rise to high levels, trading positions become crowded, and enterprise AI productivity monetization remains in the early stages, return divergence within the AI sector and market volatility will significantly intensify. The report also points out that although infrastructure construction is bustling, commercial verification at the application layer still requires time, and investors need to find a balance between strong capital expenditure and potential return uncertainty.

Core views

Capital expenditure expectations have significant upward revision potential. Current analyst consensus expectation for 2027 hyperscaler capital expenditure is $920 billion (YoY +22%), implying a significantly slowed growth rate. Goldman Sachs estimates that if AI incremental investment reaches 2%-3% of GDP (similar to peak levels of historical railway and automotive infrastructure), 2027 capital expenditure could reach $1.1 trillion to $1.25 trillion (YoY +45%-65%). In an extremely optimistic scenario, considering cash flow generation capacity and investment-grade credit market capacity, spending could even touch $1.4 trillion. Evidence supporting this includes: combined Google Cloud and AWS order backlogs surged from $358 billion to $832 billion in six months; Google explicitly stated 2027 capital expenditure will 'increase significantly' and raised funding via equity financing; token consumption volume expected to grow 24x by 2030. Infrastructure stocks face dual challenges of valuation and volatility. Although the AI infrastructure portfolio rose 40% since Q2, and most gains were driven by earnings growth, valuation expansion in some sub-segments has outpaced earnings revisions. Median forward P/E of GS TMT Data Center basket has risen to 26x, highest since ChatGPT launch. Meanwhile, AI-related stock return dispersion in Q2 surged to 53 percentage points, creating a historic high. This divergence mainly manifests as positive return differences within infrastructure, and violent contrast between positive/negative returns in application layers (e.g., software). Popular sectors like optical modules have market cap increment already 41x forward earnings increment, while memory chips and hyperscaler expectations remain relatively conservative. Enterprise AI applications remain in budding phase, monetization capability awaiting verification. Despite 54% of S&P 500 companies mentioning AI productivity in earnings conference calls, only 11% quantified specific use case benefits, and only 2% quantified impact on earnings. Companies discussing AI compared to those not discussing AI showed no significant difference in margin expectations or stock price reaction. Currently, AI-generated benefits are concentrated in cost savings (e.g., Verizon saved $200 million in energy costs, IBM cumulatively saved $4.5 billion in operating costs), rather than revenue growth. Additionally, token usage costs have become a new focus, potentially limiting profit elasticity brought by productivity improvements. For the software industry, since 85% of value relies on terminal value, market concerns over AI disruption risks lead to extreme sensitivity of valuations to long-term growth assumptions, with performance diverging continuously throughout the year.

Analysis framework

The report adopted a dual framework of 'historical analogy + macro constraints' to forecast AI capital expenditure. Unlike pure bottom-up analyst aggregation, the institution benchmarks current AI investment as % of GDP against peak investment levels in historical major technology cycles such as railways, electricity, and telecommunications, thereby judging the theoretical ceiling of capital expenditure. At the same time, combining credit market capacity (investment-grade bond index weight upper limit) and corporate balance sheets (net leverage ratio <1x) to assess financing feasibility, thus constructing three capital expenditure scenarios: baseline, optimistic, and extreme. In evaluating market risks, the report introduced the 'Market Cap Increment / Earnings Increment' ratio as a proxy indicator for measuring market expectation durability. By comparing market cap changes and forward earnings changes of different AI sub-sectors (e.g., semiconductors, optical modules, cloud vendors) since 2023, identifying which sectors priced in excessive long-term growth expectations. Additionally, through structured statistics of earnings call texts (percentage of companies mentioning AI, quantifying use cases, quantifying earnings), quantitatively tracking the actual progress of enterprise-side AI implementation to verify the final demand support for infrastructure investment.

Methodology notes

• Macroeconomic framework

  Historical Analogy of Investment as % of GDP

  Compare the proportion of current AI incremental investment to GDP (approx. 1.5%) with ratios during peak periods of common technology infrastructure like railways (3.4%) and electricity (2.2%) to anchor the potential peak range of this round of AI capital expenditure, avoiding linear extrapolation bias.

• Valuation MethodDiscounted Cash Flow (DCF)

  Terminal Value Sensitivity Analysis

  Report notes software industry 85% of value comes from terminal value, and via DCF model shows: only slight adjustments to long-term growth rate or margin assumptions can explain the violent volatility in valuations this year. This suggests investors in industries with high terminal value proportion should focus more on long-term competitive landscape rather than short-term performance.

• Quant / factor / portfolio theory

  Market Cap Increment / Earnings Increment Ratio

  Using 'Market Cap Change Since 2023 divided by Forward Earnings Change' as a proxy indicator to measure the degree of market pricing on earnings sustainability. Higher ratio (e.g., optical modules reach 41x) indicates the market incorporates more remote growth expectations and is more sensitive to future disproof.

• Industry/Industrial Analysis FrameworkPenetration Rate S-Curve

  Enterprise AI Adoption Rate Tracking

  Track enterprise AI adoption rate (currently approx. 19.5%) and quantified benefit ratio via statistical survey and earnings text mining. The indicator is in the early stage of the S-curve, meaning infrastructure-first construction is reasonable, but application layer explosion still awaits crossing the penetration rate critical point.

Asset mapping & comparison

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

• Hyperscalers (AMZN, GOOGL, META, MSFT, ORCL)

  Direct decision-makers and funders of AI capital expenditure, benefit from cloud revenue acceleration and order backlogs

  Strengths

  Strong balance sheet (Net leverage 0.4x), cloud business margin exceeds expectations, order backlogs provide revenue visibility

  Weaknesses

  CapEx growth hasn't slowed yet, equity financing dilution risk, market ROI requirements increasingly strict

  Comparison

  Compared to semiconductor stocks, Market Cap/Profit Increment ratio is lower (11x vs 14x), relatively higher valuation safety margin

  Risks

  If token demand falls short or productivity gains fail to cover costs, might be forced to cut capital expenditure

• AI Infrastructure Hardware Stocks (Semiconductors/Optical Modules/Liquid Cooling/Power)

  Direct beneficiaries of capital expenditure expansion, performance highly positively correlated with cloud vendor CapEx

  Strengths

  Strong earnings growth, supply bottlenecks confer pricing power, YTD gains lead

  Weaknesses

  Valuation expansion too fast (optical module market cap increment reaches 41x earnings increment), crowded positions, momentum factor volatile

  Comparison

  Optical modules, semiconductor valuation premiums significantly higher than cloud vendors and memory chips

  Risks

  Bottleneck relief or efficiency improvement may lead to loss of pricing power; if CapEx growth slows, hit first

• Software Industry (IGV)

  Carrier of AI applications, facing dual narrative game of AI empowerment vs AI disruption

  Strengths

  Some data infrastructure companies have shown AI-driven revenue acceleration; valuation corrected from 39x to 25x releasing some risk

  Weaknesses

  85% value depends on terminal value, sensitive to long-term growth assumptions; industry-specific software and services stocks performing weakly

  Comparison

  Data infrastructure stocks outperform industry application software stocks; private AI lab valuation rises pose competitive pressure to listed software stocks

  Risks

  Low-cost AI competition may erode incumbent revenue growth and margin expectations; enterprise adoption speed lower than expected

Key data

• 2027 Hyperscaler Capital Expenditure (Consensus vs Goldman Sachs Benchmark)$920 billion vs $1.1 trillionConsensus expected YoY growth 22%, Goldman Sachs benchmark scenario (investment 2% of GDP) expected growth 45%
• Median Forward P/E of AI Infrastructure Stocks26xHighest level since ChatGPT launch, YTD valuation expansion faster than earnings revision
• Q2 AI Stock Return Standard Deviation53 percentage pointsRecord high, showing extreme divergence within sector
• Cloud Vendor Order Backlogs (Google+AWS)$832 billionDoubled over the previous six months from $358 billion, supply-demand imbalance continues
• Proportion of Enterprises Quantifying AI Earnings Impact2%Only 2% of S&P 500 companies quantified AI contribution to earnings in reports, slight quarter-over-quarter increase
• Terminal Value Proportion in Software Industry85%Estimated based on DCF model, showing valuation extremely sensitive to long-term assumptions

Impact & implications

For AI infrastructure suppliers, capital expenditure upside implies higher revenue visibility for the next 1-2 years, especially for segments with supply bottlenecks (e.g., storage, power equipment). However for secondary market investors, high valuations and crowded trades mean simple 'beta' rally might end, making stock selection harder, needing vigilance for momentum factor reversal risk. For software and application layer companies, the market is shifting from 'concept hype' to 'performance verification', only truly enterprises that can achieve cost reduction/increase efficiency or create new revenue streams through AI can earn premium, others may face long-term valuation suppression. Macro level, if US-Iran situation eases or economic recession expectations intensify, it could trigger market style rotation from AI tech stocks to other sectors.

Risks

• Enterprise AI Demand Weakening: If productivity gains cannot justify token costs, may lead to capital expenditure cuts
• Valuation and Positioning Risk: AI infrastructure stock valuations at high levels and positions crowded, momentum factor reversal may amplify drawdowns
• Supply Bottleneck Relief: Storage, power equipment segments if capacity release or efficiency improves, may weaken pricing power and earnings elasticity
• Equity Financing Dilution: Hyperscalers issuing stock to finance CapEx, may put pressure on EPS and share price
• Macro and Geopolitical Risk: US-Iran war resolution or severe economic recession may trigger market style rotation, weakening AI sector leadership
• AI Disruption Risk: Low-cost AI competition may compress traditional software company long-term growth and margin expectations

What to watch

• Hyperscaler 2027 capital expenditure guidance and equity/debt financing moves
• Change in proportion of enterprise reports quantifying AI productivity benefits and token cost disclosure
• Earnings revision speed and valuation matching of AI infrastructure sub-sectors
• Cloud revenue growth speed, order backlog scale and supply-demand balance timing
• Differentiation trend of revenue growth between AI-native apps and traditional software in software industry
• Impact of macro economic growth expectations and Federal Reserve policy path on tech stock valuations