AI Capex Could Exceed Expectations; Market Volatility to Intensify

Overview

This Goldman Sachs research report focuses on artificial intelligence-related sectors in the US equity market. The core conclusion is that consensus expectations for hyperscaler capital expenditure in 2027 are overly conservative, with actual spending potentially significantly exceeding current forecasts, driving earnings and stock prices for AI infrastructure beneficiaries. However, factors such as valuation expansion, crowded positioning, and slow enterprise adoption will exacerbate market volatility. The report supports these views through multi-dimensional analysis including historical cycle comparisons, cash flow analysis, and corporate surveys, urging investors to balance the upside potential of capex growth against potential risks.

Core views

Conservative Capex Expectations: Market consensus forecasts hyperscaler capex at $920 billion for 2027, reflecting 22% year-over-year growth, a significant slowdown from the 84% growth in 2026. Goldman Sachs deems this expectation too conservative. If incremental investment reaches 2-3% of GDP (similar to railroad and automotive construction cycles), 2027 spending could reach $1.1 trillion (up 45%); in an extreme scenario, combining cash flows with credit market capacity, it could even reach $1.4 trillion (up 89%). Valuation and Volatility Risks: The median forward P/E for AI infrastructure stocks has expanded to 26x, the highest since the launch of ChatGPT. Q2 return dispersion for AI-related stocks rose to 53 percentage points, a cyclical high, driven primarily by positive returns in the infrastructure layer versus divergent positive and negative returns in the application layer. Some stocks have recently rallied beyond earnings expectations, making risk-adjusted returns difficult to sustain. Slow Enterprise Adoption Progress: During the Q1 earnings season, 54% of companies mentioned AI related to productivity, but only 11% quantified productivity improvements in specific use cases, and merely 2% quantified the impact on earnings. Companies discussing AI showed no significant difference in profit margins or stock price reactions compared to those that did not, indicating the market has not yet priced in AI productivity gains. Software Sector Valuation Sensitivity: Software sector P/E multiples have declined from a high of 39x last March to the current 25x. DCF models show 85% of their value is concentrated in terminal value, making them highly sensitive to minor changes in long-term growth and margin assumptions. Investor concerns regarding AI disruption uncertainty will continue to drive return dispersion.

Analysis framework

Goldman Sachs structures its analysis around the 'Capex - Valuation - Enterprise Adoption' framework. First, it compares historical technology cycles (e.g., 1990s telecom, railroad construction) to argue there is still upside room for AI investment as a share of GDP. Second, it analyzes hyperscaler cash flows, debt capacity, and equity fundraising capabilities to demonstrate that capital can support higher spending. Third, it combines corporate surveys and earnings commentary to quantify AI adoption progress and earnings impact. Finally, it uses DCF models and valuation multiple comparisons to reveal the sensitivity of the software sector to terminal value. The methodology integrates supply-demand frameworks, valuation modeling, and industry cycle analysis.

Methodology notes

• Industry/Industrial Analysis FrameworkSupply and Demand Framework

  Judge the upside potential for capital expenditure by comparing AI investment as a share of GDP with historical technology cycles (railroads, automobiles, telecommunications)

  The report contrasts the current AI investment impulse (estimated at 1.5% of GDP in 2026) with historical peaks (railroads 3.4%, telecom 1.4%), arguing that if investment intensity rises to 2-3% of GDP, capital expenditure will significantly exceed expectations.

• Valuation MethodDCF Cash Flow Discounting

  Use DCF models to analyze the sensitivity of software sector valuations to terminal value assumptions

  The report notes that 85% of software stock value is concentrated in terminal value; minor downward revisions in long-term growth or margin assumptions can explain valuation contraction, helping readers understand the fragility of high-valuation sectors.

• Cycle and Business Cycle FrameworkInflection Point Analysis

  Assess whether the industry is in an early adoption phase through surveys on AI adoption rates and earnings commentary

  With only 2% of companies quantifying AI's impact on earnings, it shows productivity conversion is still in its early stages, helping readers understand why the market has not yet priced in AI gains.

• Corporate Fundamentals and Financial FrameworkFree cash flow analysis

  Analyze the match between hyperscaler operating cash flow and capital expenditure to assess financing capability

  The report points out that hyperscalers have a net debt/EBITDA ratio of only 0.4x, meaning they can significantly increase debt while maintaining an investment-grade rating, demonstrating financial support for capital expenditure.

Asset mapping & comparison

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

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

  Direct beneficiaries of rising capex; cloud revenue acceleration and backlog growth support earnings

  Strengths

  Strong cash flows; net debt/EBITDA only 0.4x; ample financing capacity

  Weaknesses

  Equity issuance may dilute earnings; risk of capex slowdown

  Comparison

  Valuation expectations are more conservative than semiconductors; less sensitive to capex slowdowns

  Risks

  Weakening demand, equity dilution, rising interest rates increasing financing costs

• Semiconductors

  Core beneficiary layer for AI infrastructure; market cap increment is 14x the change in forward earnings

  Strengths

  Strong demand; high pricing power

  Weaknesses

  Significant valuation expansion; high volatility

  Comparison

  Higher valuation multiples than hyperscalers; greater return dispersion

  Risks

  Supply-demand balance delayed until second half of 2027; risk of cyclicality returning

• Software

  Coexistence of AI empowerment and disruption; significant return dispersion

  Strengths

  Data infrastructure stocks favored; some companies showing AI-driven revenue growth

  Weaknesses

  85% of value concentrated in terminal value; sensitive to assumptions

  Comparison

  Sector-specific software and services stocks avoided; valuations dropped from 36x to 25x

  Risks

  Low-cost competition suppressing earnings; downward revision of terminal value assumptions

Key data

• 2026 Hyperscaler Capex Consensus Estimate$757 billionYear-over-year growth of 84%
• 2027 Hyperscaler Capex Consensus Estimate$920 billionYear-over-year growth of 22%; viewed as conservative by Goldman Sachs
• Goldman Sachs 2027 Capex Scenario Forecast$1.1 trillion to $1.4 trillionCorresponds to scenarios where GDP share is 2-3% or based on cash flow/credit capacity
• Median P/E for AI Infrastructure Stocks26xHighest since ChatGPT launch
• Q2 AI Stock Return Dispersion53 percentage pointsHighest since ChatGPT launch
• Percentage of Companies Quantifying AI Impact on Earnings2%Data from Q1 earnings season
• Proportion of Terminal Value to Present Value for Software Stocks85%Estimated using DCF model

Impact & implications

The report posits that upward momentum in AI capital expenditure will benefit infrastructure beneficiaries (such as semiconductors, power, and optical networks) in the short term, but valuation expansion and crowded positioning will increase volatility risks. Medium-term, investors must balance slowing expenditure growth with uncertainties regarding earnings sustainability. In the software sector, due to AI disruption risks and terminal value sensitivity, return dispersion will persist. As hyperscalers increase equity fundraising activities, investors will focus more on return on invested capital. Macro-wise, if tensions between the US and Iran ease, market breadth may expand; however, an economic recession would drag down the overall equity market.

Risks

• Slowing capex growth or weakening demand
• Enterprise AI adoption lagging expectations; slow productivity conversion
• Earnings dilution from hyperscaler equity fundraising
• Physical capacity constraints (data centers, power, labor)
• Rising interest rates increasing financing costs
• Downward revision of terminal value assumptions for the software sector

What to watch

• Evidence of accelerating AI-related revenue
• Whether capex growth is slowing
• Macroeconomic shift from acceleration to deceleration
• Cloud business backlogs and earnings sustainability
• Changes in the proportion of companies quantifying AI productivity impact