AI investment is approaching 2% of U.S. GDP, but nationwide crowding-out effects remain relatively moderate for now

Overview

The report assesses whether U.S. AI investment is significantly crowding out other economic activity. Goldman Sachs believes that although AI investment in 2026 is expected to approach $600 billion and has become an important component of corporate capital expenditure, the nationwide crowding-out effect remains relatively limited for now. The main reasons are that hyperscale cloud service providers have previously been able to finance spending by reducing share repurchases and increasing borrowing, corporate purchases of AI services mainly replace other intermediate services, and the data center construction boom has coincided with a decline in subsidized manufacturing plant construction. At the same time, a large amount of AI equipment comes from imports, and some semiconductor and service activities are not fully captured in GDP statistics, so there are significant differences among total AI spending, its officially measured contribution, and its true economic contribution.

Core views

The report identifies three crowding-out channels. First, hyperscale cloud service providers and AI service buyers may cut other technology and software spending, with incremental crowding out in 2026 expected at about $30 billion. Second, data center construction will compete for construction resources in some states, but nationwide material shortages and construction wage pressures remain limited; the estimated crowding out of other construction activity is slightly above $10 billion. Third, increased AI-related debt issuance has expanded the supply of market duration and credit risk, which is estimated to raise borrowing rates by about 5 basis points and reduce corporate investment by about $10 billion. The three items together crowd out about $50 billion of final spending. AI investment still makes a positive contribution to economic growth, but both its growth contribution and the degree of crowding out are lower than some market commentary has claimed.

Analysis framework

The report first estimates the overall scale of AI-related equipment, data center, power infrastructure, software, and service spending, then distinguishes imports, final spending, intermediate inputs, and statistically undercounted items according to national accounts rules. It then quantifies the three crowding-out channels separately through evidence such as corporate capital allocation, information technology spending surveys, state-level construction activity, construction wages, construction material supply, investment-grade bond issuance, and credit spreads. Finally, it combines the direct investment contribution with indirect effects such as stock market wealth effects, energy and other price increases, and corporate investment crowding out, comparing the difference between officially measured GDP and real economic activity.

Methodology notes

• Macro transmission frameworkThree-channel crowding-out effect framework

  Technology spending, construction activity, and financing costs

  Separately assesses whether AI investment substitutes for other technology spending, competes for construction resources, and raises financing costs for non-AI companies through debt issuance.

• National economic accountingExpenditure-side GDP decomposition

  Imports, final spending, and intermediate inputs

  Breaks total AI spending into domestic final demand, imports, and intermediate inputs to explain why a spending scale of nearly 2% of GDP corresponds to only about a 0.5% direct impact on officially measured GDP.

• Scenario estimationRule of thumb for financing supply shocks

  Bond supply, interest rates, and the cost of capital

  References empirical estimates of the impact of quantitative tightening on long-term rates and adjusts for corporate bond credit risk, deriving an approximately 5-basis-point impact on borrowing rates and about a $10 billion decline in investment from the scale of AI-related issuance.

• Comprehensive impact assessmentCombining direct and indirect effects

  Investment contribution, wealth effects, price shocks, and crowding out

  Adds AI-related stock wealth effects, the impact of rising energy and other prices on households’ real income, and corporate spending crowding out to the direct investment contribution to assess a more complete growth impact.

Asset mapping & comparison

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

• U.S. real GDP

  AI investment directly supports growth through business fixed investment, but imports, statistical undercounting, and crowding out of other spending weaken the officially measured contribution.

  Strengths

  AI capital expenditure is large in scale and drives investment in data centers, power infrastructure, equipment, and intellectual property.

  Weaknesses

  A large amount of technology equipment depends on imports, and some semiconductor and service activities are not correctly counted as final investment.

  Comparison

  Total AI spending approaches 2% of GDP, but the officially measured direct scale impact is only about 0.5% of GDP, and the direct contribution to 2026 growth is about 0.1 percentage point.

  Risks

  If crowding-out effects, energy price shocks, or rising financing costs intensify, the net growth contribution may decline further.

• Semiconductor and AI infrastructure supply chain

  The expansion of AI investment directly increases demand for chips, servers, data centers, power facilities, and supporting services.

  Strengths

  Demand is growing rapidly, AI spending already accounts for about 15% of equipment investment, and long-term infrastructure construction demand remains strong.

  Weaknesses

  Part of demand is met by imports, so the boost to U.S. domestic output is weaker than the nominal spending scale.

  Comparison

  Compared with general corporate equipment investment, AI-related equipment is an important current source of capital expenditure growth, but its domestic value-added share is relatively low.

  Risks

  Capital expenditure returns falling short of expectations, financing constraints, misjudgment of statistical treatment, and supply-chain and power bottlenecks.

• Data centers and engineering construction

  Data center spending has become an important component of private nonresidential construction and is crowding out other projects in some states.

  Strengths

  Data center projects have relatively high profit margins and create strong demand for construction, electrical engineering, and power infrastructure.

  Weaknesses

  Nationwide signals of construction wage pressure and material shortages are currently limited, and part of the incremental demand has been offset by the decline in manufacturing plant construction.

  Comparison

  Data centers account for about 9% of private nonresidential construction spending; in a small number of states, their share of nonresidential construction spending has reached about half or higher.

  Risks

  If construction scale continues to rise over the next one to two years, it may lead to shortages of construction resources, wage increases, project delays, and stronger nationwide crowding out.

• Hyperscale cloud service provider stocks

  Related companies finance AI capital expenditure by reducing share repurchases and increasing borrowing.

  Strengths

  They have a strong cash flow base and currently have a relatively high tolerance for higher interest rates, allowing continued support for AI investment.

  Weaknesses

  Capital expenditure may exceed cash flow, while reduced buybacks and higher leverage will weaken short-term shareholder returns.

  Comparison

  Compared with cutting core operating spending, hyperscale cloud service providers have previously absorbed AI investment more through adjustments to buybacks and financing structures.

  Risks

  Uncertain returns on AI investment, debt growth, pressure on free cash flow, and valuations being overly sensitive to long-term growth expectations.

• U.S. investment-grade corporate bonds

  AI-related financing has approached one-quarter of total investment-grade bond issuance, increasing market bond supply.

  Strengths

  Credit spreads for non-AI companies remain close to historical lows, indicating limited current financing spillovers.

  Weaknesses

  A greater supply of duration and credit risk may modestly lift market-wide borrowing costs.

  Comparison

  The report estimates that AI-related issuance raises borrowing rates by only about 5 basis points, an impact smaller than the issuance share would suggest.

  Risks

  Continued acceleration in issuance, benchmark rates remaining high, or repricing of credit risk could amplify the crowding out of investment by non-AI companies.

Key data

• U.S. AI investment in 2026Approaching $600 billionEquivalent to nearly 2% of U.S. GDP.
• AI spending as a share of business fixed investmentMore than 10%This is the recent quarterly level; it accounts for about 15% of equipment investment.
• Direct scale impact of AI on officially measured GDPAbout 0.5% of GDPSignificantly below total AI spending at nearly 2% of GDP, mainly due to imports and statistical classification differences.
• Incremental crowding out through the technology and services spending channelAbout $30 billionIncludes substitution of hyperscale cloud service provider capital expenditure and substitution of AI services for other final spending.
• Incremental crowding out through the construction channelSlightly above $10 billionAssumes one-quarter of the roughly $45 billion increase in AI-related construction over the past year crowds out other construction activity.
• Interest-rate impact of AI-related debt issuanceAbout 5 basis pointsThis is expected to reduce other corporate investment by about $10 billion.
• Total crowding out from the three channelsAbout $50 billionOnly counts other activities regarded as final spending in the national accounts.
• Direct contribution of AI to official GDP growth in 2026About 0.1 percentage pointA large amount of equipment comes from imports, and some AI activity is not fully captured in official GDP statistics.
• Direct contribution of AI to real GDP growthAbout 0.3 percentage pointIncludes investment in training semiconductors misclassified as intermediate inputs and statistically undercounted exports of chip design services.
• Impact of indirect effects on the 2026 growth contributionReduced by about 0.1 percentage pointAn estimate after combining stock wealth effects, price increases, and corporate investment crowding out.
• Data centers as a share of private nonresidential construction spendingAbout 9%Has risen rapidly since 2024, but the simultaneous decline in subsidized manufacturing plant construction has provided a partial offset in resource demand.
• AI-related financing as a share of total investment-grade bond issuanceApproaching one-quarterCredit spreads for non-AI companies remain close to historical lows, indicating limited current spillover effects.

Impact & implications

At the macro level, AI capital expenditure remains an important growth driver for U.S. business investment, but the spending scale of nearly 2% of GDP cannot be directly equated with a contribution to U.S. GDP, because imports and national accounts classifications significantly weaken the officially measured effect. At the industry level, semiconductors, data centers, power infrastructure, and related construction services continue to benefit, but high-margin data center projects have already drawn construction resources away in some states. At the capital markets level, hyperscale cloud service providers are supporting capital expenditure by reducing buybacks and increasing borrowing, which may alter shareholder returns and leverage structures; AI-related credit financing is expanding rapidly, but the spillover to financing costs for non-AI companies remains small for now. Over the next one to two years, if data center construction continues to grow as forecast while manufacturing plant construction no longer provides an offset to resource demand, nationwide construction crowding out and cost pressures could increase significantly.

Risks

• If data center construction continues to expand as forecast, it may create more pronounced nationwide crowding out of construction resources over the next one to two years.
• Further growth in AI-related debt issuance could raise financing costs for non-AI companies and suppress broader corporate investment.
• Increases in electricity and other input prices may erode households’ real income and consumption, offsetting the wealth effects brought by AI.
• AI investment relies heavily on imported equipment, which may prevent nominal capital expenditure growth from translating into an equivalent scale of U.S. domestic output.
• After hyperscale cloud service provider capital expenditure exceeds cash flow, they may face pressure from rising leverage, reduced buybacks, or adjustments to other spending.
• Official statistical classification or undercounting of semiconductor investment and chip design service exports may lead to misjudgment of AI’s true economic contribution.
• Although enterprise AI service costs currently account for a low share of information technology budgets, a rapid increase could intensify substitution away from software and other service spending.

What to watch

• Whether AI capital expenditure in 2026 reaches the forecast scale of nearly $600 billion.
• Changes among hyperscale cloud service provider capital expenditure, free cash flow, debt financing, and share repurchases.
• The share of enterprise AI inference costs in information technology budgets, and the funding-source structure between incremental budgets and cuts to other spending.
• The share of data center construction in private nonresidential construction spending and state-level concentration.
• Wages of construction workers and electricians, construction material supply, and profit margins of data center projects.
• Whether the decline in subsidized manufacturing plant construction can still offset data center demand for construction resources.
• The share of AI-related investment-grade bond issuance, credit spreads for non-AI companies, and long-term borrowing rates.
• Spillover effects of AI demand on electricity prices, households’ real income, and consumer spending.
• Adjustments to official GDP accounting for training semiconductor investment and chip design service exports.