U.S. AI Investment Is Approaching $600 Billion, but Its Net Boost to GDP Is Lower Than Market Intuition Suggests

Overview

The report evaluates whether U.S. AI investment weakens economic growth by replacing other technology spending, crowding out other construction activity, and pushing up financing costs for non-AI companies. The study finds that although AI spending has become very substantial, hyperscale cloud companies’ strong cash flow and financing capacity, reallocation among enterprise services spending, and the coincidence in timing between data center construction and the pullback in subsidized manufacturing construction have kept crowding-out effects at the national level relatively limited so far. At the same time, because a large amount of equipment depends on imports, and because some semiconductor investment and technology service exports are not fully captured in official GDP statistics, there is a significant gap between total AI spending and its contribution to official GDP.

Core views

First, U.S. AI investment is expected to approach $600 billion in 2026, but its officially measured direct GDP impact is only about 0.5% of GDP, with a direct boost to that year’s GDP growth of about 0.1 percentage point. Second, after correcting for statistical omissions in semiconductor investment and chip design service exports, the direct boost to real GDP growth is about 0.3 percentage point. Third, the three crowding-out channels together add up to about $50 billion of new crowding out, including about $30 billion from substitution in technology and services spending, slightly more than $10 billion from substitution in construction activity, and about $10 billion of reduced investment caused by higher interest rates. Fourth, current credit spreads for non-AI companies, nationwide construction wages, and material shortages do not indicate severe crowding out, but clear local competition for resources has already emerged in areas with concentrated data centers. Fifth, after accounting for stock market wealth effects, erosion of household real income from rising electricity prices and other prices, and investment crowding out, indirect effects will reduce the contribution to 2026 GDP growth by about 0.1 percentage point.

Analysis framework

The report first estimates total spending on AI-related equipment, software, intellectual property, power infrastructure, and data center construction, then distinguishes imports, final expenditure, intermediate inputs, and potentially omitted items according to national accounting rules. It then estimates crowding-out effects through three channels—reallocation of technology and services budgets, competition for construction resources, and AI-related debt issuance—and cross-checks the results using the GS IT Spending Survey, state-level construction activity and wage data, credit spreads, and duration supply rules of thumb from quantitative tightening research. Finally, it combines the direct investment contribution with stock market wealth effects, price shocks, and crowding-out effects to assess the overall impact on 2026 GDP growth.

Methodology notes

• Macroeconomic transmission analysisThree-channel crowding-out framework

  Measures the substitution of AI investment for existing economic activity through three paths: other technology spending, other construction activity, and financing costs.

  This framework only includes activities that would originally have been treated as final expenditure in GDP statistics, avoiding misclassifying all substitution among different intermediate business services as GDP losses.

• National income accountingExpenditure-side GDP decomposition

  Breaks down total AI spending into imports, domestic final investment, intermediate inputs, and potentially underestimated investment and exports.

  AI spending approaching 2% of GDP does not mean it contributes equally to GDP, because imports are deducted from GDP, some services are intermediate inputs, and some semiconductor investment and technology service exports may not be fully counted.

• Survey researchGS IT Spending Survey budget reallocation analysis

  Uses the share of enterprise AI inference costs in IT budgets and their funding sources to determine whether AI service spending replaces other final expenditure.

  The survey shows that for 89% of respondents, AI inference costs account for only 1% to 5% of IT budgets, while another 10% report 5% to 10%; about one-third of the costs come from new budgets, and about two-thirds come from cuts to other spending.

• Regional resource constraint analysisState-level construction crowding-out test

  Compares the share of data center construction, changes in other nonresidential construction, construction wages, and material shortages.

  In some states, data center construction already accounts for about half or more of nonresidential construction spending, indicating local crowding out, but national statistical relationships, wages, and material data do not yet show broad resource shortages.

• Interest rate sensitivity analysisDuration supply rule of thumb

  Draws on quantitative tightening research by treating AI-related corporate bond issuance as an increase in duration supply to the market, thereby estimating its impact on interest rates and investment.

  The report starts from the assumption that Treasury and mortgage-backed securities supply equivalent to 1% of GDP raises the 10-year yield by about 2 basis points, then adjusts this to 3 basis points after accounting for corporate bond credit risk, estimating that AI-related issuance raises borrowing rates by about 5 basis points.

• Comprehensive impact assessmentCombining direct and indirect effects

  Combines the direct investment contribution with stock market wealth effects, household real income losses, price pressures, and investment crowding out.

  This method is used to distinguish AI’s contribution to officially measured GDP, corrected real GDP, and overall growth including macro spillover effects.

Asset mapping & comparison

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

• U.S. real GDP

  AI investment provides a moderate positive growth contribution, but there is a significant gap between total spending and domestic value added.

  Strengths

  Capital expenditure expansion, related intellectual property investment, and technology service exports support growth.

  Weaknesses

  A large amount of equipment depends on imports, and some AI services are intermediate inputs.

  Comparison

  AI spending is close to 2% of GDP, but its officially measured direct GDP scale impact is about 0.5%.

  Risks

  Crowding out other investment, raising electricity and other prices, and statistical classification bias.

• Hyperscale cloud computing companies

  They are the main bearers and financing entities for AI capital expenditure.

  Strengths

  Abundant cash flow allows them to support new capital expenditure through reduced share repurchases and debt financing.

  Weaknesses

  Capital expenditure is expected to exceed cash flow, and reliance on external financing may rise.

  Comparison

  Compared with ordinary companies, they have stronger financing capacity and therefore currently face less pressure to cut other spending.

  Risks

  Rising leverage, pressure on free cash flow, reduced share repurchases, and investment returns falling short of expectations.

• Semiconductors and AI equipment

  They benefit from demand for training, inference, and data center equipment, and are the direct assets carrying AI investment growth.

  Strengths

  AI spending already accounts for about 15% of equipment investment, implying a large demand scale.

  Weaknesses

  A considerable portion of equipment comes from imports, lowering the direct multiplier for U.S. GDP.

  Comparison

  The industry’s revenue contribution may be significantly higher than its contribution to U.S. domestic value added.

  Risks

  Reversal of the capital expenditure cycle, import dependence, statistical misclassification, and demand falling short of expectations.

• Data centers and nonresidential real estate

  Data center construction drives nonresidential construction investment while crowding out other engineering resources in some regions.

  Strengths

  Demand for data center construction is strong, and related projects have higher gross margins than non-technology engineering projects.

  Weaknesses

  Nationwide construction wages and material demand have not yet seen a synchronized sharp expansion.

  Comparison

  Data center spending already accounts for 9% of private nonresidential construction spending, and in some states it reaches about half or more.

  Risks

  More evident constraints on wages, materials, power capacity, and permitting could emerge over the next one to two years.

• U.S. investment-grade credit

  AI and data center financing increase duration and credit supply in the bond market.

  Strengths

  Credit spreads for non-AI companies remain near historical lows, indicating strong market absorption capacity.

  Weaknesses

  AI-related financing already accounts for close to one-quarter of total investment-grade bond issuance.

  Comparison

  The report estimates that the additional supply raises overall borrowing rates by about 5 basis points.

  Risks

  Continued acceleration in issuance could push yields higher, widen credit spreads, and suppress investment by other companies.

• U.S. equity market and household consumption

  Rising AI-related stocks support consumption through wealth effects, but price increases erode household real income.

  Strengths

  Growth in AI companies’ market capitalization can increase household wealth and boost consumption.

  Weaknesses

  Electricity and other price increases weaken households’ real purchasing power.

  Comparison

  Positive wealth effects and negative price effects offset each other, leaving a limited overall indirect contribution.

  Risks

  A pullback in AI valuations could reverse wealth effects, while energy cost pressures may persist.

Key data

• U.S. AI investment in 2026Approaching $600 billionEquivalent to nearly 2% of U.S. GDP.
• AI spending as a share of business investmentMore than 10% of business fixed investment; about 15% of equipment investmentEstimated recent share of AI spending in the corresponding total investment categories.
• Officially measured direct GDP scale impactAbout 0.5% of GDPFar below the nearly 2% share of AI total spending in GDP, mainly because of imports and differences in national accounting classifications.
• Contribution to officially measured GDP growth in 2026About 0.1 percentage pointRefers to the direct boost from AI investment to officially measured GDP growth.
• Real growth contribution after correcting statistical omissionsAbout 0.3 percentage pointIncludes potentially misclassified semiconductor investment and insufficiently counted chip design service exports.
• Indirect effect adjustmentReduces by about 0.1 percentage pointReflects the combined impact of wealth effects, price pressures, and investment crowding out on the contribution to 2026 GDP growth.
• New crowding out across three channelsAbout $50 billionOnly activities that would originally have been counted as final expenditure in GDP are included.
• Crowding out of technology and services spendingAbout $30 billionIncludes substitution in hyperscale cloud companies’ capital expenditure and reallocation of enterprise AI services budgets.
• Crowding out of construction activitySlightly more than $10 billionThe report assumes that one-quarter of the roughly $45 billion increase in AI-related construction over the past year crowded out other construction activity.
• Interest rate impact of AI-related financingAbout 5 basis pointsExpected to reduce corporate investment by about $10 billion by raising the cost of capital.
• Share of data center construction9% of private nonresidential construction spendingThis share has risen rapidly since 2024, and in some states it has reached about half or more.
• AI-related investment-grade bond financingClose to one-quarter of total investment-grade bond issuanceCredit spreads for non-AI companies remain near historical lows, indicating that financing spillovers are currently limited.
• Sources of AI services budgetsAbout one-third from new spending and about two-thirds from cuts to other spendingMost substituted items are intermediate business services, so the net impact on GDP statistics is relatively small.

Impact & implications

For investors, AI capital expenditure can still support demand for semiconductors, data centers, power infrastructure, and related engineering, but total spending should not be equated directly with its contribution to U.S. economic growth. Hyperscale cloud companies can rely on cash flow, reduced share repurchases, and increased debt financing to delay cuts to other capital expenditure; this is favorable for revenue in the AI value chain, but may reduce shareholder returns and increase credit supply. Nationwide crowding out from data center construction is currently limited, but constraints on local construction resources, electricians, transmission and distribution equipment, and power capacity may continue to intensify. At the macro level, AI’s positive effect on production and investment will be partially offset by import leakage, financing costs, electricity prices, and pressure on household real income.

Risks

• If data center construction continues to grow rapidly as forecast while subsidized manufacturing construction has completed normalization, nationwide crowding out of construction resources could intensify significantly over the next one to two years.
• Further expansion of AI-related bond issuance could push up market interest rates and credit spreads, suppressing investment by non-AI companies.
• After hyperscale cloud companies’ capital expenditure exceeds cash flow, they may increase leverage, reduce share repurchases, or begin cutting other spending.
• Shortages of electricity, transmission and distribution capacity, and construction labor could push up costs and erode household real income and consumption through price increases.
• The high import share of AI equipment means growth in industry spending cannot translate proportionally into U.S. GDP.
• Insufficient classification or measurement of semiconductor investment and technology service exports in GDP statistics creates a gap between official data and the true economic contribution.
• The report’s estimated crowding-out scale depends on empirical assumptions such as budget substitution ratios, construction crowding-out ratios, and interest rate sensitivity, and actual outcomes may deviate from the estimates.
• If valuations of AI-related stocks decline, the current wealth effect supporting household consumption may weaken or turn negative.

What to watch

• Changes in hyperscale cloud companies’ capital expenditure, free cash flow, share repurchases, and net debt.
• The share of AI services in enterprise IT budgets, and the ratio of new budgets to cuts in other spending such as software.
• Whether data center construction spending can reach forecast levels and its share of private nonresidential construction.
• Construction wages, electrician wages, material shortages, and starts of other nonresidential projects in states with concentrated data centers.
• Whether the pullback in construction of manufacturing facilities supported by the Inflation Reduction Act and CHIPS Act has fully ended.
• The share of AI-related investment-grade bond issuance, the 10-year U.S. Treasury yield, and credit spreads for non-AI companies.
• The impact of electricity prices, grid access, power generation, and transmission and distribution investment on household real income and corporate costs.
• Statistical revisions in official GDP data for semiconductor investment, AI services, and chip design service exports.
• Whether wealth effects from AI-related stocks can continue to offset financing costs and price pressures.
• Whether U.S. real GDP and business fixed investment in 2026 are close to the report’s forecasts of 2.1% and 6.9%.