Goldman Sachs estimates global AI investment at approximately $1 trillion in 2026, with the United States slightly below $600 billion

Overview

This Goldman Sachs Global Economics Analyst report focuses on the challenges of measuring the pace of AI investment. The market frequently cites nearly $800bn in projected 2026 capital expenditures by U.S. hyperscalers, but this measure omits investment by private and overseas companies while also including non-AI and non-U.S. investment. By removing the pre-AI-boom capital expenditure baseline, adding spending by private and overseas AI-related companies, and allocating global capital expenditures according to project location, the report estimates 2026 global AI investment at approximately $1019bn and U.S. investment at approximately $581bn.

Core views

The core conclusion is that the commonly used U.S. hyperscaler capital expenditure measure understates global AI capital expenditures by approximately $200bn while overstating domestic U.S. AI investment by approximately $200bn. Multiple methods jointly point to approximately $1 trillion in global AI investment and slightly below $600bn in U.S. investment in 2026. Through 2028, AI capital expenditures as a share of GDP may continue to rise, but overall investment would remain comparable to historical general-purpose technology buildout cycles. Short-term leading indicators show continued strength, although trade-data nowcasts suggest moderate deceleration in June-July. At the same time, rising cost inflation and high hardware import content limit the contribution to the level of real U.S. GDP.

Analysis framework

The report uses a three-layer validation framework. First, it expands and adjusts hyperscaler capital expenditure data by adding U.S. non-hyperscalers, key private AI companies, and overseas AI-related companies, while deducting pre-2022 baseline capital expenditures. Second, it uses gross-margin realization and forecast revisions for listed AI supply-chain companies relative to their 2022 forecasts as a proxy for incremental end-demand. Third, it uses the U.S. national accounts commodity-flow method, AI hardware imports and exports, inventory changes, and global trade data to nowcast and extrapolate AI investment in the United States and overseas.

Methodology notes

• capex_adjustmentExpanded hyperscaler capital expenditure method

  Starting with hyperscaler capital expenditures, add private and overseas AI-related companies and deduct the pre-2022 non-AI baseline.

  This method attempts to correct four biases in the commonly used $800bn measure: omitting private companies, omitting non-U.S. companies, including pre-AI-boom existing capital expenditures, and treating overseas investment by U.S. companies as domestic U.S. investment.

• cross_checkGross-margin forecast revision method

  Estimate incremental AI end-demand using gross-margin realization and forecast revisions for listed AI-related companies relative to their 2022 forecasts.

  Compared with revenue-based measures, the gross-margin approach better reduces the risk of double counting internal transactions within the AI supply chain.

• macro_nowcastNational accounts and global trade data method

  Use the U.S. commodity-flow method and global AI hardware trade data to estimate the current pace of investment.

  For the United States, hardware investment is estimated using domestic production, net imports, and inventory changes, with AI-related R&D and IPP investment added. For other economies, estimates are derived from the relationship between AI-related net imports and U.S. investment.

• dashboardLeading-indicator dashboard

  Track indicators such as semiconductor equipment imports, PMI subcomponents, import prices, memory procurement prices, and GPU rental prices.

  The report believes a dashboard approach is more suitable than a single indicator for determining when AI capital expenditure growth will slow.

Asset mapping & comparison

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

• Global macro and GDP

  AI capital expenditures are an investment-cycle variable currently attracting macro-market attention.

  Strengths

  The investment scale is large and leading indicators remain strong, potentially continuing to support capital formation and related trade flows.

  Weaknesses

  Transmission to the level of real GDP is constrained by import content, statistical definitions, and price inflation.

  Comparison

  Through 2028, AI investment as a share of GDP will remain below or within the historical 2%-5% range for general-purpose technology buildout cycles.

  Risks

  If capital expenditure growth slows faster than expected, it could weaken market pricing of AI's macroeconomic contribution.

• U.S. AI infrastructure investment

  The United States is one of the primary destinations for AI capital expenditures.

  Strengths

  U.S. AI investment is estimated at nearly $600bn in 2026, and approximately 70% of announced hyperscaler investment is located in the United States.

  Weaknesses

  The commonly used U.S. cloud-company capital expenditure measure overstates domestic U.S. investment, while part of nominal growth reflects price increases.

  Comparison

  Under the expanded measure, U.S. investment is approximately $581bn, below the approximately $794bn in total U.S. cloud-company capital expenditures frequently cited by the market.

  Risks

  Cost inflation, imported hardware deductions, and statistical measurement biases may reduce the contribution to real growth.

• AI hardware and semiconductor supply chain

  AI investment measurement relies heavily on data related to hardware, semiconductor equipment, memory, and GPUs.

  Strengths

  Trade data from supply chains in Taiwan, South Korea, and elsewhere provide high-frequency leading signals.

  Weaknesses

  Supply chains are highly interconnected, making revenue-based measures prone to double counting.

  Comparison

  The report uses the gross-margin revision method to reduce double counting and cross-checks it against official trade data.

  Risks

  If preliminary export data continue to slow, this may signal a decline in AI hardware capital expenditure growth.

Key data

• Global AI investment in 2026, expanded capital expenditure method$1019bnGoldman Sachs' preferred estimate based on expanded hyperscaler capital expenditures.
• U.S. AI investment in 2026, expanded capital expenditure method$581bnEstimate of domestic U.S. investment allocated according to project location.
• Global AI investment in 2026, gross-margin forecast revision methodApproximately $1060bnInvestment scale implied by gross-margin realization and forecast revisions for listed AI-related companies.
• Global AI investment in 2026, trade and national accounts method$1002bnEstimate derived from official statistics and global trade data.
• U.S. AI investment in 2026, trade and national accounts method$545bnCross-check estimate presented in the report summary.
• Annualized level of U.S. AI hardware investment in 2026Slightly below $500bnEstimate through May 2026 using the U.S. national accounts commodity-flow method.
• U.S. AI investment as a share of GDP in 20261.8%The corresponding global figure is 0.9%.
• U.S. AI investment as a share of GDP in 20282.8%The corresponding global figure is 1.4%.
• Cumulative global AI investment through the end of 2026Approximately $1.8tnAverage estimate from multiple methods for cumulative investment since 2022.
• Domestic share of hyperscaler capital expenditures by U.S. companiesApproximately 70%Estimated according to the location of announced investment projects, with approximately 15% in Asia and 9% in Europe.

Impact & implications

For macro markets, AI investment has become an important variable capable of affecting the capital expenditure cycle, trade flows, and certain GDP components. The report supports the view that AI infrastructure investment remains in expansion, but cautions investors against simply equating total capital expenditures by large U.S. cloud companies with U.S. AI investment. Because of high imported hardware content and measurement biases in U.S. national accounts regarding semiconductor purchases, the direct contribution of increased nominal spending to the level of real U.S. GDP may be limited.

Risks

• Estimates of AI investment scale rely on multiple assumptions that are difficult to fully verify.
• Property, plant, and equipment investment and finance leases are not clearly disclosed in some companies' capital expenditures, creating a risk of double counting.
• Internal transactions within the AI ecosystem are highly intertwined, and revenue-based measures may overstate final demand.
• Early June-July trade data suggest that investment growth may moderate from extremely strong levels.
• The rising share of cost inflation in nominal spending growth may weaken the increment to real investment.
• The high import content of U.S. AI hardware may limit its net contribution to total GDP.

What to watch

• Semiconductor equipment, AI hardware exports, and bilateral trade data from Taiwan and South Korea.
• U.S. imports, inventories, and domestic production of AI-related hardware.
• Semiconductor manufacturing equipment imports, related PMI subcomponents, and order indicators.
• Memory procurement prices, GPU rental prices, and AI hardware import prices.
• Whether consensus estimates for large technology companies' capital expenditures continue to be revised upward in 2027-2028.
• The price decomposition between nominal AI spending and real investment in official U.S. data.