Global artificial intelligence spending is far greater than commonly believed, according to a new research note from Goldman Sachs. The widely cited estimate of approximately $800 billion in capital expenditure by hyperscale cloud providers significantly underrepresents the total. When factoring in investments from private companies and non-U.S. firms, and stripping out non-AI related spending, Goldman Sachs projects global AI investment will reach $1.019 trillion by 2026.
Goldman Sachs economists Joseph Briggs and Sarah Dong detailed in an August 2 Global Economic Analysis report that the frequently referenced $794 billion hyperscaler capex figure underestimates total global AI spending by about $200 billion. Simultaneously, it overestimates U.S. investment by roughly $200 billion. After adjustments, Goldman Sachs estimates U.S.-based AI investment at $581 billion in 2026, with the global total reaching $1.019 trillion. Two separate cross-validation methods—one tracking revisions to listed companies' gross profit forecasts and the other using official national accounts and trade data—arrive at similar figures of $1.06 trillion and $1.002 trillion, respectively, closely aligning with the primary estimate.
This recalculation has direct implications for the macro market: it suggests the AI capex cycle is larger and more durable than previously expected. While Goldman's leading indicators show strong near-term momentum, trade data from Taiwan and South Korea hints at a potential moderate slowdown in investment growth for June and July.
Four Major Flaws in Common Indicators
The Goldman Sachs report identifies four fundamental flaws in using hyperscaler capex as a proxy for AI investment. First, it overlooks U.S. private companies and other listed firms that play a key role in the AI ecosystem. Goldman's credit team data indicates hyperscalers will directly account for only 40% of AI-related supply by 2026. Second, the metric completely excludes investments from non-U.S. companies, particularly those in China and other parts of Asia. Third, hyperscalers had capital expenditure exceeding $150 billion before the AI boom, meaning some current spending is unrelated to AI. Fourth, U.S. hyperscalers operate globally, with a significant portion of their capex occurring outside the U.S.
To address these issues, Goldman Sachs made several adjustments to the standard hyperscaler data: incorporating capex forecasts for other listed companies in its AI investment basket, adding media-disclosed data from key private firms, including capex from non-U.S. AI-related companies, and using 2022 expenditure levels as a baseline to exclude non-AI investments.
Three Methods Converge on Over $1 Trillion
Goldman Sachs employed three independent methods to estimate global AI investment, all yielding consistent results. The primary method (an enhanced hyperscaler capex model) estimates global AI investment at $1.019 trillion in 2026, with $581 billion within the U.S. Geographically, using location data from announced hyperscaler projects, Goldman estimates about 70% of U.S. hyperscaler capex is allocated to domestic projects, 15% to Asia, and 9% to Europe.
The first cross-validation method tracks revisions to AI-related listed companies' gross profit forecasts against a 2022 baseline to measure final demand increases. This approach estimates global AI investment at approximately $1.06 trillion in 2026, with cumulative AI-related spending since 2022 exceeding $1 trillion. The second method relies on official national accounts and global trade data. U.S. national accounts show AI-related hardware investment on an annualized basis has risen to about $463 billion by May 2026 (from the 2022 baseline). Adding approximately $100 billion in AI-related R&D and intellectual property investment brings the current annualized U.S. total to nearly $600 billion. For other countries with less data disclosure, Goldman uses global trade data and historical relationships between U.S. imports and investment to extrapolate, resulting in a global AI investment estimate of $1.002 trillion. The average of all three methods suggests cumulative global AI investment from 2022 to the end of 2026 will reach $1.8 trillion.
Capex-to-GDP Ratio Expected to Climb, Consistent with Historical Tech Cycles
For the medium-to-long-term trend, Goldman Sachs extrapolates from market consensus forecasts for public company capex, expecting the AI capex-to-GDP ratio to rise steadily. Specifically, the U.S. AI capex-to-GDP ratio is projected to increase from 1.8% in 2026 to 2.5% in 2027 and further to 2.8% in 2028. On a global basis, these figures are 0.9%, 1.3%, and 1.4%, respectively.
Goldman notes these levels are within the historical range of a 2% to 5% peak GDP investment impact seen during general-purpose technology buildout cycles. Even with significant upward revisions to 2027 capex forecasts, the AI investment-to-GDP ratio would remain within a reasonable range for historical technology cycles. The report also highlights that the timing of a slowdown in AI capex growth is a key source of uncertainty for the macro market. Goldman recommends a "dashboard" approach to track multiple leading indicators, including semiconductor manufacturing equipment imports from Taiwan and South Korea, related PMI sub-indices, import prices, memory procurement, and GPU rental prices. All leading indicators remain in a high range since 2022, suggesting a robust near-term growth outlook.
Inflation Erodes Real Investment Gains, Limiting GDP Impact
Despite the continued expansion of nominal AI investment, Goldman Sachs cautions investors about the erosion of real investment gains due to cost inflation. Official U.S. data shows that 8% of the nominal increase in AI-related hardware spending through 2026 is attributable to cost inflation rather than actual investment expansion. If this trend continues through the second half of 2026, the contribution of increased AI spending to real investment in 2026 will be less than in 2025.
Goldman also emphasizes that AI investment's impact on overall U.S. GDP remains limited due to two measurement biases. First, the U.S. national accounts do not count semiconductor procurement as investment goods. Second, the high import content of AI hardware is netted out in GDP calculations. This means that even with continued rapid growth in AI capex, its direct contribution to macroeconomic aggregates is structurally constrained.
Comments