Spending on computers, software, research, and data centers is the clearest national measure of how fast the US economy is building for AI. Two St. Louis Fed researchers estimated that these four kinds of investment added 0.97 percentage points to annualized GDP growth over the first nine months of 2025, about 39 percent of that growth. The figure measures building, not how much work inside industries has changed.
Key takeaways
- The St. Louis Fed's January 12, 2026 analysis counts four categories of investment as AI-related: information processing equipment, software, R&D, and data centers.
- Their combined contribution was 1.3 points in the first quarter of 2025, 1.16 in the second, and 0.48 in the third, in annualized percentage points of GDP growth. That was 30 percent of growth in the second quarter and 11 percent in the third.
- The same four categories contributed 0.81 points in 2000, during the dot-com boom, or 28 percent of growth.
- A contribution depends on how fast spending grows, not how large it is, so a slowdown in the third quarter does not mean spending fell.
- As of October 2026, the 2025 quarterly figures are the latest published by the Fed system; the Board's July 2026 note, which reaches the first quarter of 2026, gives no figures in its text.
What was counted and how
Hannah Rubinton and Bontu Ankit Patro built their estimate from components of business investment in the national accounts. Information processing equipment, software, and R&D come from the Bureau of Economic Analysis (BEA). Data centers come from Census Bureau construction spending, available from 2014. Each category's contribution is roughly its real growth rate multiplied by its share of GDP, expressed as annualized percentage points. An illustrative example: a category equal to two percent of GDP that grows 10 percent in a year adds about 0.2 points.
The authors call the choice a proxy. Much R&D has nothing to do with AI, although work on new models does land there, and computers and software are bought for many purposes besides AI.
The 2025 numbers, quarter by quarter
The figures below are the contributions published in the St. Louis Fed post, in percentage points. The real GDP rates are annualized and were estimates when the post appeared; BEA revises them.
| Quarter | Real GDP growth | Equipment | Software | R&D | Data centers |
|---|---|---|---|---|---|
| 2025 Q1 | -0.6% | 0.90 | 0.41 | -0.05 | 0.04 |
| 2025 Q2 | 3.8% | 0.22 | 0.57 | 0.22 | 0.15 |
| 2025 Q3 | 4.3% | 0.16 | 0.07 | 0.22 | 0.03 |
| Long-run average | 3.2% | 0.18 | 0.10 | 0.07 | 0.04 |
Equipment led in the first quarter, at 0.90 points. Software reached 0.57 points in the second quarter, above earlier highs of 0.44 in 1997 and 0.49 in 2021. By the third quarter, equipment and software had fallen back toward their long-run averages.
A comparison with the dot-com years
Over the first nine months of 2025, real GDP growth averaged 2.51 percent, and the four categories contributed 0.97 points (0.90 without data centers), or 39 percent of it. In 2000, when growth was 2.94 percent, the same categories contributed 0.81 points, a 28 percent share. Software made the biggest difference: 14 percent of growth in 2025 against 3.7 percent in 2000. The two spans differ in length, nine months against four quarters, and the authors describe the debate over AI's future productivity effects as unsettled.
What changed in 2026
Two later releases add context, though neither replaces the 2025 table. In the Federal Reserve Board's FEDS Note of July 17, 2026, Paul Soto, Mason Thieu, and Jeffrey Allen built a similar proxy from software, data centers, power facilities, computer and peripheral equipment, and net exports of computers, peripherals, and parts. They report that these components contributed meaningfully to quarterly GDP growth from 2025 through the first quarter of 2026, with software and computer equipment the largest positive contributors. The net-export drag offset much of the gross investment in quarters when imports rose sharply, and the authors warn that ignoring net exports would overstate AI's direct contribution. Their text gives no percentage-point figures, so none are quoted here.
BEA's advance estimate for the second quarter of 2026, released July 30, 2026, put real GDP growth at 1.5 percent, after 2.1 percent in the first quarter. The agency said the rise in investment reflected equipment and intellectual property products, with software (mainly prepackaged software) behind the second item, and that imports increased more than in the first quarter.
What these figures cannot show
Four limits apply when reading the numbers as proof that industries are being reshaped.
- Spending is not use. A server bought in 2025 may sit partly idle in 2026, and the data say nothing about what customers or employees do with it.
- Imports are subtracted. In BEA's July 30, 2026 advance estimate for the second quarter, the agency notes that imports "are a subtraction in the calculation of GDP." Where equipment is imported, domestic output rises by less than the spending suggests. The St. Louis Fed post does not adjust for this; the Board's July 2026 note does.
- Levels versus changes. The post states that contributions depend on growth, not high levels alone. A third quarter near the average does not mean investment fell.
- The St. Louis series ends in 2025. BEA's third estimate for the second quarter of 2026, published September 30, 2026, put real GDP growth at 2.2 percent, up from 1.5 percent in the advance estimate. No official percentage-point estimate of AI's contribution for 2026 was found, and the third-quarter 2026 GDP release had not appeared as of October 27, 2026.
Series worth checking next to investment data
No single number shows reshaping, so this illustrative checklist pairs the spending data with measures of use and outcomes.
- Business adoption surveys, which ask firms whether they use AI in production.
- Productivity by industry, which shows whether output per hour moved.
- Employment and hours by occupation, which show whether tasks or headcounts changed.
- Prices, which show whether cost savings reached customers.
FAQ
Does AI account for most US economic growth?
The St. Louis Fed estimates the four categories supplied 39 percent of growth in the first nine months of 2025. The share depends on which categories are counted. The authors call R&D an imperfect proxy, and the post does not adjust for imports. Other analysts publish much higher and much lower shares, and those are not repeated here.
Is AI investment bigger now than during the dot-com boom?
In the St. Louis comparison, the contribution was larger in 2025 (0.97 points) than in 2000 (0.81). The authors caution that the averaging periods differ.
Why did the contribution fall in the third quarter of 2025?
Investment growth slowed. The authors stress that a smaller contribution reflects slower growth, not lower spending.
Does this show which industries are being reshaped?
No. The data cover investment across the whole economy and do not sort spending or benefits by the industry that uses it.
Related reading
- Which industries are least exposed to AI
- Canada's survey results on AI in health care and other industries
- Why model versions and dates matter when comparing AI products
- An open-source planning and routing optimizer used in operations
More in AI and Society.
Last verified: October 2026
Sources
- Tracking AI's Contribution to GDP Growth, Rubinton H, Patro B, Federal Reserve Bank of St. Louis On the Economy, January 12, 2026
- GDP (Advance Estimate), 2nd Quarter 2026, U.S. Bureau of Economic Analysis, July 30, 2026
- GDP (Third Estimate), 2nd Quarter 2026, U.S. Bureau of Economic Analysis, September 30, 2026
- The AI Buildout and the Economy: Publicly Available Data to Assess AI's Impact, Soto P, Thieu M, Allen J, FEDS Notes, Board of Governors of the Federal Reserve System, July 17, 2026