Published estimates say artificial intelligence will add to economic growth, but by modest and uncertain amounts. Economist Daron Acemoglu's task-based model puts the gain in total factor productivity at no more than 0.66 percent over 10 years, while OECD researchers project that AI could raise annual growth in per capita real income by between 0.1 and 0.95 percentage points in their central scenario, depending on the country. Official productivity statistics for 2026 do not yet separate out AI, so the estimates remain projections, not measurements.
Key takeaways
- Acemoglu (2024) multiplies the share of tasks AI can profitably take over, 4.6 percent, by average cost savings of 14.4 percent to reach a 0.66 percent productivity gain over 10 years.
- An OECD analysis (March 2026) covering 76 countries finds gains that differ widely by country, from 0.1 to 0.95 percentage points a year in a medium-paced adoption scenario, with the United States near the top.
- U.S. nonfarm business productivity has grown at an annualized 2.1 percent since the fourth quarter of 2019, against 1.5 percent in the previous business cycle, but the Bureau of Labor Statistics release does not mention AI.
- Canadian business-sector labor productivity rose 1.0 percent in the second quarter of 2026, the largest quarterly gain since 2020, in a release that also does not mention AI.
- A growth estimate depends on three inputs: how many tasks AI can do, how much it saves on each, and how fast firms adopt it.
How one widely cited estimate is built
Acemoglu's paper, The Simple Macroeconomics of AI, appeared as NBER Working Paper 32487 in May 2024 and later in the journal Economic Policy in 2025. It starts from published exposure and experiment data and chains them into one number. The steps are short enough to follow by hand.
- About 20 percent of U.S. labor tasks, weighted by the wage bill, are exposed to AI.
- Of those exposed tasks, 23 percent could be automated at a profit within 10 years, which leaves 4.6 percent of all tasks affected.
- Average labor cost savings on those tasks are taken from experiments on writing and customer support, and translate to overall cost savings of 14.4 percent.
- Multiplying 4.6 percent by 14.4 percent gives about 0.66 percent more total factor productivity over 10 years, roughly 0.064 percent a year.
Output gains are somewhat larger because firms also invest. The paper gives a range of 0.93 to 1.16 percent of GDP over a decade if the investment response is modest, and 1.4 to 1.56 percent if there is a large investment boom. Acemoglu also argues that the figure may be overstated, because early evidence comes from easy-to-learn tasks, and applying an adjustment for harder tasks lowers the productivity estimate to under 0.53 percent.
What the OECD projects, and why countries differ
A second set of estimates comes from OECD economists Francesco Filippucci, Peter Gal, Katharina Laengle, and Matthias Schief, with Muhammed Yildirim of Harvard Kennedy School. Their March 2026 paper (OECD Artificial Intelligence Papers No. 57) feeds task-level gains, sector exposure, and adoption paths into a model of 76 countries and 45 sectors. A May 2026 summary brief by the same authors reports the central results.
In the medium-paced adoption scenario, AI adds 0.1 to 0.95 percentage points a year to per capita real income growth over the next decade. Mexico and Colombia sit near the low end, while Luxembourg, Switzerland, and the United States sit near the top because of stronger adoption and heavier exposure through finance and information technology. The rapid scenario reaches just under 1.2 points a year, close to the 1 to 1.5 points that the information technology boom of the mid-1990s is estimated to have added in the United States.
Starting points differ too. The authors estimate that high-intensity AI use in core business functions was about 4 percent of firms on average across OECD countries in 2024, from about 0.2 percent in Mexico to about 7 percent in Ireland. Their model also finds that countries gain little from other countries' adoption alone: for the median OECD country, only about 0.03 percentage points a year.
What official productivity statistics show so far
The Bureau of Labor Statistics reported on September 3, 2026 that U.S. nonfarm business labor productivity rose at an annualized 1.4 percent in the second quarter and 2.2 percent over the year. Since the fourth quarter of 2019, productivity has grown at an annualized 2.1 percent, higher than the 1.5 percent rate of the previous cycle that began in the fourth quarter of 2007. The same day, Statistics Canada reported that business-sector labor productivity rose 1.0 percent in the second quarter of 2026, the largest increase since the pandemic-driven 7.7 percent rise of 2020, with hours worked down 0.1 percent.
Neither release mentions artificial intelligence. Productivity is output per hour worked, so it records that something changed, not why. A rise can come from new equipment, business cycles, shifts in who is working, or software of any kind. Treating these figures as proof of an AI effect, or as proof of its absence, goes beyond what they measure.
Four sources side by side
| Source | Type | Reported figure | What it cannot show |
|---|---|---|---|
| Acemoglu, NBER 32487 (2024) | Task-based model for the United States | Up to 0.66 percent productivity gain over 10 years | Effects of tasks and tools not yet invented |
| OECD AI Papers No. 57 (2026) | Multi-country trade model | 0.1 to 0.95 percentage points a year in income growth, medium adoption | Actual adoption speed |
| BLS productivity release (September 2026) | Measured U.S. data | 2.1 percent a year since late 2019 | The cause of the rise |
| Statistics Canada productivity release (September 2026) | Measured Canadian data | Plus 1.0 percent in one quarter | Any trend, or any cause |
Questions to put to any growth claim about AI
- Is the number a projection from a model or a measurement from statistics?
- Which share of tasks does it assume AI can do, and who estimated that share?
- Does the figure refer to total factor productivity, labor productivity, GDP, or income per person?
- What time span and adoption speed does it assume?
- Does the source say what would make its estimate wrong?
FAQ
How much could AI add to GDP?
Published figures vary with method. Acemoglu's model gives roughly 0.93 to 1.56 percent of GDP over a decade, depending on how much firms invest, and the OECD scenario work implies yearly income-growth gains between 0.1 and about 1.2 percentage points depending on country and adoption speed.
Is the recent rise in U.S. productivity caused by AI?
The statistics do not say. The Bureau of Labor Statistics release does not mention AI, and linking productivity to specific tools takes separate studies of firms and workers.
Why do the estimates differ so much?
They rest on different assumptions about which tasks AI can do, how much each task is improved, and how quickly firms adopt it. A small change in any of the three moves the total.
Does a country benefit if only other countries adopt AI?
Only a little, according to the OECD model. Cheaper imports help, but domestic producers lose competitiveness, so the net gain for a median OECD country is about 0.03 percentage points a year.
Related reading
- What AI costs businesses and how that affects growth
- AI and job creation
- Cleaning services as an example of efficiency claims
- How deep learning works and where it is applied
More in AI and Society.
Last verified: October 2026
Sources
- The Simple Macroeconomics of AI, Acemoglu D, NBER Working Paper 32487, May 2024; Economic Policy 40(121), 2025
- AI meets trade: Global linkages and the cross-country distribution of the gains from AI, Filippucci F, Gal P, Laengle K, Schief M, Yildirim M, OECD Artificial Intelligence Papers No. 57, March 2026
- Global trade linkages and the cross-country distribution of the gains from AI, SUERF Policy Brief No. 1463, May 2026
- Productivity and Costs, Second Quarter 2026, Revised, U.S. Bureau of Labor Statistics, September 3, 2026
- Business-sector productivity, second quarter 2026, Statistics Canada, The Daily, September 3, 2026