The Potentially Large Effects of AI on Economic Growth (Goldman Sachs): Summary
A summary of Goldman Sachs' 2023 report on generative AI and growth: 300 million jobs exposed, 7% global GDP, how the numbers were made, and what Goldman and its critics have said since.
A summary of Goldman Sachs' March 2023 report on generative AI and economic growth: the headline numbers, how they were produced, and what Goldman and its critics have said since. We do not host the report. Read it on Goldman's site [1].
What the report is
The Potentially Large Effects of Artificial Intelligence on Economic Growth is a Goldman Sachs Global Economics Analyst report dated 26 March 2023. Its authors are Jan Hatzius, Joseph Briggs, Devesh Kodnani and Giovanni Pierdomenico [1]. It estimates how much work generative AI could automate and what that could mean for productivity and GDP.
The headline numbers
| Measure | Estimate [1] |
|---|---|
| Current jobs exposed to some AI automation | About two-thirds |
| Current work AI could substitute | Up to one-fourth |
| Full-time jobs exposed to automation | Equivalent of 300 million |
| Work that could be automated globally (employment-weighted) | 18% |
| Range across scenarios | 15–35% of work |
| US labour productivity growth | Just under 1.5 percentage points a year over 10 years after widespread adoption |
| Global GDP | 7%, almost $7 trillion a year, over 10 years |
How Goldman got there
The authors used O*NET data on the task content of more than 900 US occupations, later extended to over 2,000 occupations in the European ESCO database [1]. They classed 13 of O*NET's 39 work activities as exposed to AI, and counted tasks up to difficulty 4 on O*NET's 7-point scale as automatable. They assumed that jobs with much outdoor or physical work cannot be automated by AI [1].
Which jobs are most exposed
In the US, the report puts the share of tasks that could be automated at 46% for administrative work and 44% for legal work, against 6% for construction and 4% for maintenance [1]. Across US employment, it expects 7% of workers to be substituted by AI, 63% to be complemented by it and 30% to be unaffected [1].
On new work, the report cites research by David Autor and co-authors finding that 60% of US workers today are employed in occupations that did not exist in 1940 [1].
The assumptions that move the result
The productivity estimate is sensitive to how hard a task AI can handle. Across the report's scenarios, the boost to annual US productivity growth ranges from 0.3 to 3.0 percentage points. Spread over 20 years the effect is roughly half as large, and over 30 years about one-third [1].
Timing is the other assumption. The report notes that the productivity booms from the electric motor and the personal computer began about 20 years after the breakthrough, once roughly half of US businesses had adopted the technology [1]. Goldman did not add the findings to its baseline forecasts at the time. It also estimated that US investment in AI could approach 1% of GDP by 2030 [1].
What Goldman has said since
- November 2023. AI could automate around 25% of labour tasks in advanced economies and 10–20% of work in emerging economies. Goldman left its forecasts unchanged until at least 2027 for the US and 2028 for other economies, and estimated a 0.4 percentage point boost to US GDP growth by 2034 [2].
- February 2025. AI could raise US labour productivity by 15% over roughly 10 years, about $4.5 trillion of annual US GDP. The effect should start to show in productivity figures in 2027 and peak in the early 2030s. Only 6% of US companies reported using AI for regular production [3].
- August 2025. AI could displace 6–7% of the US workforce if widely adopted, within a range of 3% to 14%, with unemployment rising by half a percentage point during the transition [4].
- August 2026. Goldman expects $1 trillion of AI-related investment worldwide in 2026, including $581 billion in the US [5].
The skeptics' case
MIT economist Daron Acemoglu reaches a far smaller number. In his April 2024 paper he estimates no more than a 0.71% increase in total factor productivity over 10 years, and says the figure is likely below 0.55% [7]. MIT News summarised his estimate as a GDP increase of 1.1% to 1.6% over 10 years, with a productivity gain of roughly 0.05% a year [8].
Goldman set the two views side by side in a June 2024 issue of its Top of Mind series. There Acemoglu forecast that AI would raise US productivity by only 0.5% and GDP growth by only 0.9% cumulatively over the next decade, and said only 4.6% of all tasks would be affected. Briggs expected AI to automate 25% of work tasks and raise US productivity by 9% and GDP growth by 6.1% cumulatively over the decade [6].
How to read the report today
The authors presented the 2023 figures as an estimate of potential, not a forecast [1]. Goldman's own later work puts the start of the effect in 2027 [2][3]. The measures to watch are adoption, which Goldman put at 6% of US companies for regular production in early 2025 [3], and investment, which it expects to pass $1 trillion in 2026 [5].
Related: our AI trends briefing and the AI Readiness Index guide.
Frequently asked questions
How many jobs did Goldman Sachs say AI could affect?
The March 2023 report said generative AI could expose the equivalent of 300 million full-time jobs to automation, and that about two-thirds of current US jobs are exposed to some degree. In August 2025 Goldman estimated that AI could displace 6–7% of the US workforce if widely adopted.
How much could AI raise global GDP, according to Goldman Sachs?
The report estimated that generative AI could eventually raise annual global GDP by 7%, almost $7 trillion, over a 10-year period. Goldman did not add this to its baseline forecasts at the time.
What share of work could AI automate?
The report put it at about one-fourth of current work tasks in the US and 18% of work globally on an employment-weighted basis, within a range of 15–35% across scenarios.
When will AI show up in GDP data?
Goldman's later research expects AI to start showing up in US productivity figures in 2027, with the peak impact in the early 2030s.
Why does Daron Acemoglu predict a much smaller effect?
Acemoglu expects only 4.6% of all tasks to be affected by AI over the next decade. In his 2024 paper that gives a total factor productivity gain of no more than 0.71% over 10 years.
Where can I read the full Goldman Sachs report?
Goldman Sachs publishes it on its research site; the link is in the sources below. The report is dated 26 March 2023.
Sources
- The Potentially Large Effects of Artificial Intelligence on Economic Growth — Goldman Sachs Global Economics Analyst (March 2023)
- AI may start to boost US GDP in 2027 — Goldman Sachs (November 2023)
- China's advances could boost AI's impact on global GDP — Goldman Sachs (February 2025)
- How will AI affect the global workforce? — Goldman Sachs (August 2025)
- Global AI investment is forecast to exceed $1 trillion in 2026 — Goldman Sachs (August 2026)
- Top of Mind: Gen AI: too much spend, too little benefit? (PDF) — Goldman Sachs (June 2024)
- The Simple Macroeconomics of AI (PDF) — Daron Acemoglu, MIT (April 2024)
- What do we know about the economics of AI? — MIT News (December 2024)