This year, the atmosphere on a Lower Manhattan trade floor has been curiously divided. Venture cash pours into data centers, corporate executives talk feverishly about automation, and screens glow as AI stock tickers reach all-time highs. However, the most recent GDP statistics show that… typical.
Goldman Sachs experts are now referring to this discrepancy as the “productivity paradox.” According to their estimates, artificial intelligence made virtually no contribution to the growth of the U.S. GDP in 2025, despite massive investments in the field. The ending is startling. And possibly one that is uncomfortable.
| Category | Details |
|---|---|
| Institution | Goldman Sachs |
| Lead Economists | Joseph Briggs & Devesh Kodnani |
| Key Finding | AI contributed nearly zero to U.S. GDP growth in 2025 |
| Forecast Impact | +0.4 percentage points to U.S. GDP by 2034 |
| Concept | “J-curve” productivity effect |
| Reference | https://www.goldmansachs.com/insights |
Their research indicates that imports, such as specialist gear and advanced semiconductors, have accounted for a large portion of the spending increase. Yes, expenditure on high-end semiconductors by American companies is recorded as investment, but it also reduces growth through imports in the GDP calculation. For the time being, the overall impact is minimal.
The AI explosion seems to be occurring in server rooms and earnings calls rather than in the national accounts just now. The economists at Goldman refer to the pattern as a traditional “J-curve.” The first step is investment. Then there is disruption. The quantifiable productivity increases come later. Occasionally, much later.
According to their base case, broad AI deployment over a ten-year period may increase U.S. productivity growth by 1.5 percentage points a year. According to their estimates, AI might increase GDP growth in the United States by about 0.4 percentage points by 2034, with lesser increases in other developed and emerging nations. Those are not insignificant figures. However, they are far away.
The dichotomy becomes apparent when strolling around a corporate office park in New Jersey. Consultants are testing AI technologies that can quickly summarize contracts inside one building. In another, human resources departments are reorganizing teams, redistributing responsibilities, and subtly cutting staff. These adjustments might eventually free up employees for higher-value work. However, a large portion of the work at the moment is rearrangement.
Productivity figures rarely reflect reorganization right away. The analysis’s Goldman economists, Briggs and Kodnani, contend that automation will reduce labor costs and free up time for new projects. However, how well AI develops and how businesses use it will determine how big of an impact that is.
It appears that investors think AI is already changing the world. A future with stronger margins and leaner operations is reflected in valuations. GDP, however, reflects actual production rather than expected efficiency.
Additionally, there is a measuring problem. Quality and speed increases are difficult for GDP to measure. The statistical improvement might be negligible if a marketing team uses AI to create campaigns twice as quickly but producing about the same amount of billable work. The macro data appears stable, and the experience feels transforming.
Whether AI will lead to the kind of regime change that occurred during the first and second industrial revolutions is still up in the air. According to scholarly study, overall factor productivity generally declines with economic maturity, with the exception of few technological turning points. AI, according to some, is just that. Some people are not persuaded.
Executives confidently discussed replacing monotonous cognitive work at a technology conference earlier this year in San Francisco. However, several acknowledged in private discussions that assimilation is difficult. Change is resisted by legacy systems. Workers need to be trained. Data must be cleaned. Time and money are lost due to these transitory frictions.
Goldman’s staff warns against double counting in the meantime. Long-term growth projections already incorporate technological advancement. For many years, investments in information and communication technology have been the main driver of productivity increases. Exaggeration is possible when AI estimates are simply layered on top.
There is a strange contradiction when markets rise on AI stories yet economic growth is slow. Expectations seem to have outpaced the data, at least for the time being. The situation is further complicated by emerging markets. In many emerging nations, where jobs are still concentrated in industries with less exposure to AI, like construction and agriculture, Goldman predicts lesser productivity improvements. They say adoption will take longer.
Global growth disparities may worsen rather than close as a result of such uneven influence. Back in Manhattan, an analyst looks for indications of margin growth directly related to AI while perusing quarterly data. Some businesses mention early savings. Some people talk about “strategic investments.” The wording is cautious.
We can be living in the calm midst of a change, the costly, perplexing stage before quantifiable benefits. The foundation is being laid via capital deployment, retraining initiatives, and corporate restructuring. However, skylines are not instantly raised by foundations.
There is nothing new about the productivity dilemma. Similar observations were made by economists in the early days of the internet. Computers were widely used for years without clearly increasing productivity. Then, slowly, the information changed. It’s unclear if AI will go that route.
GDP growth is currently slowing down. The rate of inflation fluctuates. And the benefits of artificial intelligence are still to be realized, somewhere between spreadsheets in Washington and server farms chumming in Nevada. Maybe things will be different in 2027. Or maybe early expectations will change.
