The latest official American figures show sustained growth accompanied by a cooling of the labor market and a stabilization of inflation, all factors which point to an increase in productivity linked to AI.
“AI is everywhere, except in macroeconomic data,” Torsten Slok, chief economist at Apollo Management, an American asset manager, recently wrote, echoing the famous paradox stated by Robert Solow in 1987: at the time, despite the computer revolution, the productivity of American workers was stagnating.
The same is true today with AI, a technology that everyone (or almost) presents as revolutionary, but whose concrete economic impact is slow to appear, raising fears of a financial bubble similar to that of the dot-com. However, things could be changing.
Acceleration of growth and slowdown of the job market
Indeed, Erik Brynjolfsson, who directs the Digital Economy Laboratory at Stanford University, claims to have identified the first signs of clear productivity gains enabled by AI in the latest American macroeconomic data. “The data published this week drastically corrects the idea that the impact of AI on the American economy is still a long time coming,” he wrote in a note published this week.
The Bureau of Labor Statistics, the US government’s principal agency in the field of labor economics and statistics, has significantly revised downwards the number of jobs created by the US economy in 2025: from 584,000 jobs initially estimated, the number fell to 181,000, or 403,000 fewer, making last year the weakest in terms of job creation, excluding recession periods, since 2003. Strikingly, this reality combines with a robust growth rate, still estimated at 3.7% for the last quarter. For Erik Brynjolfsson, this decoupling, marked by maintaining high production with a significantly lower labor input, clearly points to an increase in productivity.
A lagging measure of productivity
An observation that, according to him, is reinforced by certain micro-economic analyzes. With his team, he claims to have identified a drop in recruitment in sectors exposed to AI, particularly at the start of their career: according to his work, hiring for junior positions fell by around 16%, while those who used AI to strengthen their skills saw their employment rate increase. This suggests, he said, that companies are starting to use AI for some simple tasks once done by juniors, an analysis also shared by other experts.
How can we explain that the impact of AI on productivity is only becoming apparent now, more than three years after the launch of ChatGPT? According to the expert, this is due to the natural adoption cycle of technology. “General-purpose technologies, from the steam engine to the computer, do not bring immediate gains. Instead, they require a period of massive, often unmeasured, investments in intangible capital: reorganization of business processes, retraining of the workforce, and development of new business models. During this phase, measured productivity is underestimated, as resources are diverted to investments. Updated 2025 data in the United States suggests that we are exiting now from this investment phase to enter a harvest phase where these previous efforts begin to manifest themselves in the form of measurable production,” he writes.
A recent survey of 12,000 American and European companies concluded that labor productivity had increased by 4%, with no loss of jobs.
A self-fulfilling prophecy?
If these figures are enough to reassure those who worry about the formation of an AI bubble, they must nevertheless be taken with a grain of salt, as stated by Erik Brynjolfsson himself, who specifies that productivity indicators are notoriously volatile, and that it will therefore take several additional periods of sustained growth to confirm the trend.
Other recent research, while also concluding a notable impact of AI on productivity in the United States, specifies that the latter remains modest, and that we also note productivity gains in industries little exposed to AI. “For example, the contribution of the manufacturing industry has increased more than that of information, despite less use of AI,” note two researchers from the Federal Bank of Kansas City in a study which has just been published. “We find that while AI adoption aligns with faster productivity growth across industries, it explains only a small portion of the overall productivity gain,” they conclude.
It is also very likely that the pharaonic spending deployed by American big tech on AI infrastructures has a lot to do with the latest good growth figures across the Atlantic. Far from demonstrating a significant impact of AI on productivity, the latest macroeconomic figures from the United States would rather illustrate a self-fulfilling prophecy, with spending invested in AI contributing to artificially inflate growth… with the risk of a severe backlash if the expected gains do not materialize.
Competition from Chinese open source
More worryingly for the American tech giants, some experts believe that, even if the productivity gains of AI become evident, it is not certain that they are the ones who get the brownies out of the fire. This is the thesis defended by Louis-Vincent Gave, head of the asset management consulting company Gavekal, during a recent interview. According to him, the rise of open source AI constitutes a major threat to the future profits of American giants.
“Where will the economic benefits of productivity gains go? American technology giants have opted for what I call the fortress approach: they want to build the best large language model (LLM) in a closed system, and think that users will pay fifty dollars a month to access it. Chinese companies, for their part, have chosen an open source approach,” he summarizes. However, this open source AI is quickly gaining market share, including across the Atlantic.
“The Andreessen Horowitz venture capital fund recently said that 80% of the start-ups that come to its door use Chinese LLMs,” he says. “The competition is very fierce. You have read the news that Airbnb has moved from ChatGPT to Alibaba’s Qwen because its open source system can easily be modified according to their needs. The valuation of many American companies is essentially based on a winner-takes-all vision. In my eyes, this seems a very daring hypothesis.” More than on productivity gains, it is perhaps on the performance of Chinese AI that American companies now have their eyes fixed.
