The numbers are brutal. A National Bureau of Economic Research study surveying 6,000 executives across the US, UK, Germany, and Australia found that nearly 90% of firms say AI has had zero measurable impact on employment or productivity over the past three years. This comes despite 374 S&P 500 companies talking up AI in earnings calls and $250 billion poured into AI investments in 2024 alone. About two-thirds of executives say they use AI, but that usage averages just 1.5 hours per week. A quarter of respondents don't use it at work at all.
Apollo chief economist Torsten Slok summed it up bluntly: "AI is everywhere except in the incoming macroeconomic data." Outside the Magnificent Seven tech giants, he noted, there are "no signs of AI in profit margins or earnings expectations." The disconnect has economists invoking Robert Solow's famous 1987 observation about computers being "everywhere but in the productivity statistics." Solow was describing a productivity paradox that lasted from the 1970s until the late 1990s, when IT investments finally translated into measurable gains.
The academic picture is messier. A 2023 MIT study claimed AI could boost worker performance by 40%. But a 2024 MIT study led by Nobel laureate Daron Acemoglu projected only a 0.5% productivity increase over the next decade. "That's better than zero," Acemoglu said. "But it's just disappointing relative to the promises that people in the industry and in tech journalism are making." The Federal Reserve Bank of St. Louis found a modest 1.9% excess cumulative productivity growth since ChatGPT launched. Meanwhile, ManpowerGroup's 2026 Global Talent Barometer found that while regular AI use climbed 13%, worker confidence in the technology's utility dropped 18%. BCG research added another wrinkle: productivity actually fell when workers used four or more AI tools, with respondents reporting brain fog and more mistakes.
The integration problem is real. Most companies run on decades-old systems with proprietary data structures and messy dependencies that don't play nice with modern AI tools. That 1.5-hour weekly usage stat tells you everything about how hard it is to fit AI into existing workflows. But some economists see reasons for optimism. Stanford's Erik Brynjolfsson noted that fourth-quarter GDP was tracking up 3.7% while job growth stayed flat, a decoupling pattern that mirrors what happened when office automation finally clicked in the 1990s. Slok suggested AI productivity could follow a "J-curve," an initial dip followed by exponential gains. The question is whether companies have the patience, and the integration chops, to get there.