The economists cannot agree on whether artificial intelligence did anything for the American economy in 2025. That disagreement is the story.
Goldman Sachs chief economist Jan Hatzius told clients in February that AI-related investment contributed basically nothing to US GDP growth last year. The United States spent roughly $400 billion on AI infrastructure in 2025. According to Goldman, the return on that investment, measured in the official growth statistics, was zero.
The St. Louis Fed reached the opposite conclusion in a January analysis. "Our analysis suggests that the recent investments in AI-related categories have contributed significantly to the real GDP growth in 2025," it wrote, adding that AI's contribution had surpassed the information-technology contribution during the dot-com boom. Source: https://www.stlouisfed.org/on-the-economy/2026/jan/tracking-ai-contribution-gdp-growth
The San Francisco Fed split the difference in a February economic letter, finding that "most macro-studies of productivity growth find limited evidence of a significant AI effect." Even firms that report finding AI useful, the letter noted, show little evidence of transformative gains at scale. Source: https://www.frbsf.org/research-and-insights/publications/economic-letter/2026/02/ai-moment-possibilities-productivity-policy/
This is not a technical dispute. It is an argument about timing, distribution, and measurement — and it arrives at exactly the wrong moment for the AI industry's credibility.
The Bureau of Labor Statistics does not share the optimism either. Nonfarm business productivity grew 2.1 percent in 2025, up from an annual average of 1.4 percent over the preceding decade. Source: https://www.bls.gov/news.release/prod2.nr0.htm Federal Reserve chairman Jerome Powell told Congress in February that some of that improvement "may be AI" — but he could not isolate the contribution. The number cannot be pulled out of the aggregate data.
The gap between what companies say they are doing and what the macroeconomic data shows is not new. Economists have a name for it: Solow's Paradox, from a 1987 quip attributed to Robert Solow that you could see the computer age everywhere except in the productivity statistics. The same pattern followed the internet. It followed mobile computing. Each time, the investment preceded the measured payoff by years.
The Council of Economic Advisers under Barack Obama documented the pattern in a 2016 report: general purpose technologies routinely take fifteen to thirty years to show up in aggregate productivity data, because the gains are initially concentrated in islands of early adoption while the rest of the economy absorbs the disruption.
The current cycle may be shorter — the infrastructure buildout is faster than previous technology waves — but the basic dynamic appears similar. The San Francisco Fed's literature review cited work by Alex Imas, Daron Acemoglu, and others finding that realized AI productivity gains remain hard to document at the macro level even as firm-level case studies accumulate.
What is easier to count is the anticipated displacement. A survey of 750 chief financial officers by Duke University and the Federal Reserve Banks of Richmond and Atlanta found that the respondents expected to eliminate approximately 502,000 positions to AI in 2026 — roughly half a million jobs, a ninefold increase from the 55,000 AI-attributed losses in 2025. That figure represents less than half a percent of the total American workforce, but the direction of movement is consistent: companies are planning around AI-driven workforce reduction in numbers that would have been unthinkable two years ago. Source: https://fortune.com/2026/03/24/cfo-survey-ai-job-cuts-productivity-paradox-2026/
The contradiction is not that one side is wrong and the other is right. The BLS data is real. The Goldman data is real. The St. Louis Fed analysis is real. They are measuring different things at different levels of aggregation over different time horizons. The micro-level evidence of AI making individual workers more productive — in customer support, in software development, in legal document review — is substantial enough that multiple analysts, including Goldman, put the gains for top-performing adopters at around 30 percent in specific task categories. The macro-level evidence remains elusive.
What changes if the measurement finally catches up? The Penn Wharton Budget Model projected in September 2025 that AI would lift annual productivity growth by 1.5 percent by 2035, approaching 3 percent by 2055. Those are long-horizon numbers. The disagreement about 2025 is, at one level, a disagreement about when the J-curve breaks upward. Source: https://budgetmodel.wharton.upenn.edu/issues/2025/9/8/projected-impact-of-generative-ai-on-future-productivity-growth
In the meantime, the half-million-job estimate sits uneasily alongside the zero-contribution headline. Companies are acting on AI's promise before the data confirms the promise has arrived. That is the definition of anticipation. Whether the productivity follows the anticipation, and on what timeline, is the economic question of the next decade.
