The Federal Reserve has published a staff note that pulls together three surveys of AI use across U.S. firms, workers and executives.
Together, they show a market moving on several tracks at once: the Census Bureau put firm-level AI adoption at about 18% at year-end 2025, the Real-Time Population Survey put work-related generative AI use at about 41% of workers in November 2025 and the Atlanta Fed’s Survey of Business Uncertainty estimated that 78% of the labor force works at firms that have adopted AI.
Why the numbers look so far apart
The spread shows not only speed, but also how adoption is being measured. The Fed says the differences reflect units of analysis, question framing, materiality of use, information asymmetries between respondents and possible social desirability bias.
The Census series changed in November 2025, when BTOS broadened its question from AI use in “producing goods or services” to AI use in “any of its business functions,” and the bureau said the wording change was large enough to require a new time series.
That methodological break is central to the 18% firm-adoption figure. The Fed notes that the legacy BTOS series had reached about 10% adoption before the revision, while the new broader series averaged about 18% by year-end 2025 and put planned adoption above 20% for the first half of 2026.
Late-2025 adoption, then, cannot be read as a simple continuation of the earlier firm-use series without accounting for the broader definition.
Where adoption is actually concentrated
The sector picture is clearer. The Fed says professional services and finance lead in both the firm and worker data: in BTOS, adoption in those sectors stood at about 33% and 30%, while the worker survey put work-related AI use at 62% and 63%.
Manufacturing showed the fastest year-over-year growth in worker-reported use, but finance and professional services remained the highest-adoption sectors.
Firm size also helps explain why the numbers look so far apart. The Fed’s tables show that firms with 250 or more employees account for just 0.9% of U.S. firms but 56.2% of employment, while firms with 1 to 49 employees make up 95% of firms and 26.6% of employment.
That distribution helps reconcile an 18% firm-weighted adoption rate with a 78% employment-weighted estimate: large employers are adopting earlier, and their decisions reach a much larger share of workers.
The productivity question
The productivity question remains less settled. The Fed note places the adoption data beside an infrastructure buildout that pushed combined 2025 capital expenditure at Amazon, Google, Meta, Microsoft and Oracle to $412 billion, or about 1.31% of U.S. GDP, but it does not claim a direct economywide payoff yet.
A March Atlanta Fed working paper offers a narrower read, finding positive labor productivity gains that vary by sector and are expected to strengthen in 2026, especially in high-skill services and finance, while near-term aggregate job losses remain limited.
What 2026 worker data adds
More recent worker data extend the same pattern into 2026. Economists at the St. Louis Fed reported that 43% of U.S. workers said they used AI on the job in January-February 2026, and that 5.2% of all U.S. work hours were already being spent using it.