TRANSMISSION // 2607.08920v2
AI is entering big business—but productivity gains remain unproven
Source published
A study estimates growing deep AI adoption among S&P 500 firms, especially in technology. It reports a profitability pattern but no measured differences in productivity or capital spending.

How many companies have moved beyond talking about AI to building it into their operations? A study uses corporate filings to estimate adoption among S&P 500 enterprises. It finds that deep integration is growing, while measured productivity differences remain elusive.
From AI talk to business integration
AI tools are widely available for individual tasks, but using a chatbot is not the same as building AI into a company’s operations. The authors of AI Adoption in S&P 500 Firms focus on this deeper form of adoption: whether AI is integrated into business processes, or used in producing goods and delivering services.
They examine S&P 500 firms from 2016 to 2025. Their measure draws on companies’ annual reports filed with the U.S. Securities and Exchange Commission (SEC). The authors argue that these filings can help distinguish concrete business activity from AI-related publicity because companies are legally barred from making materially false or misleading statements in them.
Adoption is growing, but concentrated
The study estimates that in 2025, 11% of S&P 500 enterprises had deeply integrated AI into business processes. A further 10% were using AI in production or service delivery. The authors report that adoption had more than quadrupled from 5% in 2022, with adoption slowly accelerating among non-technology firms. The technology sector adopted AI especially aggressively and accounted for two-thirds of deeply integrated enterprise adoption.
The researchers also report a “J-curve” in profitability as firms move from no adoption to deep adoption. This is an observed pattern, not proof that adopting AI causes profits to rise. The abstract does not establish that AI alone explains profitability differences. The study reports no corresponding differences in capital spending or productivity. That finding should not be read as evidence that AI has already produced measurable productivity gains across the firms studied—or as proof that no individual firm gained productivity.
Within technology firms, adoption was higher among companies with more employees and higher Tobin’s q, a market-based measure related to a firm’s value. The authors report no such relationship among other firms. These are observed associations; the abstract does not establish why they occur.
What the evidence can—and can’t—tell us
Company filings offer a way to study large enterprises systematically, but they are not a direct record of every AI tool used by every employee. The results concern S&P 500 firms, so they do not by themselves describe businesses outside that group. The abstract does not provide enough methodological detail to assess how adoption was classified, how that classification was validated, or how the analysis handled other factors that might influence adoption and profitability.
This account is based on the paper’s abstract, not a full-text examination. The source is an arXiv preprint, and the supplied material does not establish its peer-review status. The findings are best treated as an estimate rather than a final verdict on AI’s economic effects. The narrower takeaway is that deep adoption appears to be increasing among S&P 500 enterprises, while the study reports no measured differences in productivity or capital spending.
- artificial intelligence
- AI adoption
- S&P 500
- business productivity
- corporate filings
Related reading
Automatically matched by shared topic keywords. These links do not imply that the sources agree or establish a consensus.
AI Read the News, but the Opening Price Had Already Moved
Source published
Shared keywords: artificial intelligence
How lost practice could make AI use hard to reverse
Source published
Shared keywords: ai adoption
When one dominant signal distorts sparse autoencoder scores
Source published
Shared keywords: artificial intelligence
A public image collection aims to give ERCP AI a clearer test
Source published
Shared keywords: artificial intelligence
Looking Inside Mixture-of-Experts Models, One Neuron at a Time
Source published
Shared keywords: artificial intelligence