AI Adoption in S&P 500 Firms
summary
The gist
The adoption of artificial intelligence (AI) by large enterprises is an important potential source of aggregate productivity improvement and labor market impact.
In short
The study measures deep AI integration in S&P 500 firms using SEC filings to distinguish it from hype. In 2025, 11% of these firms had deep integration, and adoption has quadrupled since 2022. Deep integration boosts net profits but does not significantly affect productivity or capital expenditure. Technology firms show stronger ties between deep AI use and high valuation (Tobin's Q).
Key concepts
- Deep AI Adoption Measure
- The authors created a new way to measure how deeply AI is used in a company, based on SEC 10-K filings. This score ranges from 1 (no mention of AI) to 5 (AI is central to strategy and performance), helping separate real integration from mere hype.
- J-Curve Effect
- This describes how profitability changes as a firm moves from having no AI adoption to deep integration. Firms experience a 'J-curve' in net profit margins, meaning profits increase sharply as they adopt the technology, then slow down.
- Tobin's Q
- Tobin's Q is a valuation metric used to assess how much the market values a company relative to its assets. The paper finds that technology firms with deep AI integration have significantly higher Tobin's Q ratios, indicating greater market confidence in their future value.
- Sector-Specific Adoption
- AI adoption rates vary by industry. While non-tech firms are adopting AI more slowly, the technology sector is aggressively adopting it, accounting for two-thirds of deep integration. This suggests that tech firms are currently leading the charge in business process automation.
Terminology used across episodes
This episode discusses
- AI Adoption in S&P 500 Firms · Paper Radio
- Crashing Waves vs. Rising Tides: Findings on AI Automation from Thousands of Worker Evaluations of Labor Market Tasks
- A Constructive Prediction of the Generalization Error Across Scales
- The AI risk repository: A meta-review, database, and taxonomy of risks from artificial intelligence
The paper
AI Adoption in S&P 500 Firms · Read on arXiv
Alfred P. Sloan Foundation Grant No. G-2025-25158 · MIT FutureTech · Carnegie Mellon University
Transcript
Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.
Ruth: I'm Ruth.
Warren: And I'm Warren, here with Iris.
Ruth: We're looking at the paper: "AI Adoption in S&P 500 Firms".
Paper summary: Ruth: Welcome back. We're talking about this paper today, "AI Adoption in S andP five hundred Firms <ref:2607.08920#pg1,AI Adoption in S&P 500 Firms>." The main idea is that AI adoption isn't just about having a tool; it’s about how deeply it’s woven into the actual business processes of big companies.
Warren: Right, so we're looking at what the authors claim here: they study AI adoption among S andP five hundred firms from two thousand sixteen to two thousand twenty-five and they focus on that deep integration in business operations because that’s where the real impact happens for productivity and labor markets <ref:2607.08920#pg1>.
Iris: I'm interested in how they defined this deep integration since that seems central to their whole argument. They developed a novel measure based on SEC ten-K filings, which is interesting because it ties it to what companies have legally committed to reporting <ref:2607.08920#pg1,based on SEC 10-K filings>.
Ruth: Exactly, Iris. The paper says this rubric ranges from one, which means no mention of AI at all, up to five where AI is a core component of the firm’s strategy and financial performance deeply embedded across business functions and operations.
Warren: And they give some concrete numbers on that integration. In two thousand twenty-five they found that eleven percent of S andP five hundred enterprises had AI deeply integrated into their business processes, and another ten percent were using AI in the production of goods and delivery of services <ref:2607.08920#pg1,11% of S&P 500 enterprises had AI deeply integrated into their>.
Iris: That's a significant jump from where they started back in two thousand twenty-two which was only five percent deep adoption then <ref:2607.08920#pg1>. The paper notes that this adoption has more than quadrupled since two thousand twenty-two with slow but accelerating adoption among non-technology firms, but very aggressive adoption in the technology sector <ref:2607.08920#pg1,accelerating adoption among non-technology firms, but very aggressive adoption in the>.
Ruth: That's a big piece of context. So, what does this mean for the big picture? It suggests that while some sectors are catching up slowly, the technology sector is driving most of that deep integration.
Warren: The paper highlights that generative AI tools are useful for personal or professional use, but their focus here is on the deep integration in business processes of large enterprises which they see as bellwethers for firm adoption more broadly.
Paper summary: Iris: And they point out some specific sectors leading this charge. In two thousand twenty-five the technology sector leads in AI adoption, with over fifty percent of those firms scoring four or five on their rubric <ref:2607.08920#pg1>. The leading-edge sectors are software and services and semiconductors and technology hardware and equipment, with over fifty percent of the firms scoring four or five.
Ruth: That’s a lot of concentration. Now, let's talk about what this deep adoption actually does to the company's bottom line according to the paper.
Warren: The authors look at firm profitability and they find a "J-curve" effect as firms move from no adoption to deep adoption in business processes on profitability, but they observed no differences in capital expenditure or productivity.
Iris: That’s a key detail for me, Warren. If there's no difference in capital expenditure or revenue per employee, it suggests the productivity gains aren't immediately showing up in those specific measures.
Ruth: And the paper says that across all sectors, those firms using AI in production and having deeply integrated AI in business processes report higher net profit margins with that J-curve effect.
Warren: They did find a positive relationship between deep AI integration and net profit margin across both technology and non-technology sectors, with fifteen percent higher for non-technology firms and only two point six percent higher for technology firms.
Iris: That small percentage difference for the technology sector is something I'd like to dig into later, especially since we see other relationships here. The paper also looked at productivity and valuation metrics.
Ruth: What about that? Does deep AI integration actually change how much revenue a firm generates per employee or how much capital it invests?
Iris: The paper says there is no significant association between AI adoption and productivity, as measured by revenue per employee, and there is no significant correlation between AI adoption and capital expenditure.
Warren: But they did find a statistically significant relationship between technology sector AI adoption and Tobin’s q. Furthermore, firms with deep AI integration exhibit Tobin’s q ratios that are forty-seven point eight percent higher.
Paper summary: Ruth: So the value of the firm, measured by Tobin's q, seems to be positively linked to this deep integration in the tech sector, even if it doesn't translate directly into higher revenue per employee.
Iris: And I think what’s interesting is how this concentration plays out among larger firms. The paper notes that deeper AI adoption concentrates among larger firms in the technology sector, and for those non-technology firms, the association between the AI score and headcount is weak.
Warren: But in the technology sector, they found a much stronger positive pattern among those firms with more employees when AI adoption scores four or five. This suggests that large technology firms are better positioned to adopt AI at scale because they already have more complete data architecture and data pipelines.
Ruth: So, the finding is that scale matters for technology firms in terms of deep adoption, linking it to their existing infrastructure. That seems like a practical implication for how companies should approach this shift.
Iris: And what about the limitations? The paper does flag that while adopting new technologies can improve profits, the adoption of AI requires organizational transformation and skill reconfiguration, which can be challenging and introduce risks. They note that management practices are a central source of productivity differences across firms (Bloom and Van Reenen, two thousand ten) <ref:2607.08920#pg1>.
Warren: So the authors aren't suggesting AI is an automatic profit booster; it demands a lot of internal change. That’s a necessary caveat for anyone looking at this research.
Ruth: So, to wrap up this look at "AI Adoption in S andP five hundred Firms," the paper shows that deep integration, measured by their specific rubric, is growing rapidly in tech-heavy areas and it links positively to firm value metrics like Tobin's q <ref:2607.08920#pg1,AI Adoption in S&P 500 Firms>.
Warren: And it’s important to remember that while adoption is spreading, the paper cautions that the actual economic benefits aren't immediately reflected in simple measures of productivity or capital spending.
Iris: It really highlights the difference between using AI as a tool and making it a core part of how a business operates.
Ruth: That’s what they are pointing toward for their future work, focusing on those organizational transformations needed to make this deep adoption actually pay off in the long run.
Conclusion: Ruth: So we’re looking at how AI adoption is really happening in big companies, specifically S andP five hundred firms, in this paper called "AI Adoption in S andP five hundred Firms."
Warren: Yeah, the authors are tracking that deep integration of AI into business processes over a decade.
Iris: They’re using a specific way to measure that deep adoption based on SEC filings to try and separate the real stuff from just hype.
Ruth: They found that even though things are growing, this deep integration is still relatively low, around eleven percent in twenty twenty-five for S andP five hundred firms.
Warren: It’s a jump from five percent back in twenty twenty-two, but they show it accelerating faster in the tech sector.
Iris: That technology sector really seems to be driving most of the deep integration we’re seeing right now.
Ruth: The paper says while firms are getting deeper into AI use, they aren't seeing immediate changes in things like capital expenditure or revenue per employee.
Warren: No big boost to those specific productivity metrics, just a shift in how the companies are structured.
Iris: But they did find that this deep AI integration is linked to higher net profit margins across the board with a kind of J-curve effect as adoption ramps up.
Ruth: And for firms in the tech sector, that profit margin lift was only about two point six percent, while non-tech firms saw fifteen percent.
Warren: The paper also found a link between deep AI integration and Tobin's q ratios being forty seven point eight percent higher, especially in those tech firms.
Iris: That suggests the market values these deeply integrated companies more highly than just their immediate operational efficiency shows.
Ruth: So the main thing is that deep AI adoption is moving forward, it’s concentrating in tech, and it’s starting to show up in firm value metrics.
Warren: It makes you wonder what kind of organizational changes are actually needed for this adoption to become a sustained advantage.
More episodes
- Daily Summary for 2026-11-19
- 2608.21274-Recommendation Quality and the Concentration of Consumption: Experimental Evidence from Netflix
- 2408.02757-A nonparametric test for diurnal variation in spot correlation processes
- 2306.16393-High-Dimensional Canonical Correlation Analysis
- 2605.17299-Geometric Brownian motion with intermittent entries and exits
- 2607.04753-Fooling Yourself: how narratives shape beliefs
- 2408.14671-Inference on Treatment Effects with Repeatedly Mismeasured High-Dimensional Controls
- 2201.06898-Difference-in-Differences Estimators for Treatments Continuously Distributed at Every Period
- 2501.15307-Influence Functions: A Unified Differentiation-Based Framework
- The day in conversation for 2026-11-19