Only 11% of S&P 500 firms have deeply integrated AI, MIT-led study finds

Just 11% of S&P 500 companies had deeply integrated AI into core business strategy and day-to-day operations by 2025, according to a new study posted to arXiv and led by researchers at MIT FutureTech and Carnegie Mellon University. The figure is striking given the volume of executive commentary about AI transformation over the past three years, and it carries a direct implication for enterprise technology and procurement leaders: most large organizations, even well-resourced public companies, are still running pilots.

The researchers analyzed more than 4,400 firm-year observations from 510 S&P 500 members between 2016 and 2025. Rather than relying on executive surveys, they built a five-level adoption scoring framework and applied it to AI-related passages in annual SEC 10-K filings, where companies have legal obligations to avoid materially misleading disclosures. Passages were extracted using a broad set of AI keywords and then scored using GPT-5-mini against the rubric. The team validated scores against U.S. Census Bureau business survey data and spending data from fintech company Ramp, finding strong alignment with both.

A five-level picture of where enterprises actually stand

The scoring system ranged from companies that mentioned AI only as a general industry trend at the bottom, up through exploratory efforts, pilot programs, and production deployments, to the highest tier: AI embedded across operations and central to business strategy. By 2025, 21% of S&P 500 firms had reached at least the production-deployment level. Of those, 11% had achieved the top tier. Overall AI adoption has more than quadrupled since 2022, largely driven by the post-ChatGPT acceleration that began in 2023.

Technology companies accounted for roughly two-thirds of firms at the deepest integration level. Software companies led all sectors at 70% deep adoption, with semiconductor and hardware firms close behind. Financial services companies showed meaningful activity, particularly in AI-enabled products and select operational processes, though most remained in pilot rather than full production mode. Consumer staples, food producers, household products companies, and utilities were the furthest behind, with roughly a quarter or more of firms in those sectors making no substantive mention of AI adoption in filings.

Profit margins are moving, productivity is not yet

The study's financial findings matter for any operations or finance leader building the business case for AI investment. Companies with mature, deeply integrated AI reported stronger net profit margins than peers at lower adoption levels. The researchers described the pattern as a J-curve: early adopters absorb costs from infrastructure build-out, organizational redesign, and training before financial performance improves. For non-tech industries, the effect was especially pronounced, consistent with the idea that manufacturing, healthcare, and retail companies have more efficiency headroom because they are starting from lower baselines of digital automation.

Productivity is a different story. The researchers found no consistent link between AI adoption and higher revenue per employee, their measure of productivity. That absence suggests the broad efficiency gains widely anticipated from generative AI have not yet shown up at scale in financial data, even among the most advanced adopters. The study also found no consistent relationship between AI adoption and capital expenditure for most companies. The researchers noted that large infrastructure investments remain concentrated among a small number of major technology firms. Most other companies access AI through cloud providers and software subscriptions, meaning spending shows up as operating expense rather than capex, a distinction that matters for how procurement and finance teams categorize and track AI costs.

Employment picture is more nuanced than headlines suggest

The study found no broad evidence of workforce reductions at companies with high AI adoption. Technology firms at the deepest integration levels actually tended to employ more workers, reflecting the engineering talent and data infrastructure required to deploy AI at scale. Outside tech, company size showed little relationship with adoption level.

The researchers were careful not to interpret those numbers as reassurance that AI will have limited labor impact over time. Companies may be simultaneously eliminating some roles and hiring workers with AI-related skills, with the two movements offsetting each other in headcount figures. Job posting data and worker transition records would be needed to surface that dynamic, and the current study does not reach that level of granularity.

What enterprise operators should take from this

The study's methodology has limits the authors acknowledge openly. It covers only large publicly traded companies, and the adoption scores are derived from disclosures rather than direct observation of internal systems. Correlations in financial outcomes do not prove causation. Still, the scope of the dataset, a decade of filings across more than 500 companies, and its validation against independent spending data make it one of the more rigorous attempts to measure enterprise AI deployment at scale.

For CIOs and operations leaders benchmarking their own organizations, the numbers provide a useful frame: even among the largest, best-resourced public companies in the U.S., the overwhelming majority are still not running AI as a core operational system. The gap between the 11% that have crossed that threshold and the broader S&P 500 is where most enterprise decisions about scaling, resourcing, and governance will be made over the next two to three years. The researchers indicate that barrier factors include implementation cost, required organizational change, and uncertainty around operational and regulatory risk, not just technology readiness.

The research team notes that whether today's pilot deployments mature into genuine enterprise integration, and whether the profit-margin advantages currently visible in financial statements eventually translate into measurable productivity gains, remain open questions that future longitudinal analysis will need to answer.