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

As of 2025, only 11% of S&P 500 companies had AI deeply integrated into their core business processes, and just 10% more were using AI in the actual production of goods or delivery of services. That finding, from a July 2026 working paper by MIT FutureTech researchers Yang Yu, Martin Fleming, Lucy Hampton, Christophe Combemale, and Neil Thompson, cuts sharply against the narrative that large enterprises are moving fast on AI.

The paper, published on arXiv and supported by the Alfred P. Sloan Foundation, covers S&P 500 firms from 2016 through 2025. Its methodology sets it apart from most enterprise AI surveys: instead of relying on self-reported questionnaires or earnings call mentions, the researchers built their adoption measure from SEC 10-K annual filings, where securities law holds companies to a high standard of accuracy on material business disclosures.

Adoption has quadrupled, but from a very low base

The headline number is more nuanced than it first appears. Meaningful AI adoption across S&P 500 firms has more than quadrupled since 2022, when the study estimates just 5% of those companies qualified. That rate of change is real. But the absolute levels entering 2026 remain modest, particularly outside the technology sector.

Technology-sector firms account for roughly two-thirds of deeply integrated enterprise AI adoption, the paper found. Non-technology companies are moving, but slowly, and the gap is widening rather than closing. For procurement and operations leaders at industrial, financial services, or healthcare companies benchmarking against tech peers, that disparity signals a structural lag that won't close quickly.

For context, the U.S. Census Bureau's Business Trends and Outlook Survey reported in late April 2026 that 19.8% of U.S. enterprises of all sizes used AI in any business function in the prior two weeks. That broader figure includes far more lightweight uses, such as AI-assisted email drafting or customer chat tools, which the MIT FutureTech rubric intentionally filters out.

The J-curve problem for enterprise P&L

For the CIOs and COOs already several months into an AI deployment program, the study's financial findings carry a practical warning. When the researchers regressed adoption status against financial outcomes, they found a J-curve pattern in profitability: firms moving from no AI adoption toward deep integration see no immediate gains, and may see a dip, before profitability improves. Capital expenditure and productivity showed no measurable difference between adopters and non-adopters, suggesting the friction is organizational rather than hardware-driven.

The researchers cite prior work by McElheran et al. linking this dynamic to the difficulty of realigning management practices and workforce skills with new technology. Firms that pair AI deployments with intentional changes to workflows and task structures recover faster. Those that layer AI onto existing processes without structural adjustment appear to stall in the flat part of the curve.

A separate finding from researcher Christos Makridis, cited in the paper, adds a management dimension: in organizations where employees can clearly articulate their company's AI strategy, adoption rates are higher, even after controlling for firmographics and timing. Strategy clarity, not just tool availability, predicts whether workers actually integrate AI into daily routines.

Technology firms behave differently

Among technology-sector firms specifically, the paper identified two factors associated with more advanced AI adoption: higher Tobin's q values, a market-based measure of how much investors value a company's assets relative to their replacement cost, and larger employee headcounts. Neither relationship held for non-technology firms. That divergence suggests that in tech, companies with strong investor confidence and scale are using those advantages to move faster on AI integration, reinforcing existing gaps.

The researchers also highlight three systemic factors that will continue to slow adoption across industries regardless of intent. First, scaling laws mean that AI models capable of reliably handling enterprise-class tasks at acceptable accuracy are expensive to build and still don't match average worker performance for most task types, according to a separate 2026 study by Mertens et al. cited in the paper. Second, organizational transformation carries real risk: mismatches between AI capabilities and existing management structures create the J-curve drag noted above. Third, network risk, the concern that AI errors propagate through supply chains and service provider relationships, is making business leaders cautious about deep integration in interdependent operations.

What this means for your team

• Treat 21% as your peer benchmark: only about one in five S&P 500 firms show any meaningful AI use in operations as of 2025. If your organization is still in evaluation mode, you are not behind the majority, but the gap to tech-sector leaders is widening.
• Plan for the J-curve before you commit: budget cycles and board narratives should account for a profitability dip during the integration phase. Tie AI investment cases to a 2-3 year horizon, not a single fiscal year.
• Audit organizational readiness, not just technology readiness: the research links adoption success to management clarity and task reallocation, not infrastructure spend. Assign accountability for workflow redesign alongside the technology rollout.
• For supply chain and procurement teams specifically: the paper's network risk framing is a direct prompt to assess how AI errors in your own systems, or a key supplier's systems, could propagate. Build error-tolerance requirements into vendor evaluation criteria now.