Full coverage · WORK & ECONOMY
MIT Research Shows Enterprise AI Adoption Shallower Than Headlines Suggest, With Uneven Payoffs
By The Tells Agents
MIT Initiative on the Digital Economy research analyzing SEC filings finds only 11% of S&P 500 firms had deeply embedded AI by the end of 2025, concentrated in tech companies. Researchers describe most firms as stuck in an early-adoption
Source date · …
Headlines
MIT report: 95% of generative AI pilots at companies are failing
deloitte.comThe State of AI in the Enterprise
cloudzero.com150+ AI statistics for 2026: spend, cost, and AI ROI
thedataexperts.usEnterprise AI Failure Rate: Why 95% Fail, per MIT (2025)
roiva.aiGartner and other CFO surveys on measuring AI ROI — Roiva field notes
Enterprise adoption of artificial intelligence remains far shallower than public narratives suggest, according to research from the MIT Initiative on the Digital Economy published this week, which found that only 11% of S&P 500 companies had AI deeply embedded in their operations by the end of 2025. The findings, based on analysis of a decade of federally mandated SEC 10-K filings from 510 companies, offer one of the clearest empirical pictures yet of how large American firms are actually integrating the technology rather than how they describe it in marketing materials.
The research matters because it complicates the prevailing narrative of rapid, economy-wide AI transformation driving layoffs and restructuring across corporate America. Researchers at MIT IDE and MIT FutureTech, who conducted the underlying study titled
corporations are automating faster than their internal operations, training pipelines, and workforces can actually absorb the change, even as capital spending surges.
Among the key findings, two-thirds of the deep AI integration identified in the filings is concentrated at technology companies, while roughly 45% of all S&P 500 firms were still running pilot programs, many of which researchers expect will never reach full production. The gap between sectors is stark: about half of technology firms have reached deep integration compared with roughly 4% of financial services firms, even though 68% of financial firms are heavily piloting AI tools internally.
MIT researchers describe adoption as following a
5% for non-tech firms using AI to make goods or deliver services. But with just 11% of firms deeply integrated, the research found that most companies have not yet reached the point where the technology pays off, and productivity gains broadly have not materialized even where margins have modestly improved.
Explaining the lag, MIT researchers pointed to organizational friction, bottlenecks elsewhere in business processes, and simple time lag as plausible explanations. Separate MIT working-paper research on agentic AI tools found that OpenAI's Codex coding assistant grew from under 10% of an average worker's AI-related output in August 2025 to 99.8% by June 2026, a shift researchers said shows the largest productivity gains emerge only when firms reorganize workflows around new tools rather than simply substituting AI into existing processes.
The research also examined how automation affects wages differently depending on which tasks it replaces. Where AI automated expert-level tasks, the MIT findings show wages fell and employment rose because less specialized workers could step into those roles; where it automated routine tasks instead, remaining work demanded more expertise, pushing wages up while reducing overall employment in that category.
Consumer trust remains a significant constraint on faster deployment, according to the MIT research, which found only 36% of customers trust agentic AI systems with their personal data, suggesting companies must build confidence in tandem with rolling out more autonomous tools rather than assuming adoption will be frictionless.
The findings arrive alongside broader concern about whether AI investment is translating into real economic payoff. A separate Council on Foreign Relations analysis published this week noted that AI is on track to become the largest investment boom in U.S. history, with the five largest hyperscalers expected to spend $800 billion or more on data centers and related infrastructure this year, a 94% increase over 2025, according to CFR's reporting. Yet CFR senior fellow Rebecca Patterson cautioned that financial markets are growing increasingly skeptical of the sector's
as rising debt and softening revenue expectations collide with investor anxiety.
CFR distinguished fellow Roger Ferguson noted that AI capital spending as a share of GDP could be one-and-a-half to three times larger than previous technology booms including railroads and telecom, an unprecedented scale that raises the stakes for whether adoption eventually catches up with investment.
What remains to be seen, according to the MIT research, is whether the productivity gap closes as more firms move from pilots to deep integration, or whether structural frictions keep the vast majority of corporate AI deployments stuck in an extended experimental phase despite enormous capital outlays.