By Rajesh Beri | July 14, 2026
Global AI spending will hit $2.52 trillion in 2026 — a 44% surge year-over-year, according to Gartner. Eighty-eight percent of organizations report using AI in at least one business function. Every major enterprise has an AI strategy deck. Most have an AI budget line. Many have a Chief AI Officer.
And yet, according to a new study from MIT FutureTech and Carnegie Mellon University analyzing 510 S&P 500 companies over a decade of SEC filings, only 11% have AI deeply integrated into their core business processes. A further 10% have reached production deployment. The remaining 79% are still exploring, piloting, or not mentioning AI at all.
That's not a gap. That's a chasm between what enterprises are spending on AI and what they're actually doing with it.
The study, posted to arXiv on July 9, 2026, offers something rare in the AI adoption discourse: a methodology that can't be gamed by hype. Instead of relying on surveys where executives self-report optimistic adoption numbers, the researchers built a five-level scoring system using companies' annual 10-K filings — documents where federal law prohibits materially false or misleading statements. When you face SEC enforcement risk for exaggerating, the buzzwords disappear and reality surfaces.
The timing of this research is brutal for anyone writing AI budget proposals. The U.S. Census Bureau's Business Trends and Outlook Survey reports that only 19.8% of U.S. enterprises used AI "in any of its business functions" in the two weeks prior to the late April 2026 survey. Among firms with 250+ employees — the group most likely to have AI budgets — adoption reaches 35.3%. These aren't pilot numbers. These are "did you actually use it?" numbers. And the answer, for the vast majority of American businesses, is still no.
The 11% Club: What Deep Integration Actually Looks Like
The MIT/CMU researchers weren't measuring whether a company mentioned AI in a press release or earnings call. They built a rubric that separates signal from noise:
- Level 1: No meaningful mention of AI adoption
- Level 2: Exploring AI as a future possibility
- Level 3: Piloting AI in products or processes, but not emphasizing financial results
- Level 4: Using AI in production of goods and delivery of services
- Level 5: AI deeply embedded across operations and business strategy
The researchers extracted AI-related passages from thousands of SEC filings using keyword identification, then evaluated them using GPT-5-mini against their rubric, with manual review for consistency. The resulting database covered more than 4,400 firm-year observations from 2016 to 2025.
To validate their approach, they compared scores with both the U.S. Census Bureau's Business Trends and Outlook Survey and business spending data from fintech company Ramp. The correlations held — suggesting their metric captures genuine operational reality, not corporate narrative.
The findings: only 55 S&P 500 firms scored a 5 (deep integration). Another 51 scored a 4 (production use). That's 106 companies out of 510 — roughly one in five — that have moved beyond experimentation. Meanwhile, 18% of S&P 500 firms made no meaningful mention of AI adoption in their most recent filings. These aren't small companies. These are among the 500 largest publicly traded firms in the United States.
The Tech Sector Distortion: Two-Thirds of the Winners Come From One Place
The study reveals a concentration problem that should alarm any CXO benchmarking against "industry averages."
Technology companies account for roughly two-thirds of all firms scoring at the highest adoption levels. Within the tech sector, 62% of firms achieved Level 4 or 5 adoption. Software companies led all subsectors at 70%. Semiconductor and technology hardware firms followed closely.
Outside technology? The numbers collapse. Financial services showed substantial activity, but most of it remained in pilot deployments rather than production. Banks, insurers, and diversified financial firms reported integrating AI into selected products without yet positioning it as a core driver of financial performance.
Industries like consumer staples, food production, household products, and utilities lagged furthest — with roughly a quarter or more of companies making no meaningful AI mention at all.
The strategic implication: when McKinsey reports that 88% of organizations use AI in at least one business function, the number is technically accurate but practically misleading. There's a vast difference between an employee using ChatGPT to draft an email and a company embedding AI into its supply chain optimization, pricing engines, and strategic decision-making. The MIT/CMU study measures the latter. And the number that matters — deep integration — is 11%.
The J-Curve: Why Your AI Investment Looks Like It's Failing (When It Might Not Be)
Perhaps the study's most important finding for CFOs and boards is the profitability pattern researchers describe as a "J-curve."
Companies in early adoption stages — Levels 2 and 3 — often experienced lower profitability than firms with no AI adoption at all. The costs are real and front-loaded: digital infrastructure buildouts, organizational restructuring, employee training, governance framework development, and the operational friction of adapting workflows around AI capabilities that keep changing.
But firms that pushed through to Level 5 adoption showed significantly higher net profit margins than companies at any other level. The researchers are careful to note this is correlation, not causation — more profitable companies may simply have more resources to invest in AI. But the J-curve shape appeared especially pronounced outside the technology sector, suggesting that traditional industries may have the most to gain from deep AI integration precisely because they start from lower baselines of digital automation.
This pattern maps directly to what Kyndryl found in its 2026 People Readiness Report: only 9% of organizations — dubbed "Pacesetters" — have done three things simultaneously: redesigned roles around AI, implemented change management for their workforce, and built genuine workforce readiness. These Pacesetters are 1.5x more likely to achieve AI-related revenue growth and 1.6x more likely to report better innovation outcomes. The other 91% are somewhere on the painful downward slope of the J-curve, spending money without seeing returns.
The J-curve also explains a statistic that has puzzled analysts: a February 2026 National Bureau of Economic Research study of nearly 6,000 executives found that roughly 90% of firms using AI report no measurable impact on productivity or employment. As Forbes reported in its analysis of the AI validation gap, only 28% of AI use cases fully succeeded and met ROI expectations in a Gartner survey of 782 infrastructure leaders. Twenty percent failed outright.
The productivity gains aren't absent. They're delayed — hidden behind the organizational transformation costs that most enterprises underestimate by an order of magnitude.
The Spending Paradox: Where $2.52 Trillion Actually Goes
The MIT/CMU study found no consistent relationship between AI adoption and capital expenditures across most companies. Only a handful of major technology firms — members of the so-called Magnificent Seven — are making the massive infrastructure investments associated with building AI systems. Most companies purchase AI capabilities through cloud providers or software services, meaning AI spending appears primarily as operating expenses.
This creates a measurement problem that compounds the J-curve. When AI spending shows up as OpEx rather than CapEx, it gets scrutinized quarterly rather than amortized over years. CFOs face pressure to show immediate returns on what is fundamentally a multi-year transformation investment. The result: 100% of CIOs are budgeting for AI, but half have already blown their budgets — not because AI costs too much, but because the organizational transformation surrounding AI costs far more than the technology itself.
PwC's 29th Global CEO Survey quantifies the executive disillusionment: 56% of CEOs reported that their companies have seen neither higher revenues nor lower costs from AI. Only 12% reported that AI delivered both cost savings and revenue gains. The project ships, more investment follows, and at no point does anyone verify the underlying assumptions.
As the CloudBees "State of Code Abundance 2026" report found: only 31% of AI spend can be attributed to specific business outcomes, despite 51% of leaders expressing high confidence in their ability to measure ROI. The confidence exists. The attribution does not.
Framework #1: Enterprise AI Maturity Assessment
Use this assessment to determine where your organization sits on the MIT/CMU adoption spectrum — and whether you're positioned to break through the J-curve or stuck in the expensive middle.
Dimension 1: AI Integration Depth (Score 1-5)
| Level | Description | Diagnostic Questions |
|---|---|---|
| 1 — No Mention | AI not referenced in strategy or operations | Could you describe AI's role in your business to a board member? |
| 2 — Exploring | AI discussed as future possibility; no active projects | Do you have funded AI initiatives with assigned teams? |
| 3 — Piloting | AI tested in 1-3 use cases; no financial results attributed | Have any AI pilots generated measurable business outcomes? |
| 4 — Production | AI used in product delivery or service operations | Does AI influence revenue-generating processes today? |
| 5 — Deep Integration | AI embedded in strategy, operations, and decision-making | Would removing AI systems materially impact your P&L? |
Dimension 2: Organizational Readiness (Kyndryl Pacesetter Criteria)
| Factor | Lagging (<25th percentile) | Developing (25-75th) | Leading (>75th) |
|---|---|---|---|
| Role Redesign | No roles changed for AI | Some roles augmented | 61%+ roles redesigned around AI workflows |
| Change Management | No formal program | Training programs exist | Workforce understands new operating model + guardrails |
| Workforce Readiness | <15% workforce AI-ready | 15-30% ready | >30% workforce proficient with AI tools |
| Governance | No AI policies | Partial governance | Clear policies on AI decision boundaries + registry |
Dimension 3: Financial Tracking Maturity
| Level | Characteristic | Action Required |
|---|---|---|
| Ad Hoc | AI spend buried in IT/cloud budgets | Separate AI cost center; tag all AI-related OpEx and CapEx |
| Tracked | AI spend identified but not attributed to outcomes | Implement outcome attribution: link each AI initiative to a business metric |
| Optimized | AI spend tied to specific business outcomes with ROI measurement | Build J-curve tolerance: set 18-36 month ROI horizons, not quarterly |
Scoring Your Position
- Total Score 3-7: Pre-J-Curve. You haven't invested enough to feel the dip yet — but you also haven't started climbing.
- Total Score 8-11: J-Curve Valley. Maximum pain, maximum organizational stress, maximum temptation to cut AI budgets. This is where 70%+ of enterprises sit today. Do not retreat. The study shows firms that push through report higher margins.
- Total Score 12-15: Ascending. You're past the worst of the transformation costs and beginning to see returns. Focus shifts from proving AI works to scaling what works.
Framework #2: J-Curve Navigation Playbook — The 6-Phase Transformation Timeline
Based on the MIT/CMU findings, Kyndryl Pacesetter data, and the CMU/Accenture AI Adoption Maturity Model, here's a phased approach to moving from the J-curve valley to deep integration:
Phase 1: Baseline Reality (Months 1-2)
Objective: Know exactly where you stand — no self-deception.
- Audit every AI initiative against the 5-level MIT/CMU rubric above
- Classify each initiative: exploring, piloting, production, or integrated
- Identify your "AI honesty gap" — the delta between what you tell the board and what your SEC filing would say
- Separate AI spending from general IT/cloud budgets
- Key metric: What percentage of AI spending can be attributed to a specific business outcome? (Industry average: 31%)
Phase 2: Kill the Pilots That Won't Scale (Months 2-4)
Objective: Stop burning resources on dead-end experiments.
- Evaluate every pilot against three criteria: (1) clear business outcome, (2) path to production, (3) organizational readiness to adopt
- Terminate pilots that lack all three — Gartner data shows 20% of AI use cases fail outright
- Redirect resources to 2-3 high-conviction bets with production-ready data and executive sponsorship
- Key metric: Pilot-to-production conversion rate (industry average: <25%)
Phase 3: Redesign Work, Not Just Workflows (Months 3-8)
Objective: Close the organizational readiness gap that kills ROI.
- This is the phase most enterprises skip — and it's why they stay stuck in the J-curve valley
- Redesign roles around AI capabilities, not just layer AI onto existing processes (61% of Pacesetter organizations have already done this)
- Implement change management that helps employees understand their new operating model
- Build governance: clear policies on which decisions AI can and cannot make (only 33% of organizations have this today)
- Workforce AI readiness has dropped 6 points from 2025 to 2026 — moving faster doesn't help if your people can't keep up
- Key metric: Percentage of workforce that reports understanding the organization's AI strategy (strongest correlate of adoption per Makridis 2025)
Phase 4: Production Deployment With Attribution (Months 6-14)
Objective: Move from "AI in use" to "AI with measurable business impact."
- Deploy AI into revenue-generating or cost-reducing processes with clear baselines
- Implement outcome attribution frameworks that isolate AI's contribution from other variables
- Set 18-36 month ROI horizons — the J-curve data shows profitability improvements emerge in later stages, not immediately
- Track leading indicators (task completion time, error rates, customer satisfaction) alongside lagging indicators (revenue, margin)
- Key metric: Number of AI use cases with attributed business outcomes (not just "in production")
Phase 5: Deep Integration (Months 12-24)
Objective: Embed AI into strategy and decision-making, not just operations.
- AI informs capital allocation, pricing, product development, and strategic planning
- This is where the MIT/CMU data shows the profitability inflection point
- Technology sector firms at this stage show strong correlation with higher Tobin's q (market valuation relative to asset value)
- Non-tech firms at this stage show the largest margin improvements — the less digitally automated your starting point, the bigger the gains
- Key metric: Would removing AI systems materially impact your P&L? If yes, you've arrived.
Phase 6: Continuous Maturity Assessment (Ongoing)
Objective: Prevent regression and adapt to evolving AI capabilities.
- Re-score against the maturity assessment quarterly
- Benchmark against sector-specific adoption rates (tech sector: 62% at Level 4-5; all others: significantly lower)
- Monitor the 79% workforce readiness gap — even Pacesetters need to continuously adapt
- Watch for the "botsitting crisis": employees saving 11 hours with AI but wasting 6.4 hours managing it
The Uncomfortable Truth: Why 89% Isn't Surprising
The MIT/CMU finding that only 11% of S&P 500 firms achieved deep AI integration shouldn't shock anyone who has watched technology adoption curves. The researchers themselves cite economic literature showing that gradual adoption is optimal when facing transformative technology risks. As Acemoglu and Lensman demonstrated, the adoption path should be slow and convex — accelerating only after increased certainty that catastrophic failure won't occur.
AI adoption has more than quadrupled since 2022, from 5% at the highest integration levels to 21% at Levels 4 and 5 combined. The acceleration is real. But the Nadella paradox applies: moving faster without organizational transformation just creates more expensive failure modes.
The CMU Software Engineering Institute and Accenture recognized this when they released their AI Adoption Maturity Model in June 2026 — a 63-page framework built from reviews of 100+ existing models, 600 practitioner surveys, and real-world pilots. Their core insight mirrors the MIT study: "True AI maturity is not measured by how much AI an organization deploys, but by its ability to build trustworthy and resilient capabilities, rigorous engineering practices, and governance approaches aligned with business outcomes."
Deployment is not adoption. Spending is not integration. And enthusiasm is not maturity.
What the Next 18 Months Look Like
The MIT/CMU data shows adoption accelerating — the curve is convex, exactly as theory predicts. Non-technology sector adoption is slowly picking up, with financial services leading the charge. Gartner projects that $234 billion in enterprise application software spend is exposed to "agentic arbitrage" by 2030 — meaning companies that don't reach deep integration will watch their software vendors be disrupted by those that have.
The competitive window is narrowing. Every quarter a company spends in the J-curve valley without a clear path to Phase 5 is a quarter where the 11% pull further ahead.
The question for every CXO reading this isn't whether to invest in AI. That decision was made years ago. The question is whether you're investing in the organizational transformation required to make that AI investment productive — or just adding to the $2.52 trillion being spent on technology that most companies can't yet prove is working.
The SEC filing doesn't lie. The survey might. Know the difference.
Rajesh Beri is Head of AI Engineering at Zscaler, covering enterprise AI strategy, security, and infrastructure. Subscribe for daily analysis of the decisions shaping enterprise AI.
