88% of Firms Use AI, Only 6% See Real Profits From It

McKinsey's 2026 State of AI survey found that 88% of organizations have deployed AI in at least one function — yet only 6% qualify as "high performers" with more than 5% EBIT impact, a figure unchanged from 2025. That 82-point gap is the core of what Fortune contributor Ming Lu calls the Innovation Paradox: the faster we innovate, the slower we grow.

The argument, laid out in a Fortune.com commentary, is that enterprise AI adoption has stalled not because the technology fails, but because institutions fail to adapt.

What do the numbers actually show?

The diffusion data splits into what Lu describes as a double K-shaped pattern. The first divide runs between consumers and enterprises:

• Smartphone adoption reached 80% of Americans in seven years; enterprise resource planning systems took 33 years to reach 57% of U.S. firms.
• Consumer AI usage has hit 53% of U.S. adults barely three years after ChatGPT's launch.
• Enterprise AI in productive deployment stands at just 10%.

The second divide sits inside the enterprise. The 6% of AI high performers are three times more likely to have redesigned workflows end-to-end, according to the McKinsey survey. Only 21% of all adopters have done so.

The stock market reflects the divergence. Thirty-six S&P 500 AI companies now represent 45% of the index's market capitalization. Over three years, the S&P 500 returned 76%; strip out AI stocks and the figure drops to 32%.

Why hasn't fast adoption produced growth?

Lu points to history. When the PC colonized every desk, economist Robert Solow quipped that you could see the computer age everywhere but in the productivity statistics. Over five decades, waves of digital innovation have delivered spectacular consumer adoption alongside continuously declining productivity growth.

Electrification and early telecommunications succeeded where digital waves failed, Lu argues, because they diffused both deep and broad. Electric motors forced factories to abandon rigid layouts for continuous-flow production; telephones collapsed coordination costs; organizational redesigns followed, from Sloan's multidivisional structure at General Motors to Taylor's scientific management.

Digital technologies landed in a fundamentally different economy. Economist William Baumol identified the constraint: "stagnant" service sectors resist productivity improvement because they depend on human interaction. These Baumol sectors — healthcare, education, public administration, construction — account for roughly 50% of advanced-economy GDP.

What is the overhead trap?

Lu's cross-country metric, the Combined Overhead Ratio, captures government spending, corporate SG&A and regulatory compliance costs. In the U.S., the ratio crossed 47% of GDP around 2000. Above that threshold, GDP growth has not sustainably exceeded 2.5% — roughly half the rate achieved when overhead remained below 35% in the 1960s.

The supporting figures:

• Government spending has surged to 36% of GDP.
• Corporate SG&A has doubled since the 1980s.
• The Competitive Enterprise Institute estimates U.S. firms spend more than $2 trillion annually on compliance alone.

His within-sector analysis of S&P 500 companies from 1985 to 2025 shows 73% of sector-decade observations fall into the trap: SG&A rising while revenue growth declines.

Jack Dorsey of Block has argued that corporate hierarchy is an obsolete information-routing protocol. Lu's verdict on most current AI deployments is blunter: companies are putting a faster engine in a horse-drawn carriage.

Is AI different this time?

Yes, categorically, Lu writes. Previous technologies made it easier to process, transmit or display information. AI reasons, decides and creates — and Baumol sectors are not information-scarce but judgment-intensive, requiring diagnosis, evaluation and personalization, precisely the capabilities AI can augment.

The barriers, however, remain. Drawing on more than 40 years as a corporate executive and private equity partner, Lu cites two cases. At one consumer company, digitization delivered strong online-channel growth but failed to scale across the larger offline business because independently owned distributors refused to cede customer data to an integrated platform. In healthcare, introducing digital tools to doctors proved straightforward; end-to-end digitization of clinical workflows proved fraught, tangled in data governance, compliance and the asymmetric power of medical professionals over management.

Writer's 2026 enterprise AI survey, conducted with Workplace Intelligence across 2,400 global leaders, confirms the pattern at scale: 79% of organizations report challenges in AI adoption — a double-digit increase from 2025 — and 54% of C-suite executives admit AI adoption is "tearing their company apart."

What matters: depth and breadth, not pace

ChatGPT's consumer adoption was the fastest of any consumer technology in recorded history. It has moved no macroeconomic needle, Lu writes. What matters is whether AI penetrates core operations and redesigns workflows — depth — and whether that transformation extends beyond superstar firms into the sectors that dominate employment and GDP — breadth.

Historical evidence from electrification to digital waves supports an adoption timeline of over ten years, often decades. The technology progresses faster than any predecessor, but the organizational redesign required to absorb it proceeds at the pace of human institutions: slowly, painfully, and against fierce resistance. The Innovation Paradox persists, Lu concludes, not because technology fails to advance, but because institutions fail to adapt — and society will see real impact only when AI is adopted deeply within enterprises and broadly across large economic sectors.