Ming Lu diagnoses the innovation paradox of the AI era

Ming Lu, who spent more than 40 years as a corporate executive and private equity partner, diagnosed the innovation paradox of the AI era in a commentary published on October 11 in a U.S. business publication. He said that over the past 50 years digital innovation achieved spectacular speed in consumer adoption while productivity growth kept declining, and he called this phenomenon the innovation paradox. His definition fits in one sentence: "The faster we innovate, the slower we grow."

The starting point is how executives feel. Citing McKinsey's 2026 State of AI survey, Lu wrote that 88% of chief financial officers (CFOs) at the world's largest companies answered "not so much" when asked whether AI is transforming their bottom line. In the same survey, 88% of organizations use AI in at least one function, but high performers achieving more than 5% impact on earnings before interest and taxes (EBIT) stood at just 6%, a share unchanged from 2025.

Lu explains the gap as a double K-shaped diffusion. The first K-shape lies between consumers and enterprises. Smartphones reached 80% of Americans in seven years, while enterprise resource planning (ERP) systems took 33 years to reach 57% of U.S. firms. Consumer AI usage rose to 53% of U.S. adults three years after the launch of ChatGPT, yet enterprise AI in productive deployment stands at 10%. The second K-shape plays out inside the enterprise. Lu saw the 82-point gap between the 88% of adopters and the 6% of high performers not as a technology deficit but as an organizational transformation deficit. High performers are three times more likely to have redesigned workflows end to end, and only 21% of all adopters have done so.

The stock market has the same shape. According to Lu, 36 AI companies in the S&P 500 account for 45% of the index's market capitalization. The S&P 500 returned 76% over the past three years, but excluding AI stocks the figure falls to 32%. METAL has reported that an EY chief economist warned the AI productivity boom will create a winner-takes-all economy, and Lu's numbers show that concentration is already priced into the market.

The comparison is electrification in the early 20th century. Electricity and telecommunications spread to consumers and enterprises at a similar pace and a similar depth. Electric motors moved factories from rigid layouts to continuous-flow production, and telephones lowered coordination costs. Organizational redesigns followed, such as General Motors' multidivisional structure and Taylor's scientific management. Digital technology entered an economy dominated by services. Healthcare, education, public administration and construction, the sectors economist William Baumol identified as resistant to productivity gains because they depend on human interaction, account for about 50% of advanced-economy GDP.

The backbone of Lu's diagnosis is overhead. He built a metric called the Combined Overhead Ratio that bundles government spending, corporate selling, general and administrative expenses (SG&A) and regulatory compliance costs. In the U.S. the ratio crossed 47% of GDP around 2000, and GDP growth has not sustainably exceeded 2.5% since. That is about half the rate of the 1960s, when overhead stayed below 35%. Government spending has risen to 36% of GDP, and corporate SG&A has doubled since the 1980s. The Competitive Enterprise Institute (CEI) estimates that U.S. firms spend more than $2 trillion a year on compliance alone. When Lu split S&P 500 companies from 1985 to 2025 by sector and decade, 73% of observations fell into a trap of rising SG&A and declining revenue growth. "Most companies deploying AI today are putting a faster engine in a horse-drawn carriage," he wrote.

That does not mean Lu puts AI on the same line as earlier technology. Earlier technologies made information easier to process, transmit and display, whereas AI reasons, decides and creates. Baumol sectors are not short of information but heavy in judgment such as diagnosis, evaluation and personalization, so he saw ample room for AI to contribute. The barriers remain. At one consumer company Lu worked with, independently owned distributors refused to cede customer data to an integrated platform, so online-channel gains failed to spread into the much larger offline business. In healthcare, giving doctors digital tools is easy, but digitizing clinical workflows end to end is tangled in data governance, compliance and the power of medical professionals.

Surveys from the field point the same way. The 2026 enterprise AI adoption survey that enterprise AI platform company WRITER released on April 7 with research firm Workplace Intelligence covered 2,400 knowledge workers (1,200 executives and 1,200 employees) in the U.S., the U.K., Ireland, Benelux, France and Germany from December 17, 2025 to January 25, 2026. Of the executives, 79% said they face adoption challenges such as lagging ROI, strategy gaps and internal power struggles, and 54% said AI adoption is tearing their company apart. Leaders reporting significant ROI stood at 29% for generative AI and 23% for AI agents. According to the WRITER release METAL reviewed, 69% of executives said their company is doing layoffs due to AI, yet 39% admitted they have no formal strategy to drive revenue from AI. "Layoffs are not a viable AI strategy," said May Habib, CEO of WRITER, adding that leaders who radically redesign operations with human-agent collaboration at the center build advantages competitors cannot replicate.

Lu's conclusion is depth and breadth over pace. ChatGPT was the fastest-spreading consumer technology on record, yet it has not moved macroeconomic indicators, he wrote. What must be examined is whether AI enters core operations and changes workflows, and whether that change spreads beyond a few superstar firms into the sectors that make up a large share of employment and GDP. History from electrification to digital shows technology adoption took more than ten years and often decades. Technology advances faster than ever, but the organizational redesign needed to absorb it moves at the speed of human institutions. By Lu's diagnosis, the key to resolving the paradox lies not in model performance but in the company org chart.