The Innovation Paradox: Why Speeding Up Innovation Can Actually Slow Economic Growth

Silicon Valley loves to paint a picture of AI as the next silver bullet – a boundless source of abundance that will lift every bottom line. Yet, when you sit down with CFOs from the world’s biggest companies, the story looks a lot less glittery. In McKinsey’s 2026 State of AI survey, a whopping 88 % of them answered, “Not so much.”

We’ve seen this before. Remember the hype surrounding the iPhone in 2007? Or the boom‑time promises of broadband as the engine of a “New Economy”? Even the personal computer, once hailed as a productivity miracle, left economists like Robert Solow shaking his head, noting that you could see computers everywhere except in the productivity numbers.

What’s happening is what I like to call the Innovation Paradox: the faster the tide of new technology rises, the slower the ship of economic growth seems to move.

One way to picture it is the double “K‑shape” of technology diffusion. The first K‑shape separates consumers from enterprises. Consumers jump on smartphones at breakneck speed – 80 % of Americans were hooked within seven years. Enterprise resource‑planning (ERP) systems, by contrast, took a good 33 years to reach just over half of U.S. firms. AI follows a similar split: three years after ChatGPT arrived, 53 % of U.S. adults were already using consumer‑grade AI, but only about 10 % of companies have deployed it in ways that truly boost productivity.

The second K‑shape lives inside the corporate world itself. McKinsey’s latest survey shows that while almost every firm dabbles with AI in at least one department, only 6 % can claim they’re “high performers” – that is, achieving a 5 % lift in EBIT or more. That tiny slice is not missing tech; it’s missing transformation. Those winners are three times more likely to have completely re‑engineered their workflows, yet only about one‑fifth of all adopters have taken that step.

Even the stock market mirrors the divide. AI‑centric names now make up roughly 45 % of the S&P 500’s market cap, and over the past three years the index has returned 76 %. Strip out the AI heavyweights and you’re left with a modest 32 % gain. The numbers speak loudly.

Look back at electrification and the telephone boom in the early 20th century. Those technologies reached both consumers and factories at comparable speeds, and they reshaped how work was organized – think of GM’s multidivisional structure or Taylor’s scientific management. The diffusion was deep, not just a surface‑level adoption.

Today’s economy, however, is built on services. Hospitals, schools, government agencies – they don’t have the same supply‑chain linkages as a steel mill. William Baumol warned us decades ago that “cost‑disease” sectors, which rely heavily on human interaction, simply can’t boost productivity the way manufacturing can. Those Baumol sectors now account for about half of GDP in advanced economies, and they stubbornly resist the efficiencies that technology promises.

So why does enterprise AI stall? It isn’t a lack of algorithms. It’s the weight of organizational overhead. I’ve been tracking a metric I call the Combined Overhead Ratio – a blend of government spending, corporate SG&A, and compliance costs. In the U.S., that ratio cracked 47 % of GDP around the turn of the millennium. Once you’re above that line, sustainable growth rarely tops 2.5 % a year – roughly half the pace we saw when overhead hovered below 35 % in the 1960s.

Government outlays now sit at 36 % of GDP, corporate SG&A has doubled since the ’80s, and firms spend north of $2 trillion annually just to stay on the right side of regulation. My own dive into S&P 500 data from 1985‑2025 shows that in 73 % of sector‑year combos, SG&A was climbing while revenue growth was slowing – a classic “overhead trap.” As Jack Dorsey has quipped, corporate hierarchies are an outdated way to route information. Most AI projects today feel like swapping a faster engine into a horse‑drawn carriage.

AI does differ from earlier digital waves. Earlier tech made it easier to process, move, or display information. AI, by contrast, can reason, decide, and even create. That matters because the toughest service sectors aren’t data‑starved; they’re judgment‑starved. Diagnosis, evaluation, personalization – those are exactly the areas where AI can lend a hand.

Yet the same old barriers reappear. I’ve watched a consumer brand crush its online sales with digital tools, only to hit a wall when its offline distributors refused to share customer data. In healthcare, getting doctors to click a new app is one thing; redesigning the whole clinical workflow to integrate that app is another beast altogether – tangled in data‑governance rules, compliance hurdles, and the power dynamics of medical professionals.

These aren’t outliers; they’re the rulebook for most firms trying to ride the AI wave. The paradox isn’t that AI is ineffective – it’s that without deep, organization‑wide redesign, the technology ends up as a flashy add‑on, not a growth engine.

Bottom line: faster innovation alone won’t fix the slowdown. We need to unclutter the overhead, re‑think how work gets done, and let AI slip into the very fabric of enterprises – not just sit on a dashboard.