In 1987, economist Robert Solow quipped that "you can see the computer age everywhere but in the productivity statistics." Nearly four decades later, a landmark survey of 6,000 executives across the United States, United Kingdom, Germany, and Australia suggests we may be living through the same paradox — this time with artificial intelligence.
The study, reported by Fortune, found that roughly 90% of firms reported zero measurable impact on employment or productivity over the past three years — this despite corporations having poured more than $250 billion into AI infrastructure in 2024 alone. Nearly a quarter of respondents are not using AI at all.
Let that sink in for boards and investors: a quarter-trillion dollars deployed, and the macroeconomic needle has not moved.
Among firms that have adopted AI, average employee usage clocks in at a modest 1.5 hours per week. The forward projections are hardly more encouraging: executives surveyed forecast just a 1.4% productivity increase and a 0.8% output gain over the next three years, alongside a 0.7% reduction in employment. These are rounding errors, not the productivity revolution the industry has been selling.
The capital allocation problem
This is not primarily a technology story. It is a capital allocation story — and boards need to start treating it as one.
The mismatch between AI investment and AI returns mirrors, with uncomfortable precision, the original Solow Paradox. When computing investment surged in the 1970s and 1980s, productivity growth actually slowed — dropping from 2.9% annually during the 1948–1973 postwar boom to just 1.1% in the years that followed. The productivity payoff from computing eventually arrived, but it took roughly two decades and required businesses to fundamentally restructure workflows, not merely bolt new tools onto existing processes.
The optimists will make precisely that argument: the gains are coming, we just need to wait. And they may be right. But investors writing checks today need to price the timing risk honestly.
The academic consensus is sobering. MIT economist Daron Acemoglu projects just a 0.5% productivity increase over the next decade from AI adoption — a figure that would barely register against the capital being deployed. A Federal Reserve Bank of St. Louis study, published in 2025, found only a 1.9% excess cumulative productivity increase since the launch of ChatGPT. Against the backdrop of the MIT research that claimed AI could boost individual worker performance by 40% in controlled settings, the gap between lab and reality is vast.
Why the disconnect?
Several structural factors explain the divergence. First, most AI deployment to date has been additive rather than transformative — workers using AI assistants to draft emails faster, not fundamentally different business models. Second, the productivity gains tend to accrue at the individual task level, not the firm or economy level: a better email does not necessarily produce a better quarter.
Third, and perhaps most importantly for capital allocation decisions, the firms capturing most of the AI infrastructure spending — Nvidia, Microsoft, Amazon, Google — are not the firms that need to demonstrate productivity returns. Their customers do. And those customers are the 6,000 CEOs in this survey.
The questions boards should be asking
For executives reading this data, the right response is not to halt AI investment — the technology's long-term trajectory remains credible. The right response is to demand harder accountability from internal AI initiatives. Vague commitments to "transformation" and "competitive positioning" are not answers. Measurable targets tied to specific workflows are.
For investors, the survey is a useful reality check against the premium multiples currently embedded in AI-adjacent stocks. Markets have priced in a productivity revolution. The CEOs building with these tools have not yet seen one.
The numbers don't lie — but the investor deck sometimes does.
