In 1987, the economist Robert Solow wrote a sentence that got stuck in everyone’s head for forty years: “You can see the computer age everywhere but in the productivity statistics.”
He was talking about mainframes and personal computers. The technology was obviously transforming offices. You could see it. But national productivity data refused to move.
In February 2026, Fortune ran a story that made Solow’s paradox feel uncomfortably current. And the numbers behind it are hard to argue with.
The NBER Study That Dropped a Bomb
An NBER working paper surveyed nearly 6,000 CEOs, CFOs, and top executives across the US, UK, Germany, and Australia. The research team included scholars from Stanford, MIT, and the Bank of England — this wasn’t a LinkedIn poll.
The headline finding: 89% reported zero change in productivity from AI over the past three years. Not negative change. Not slight improvement. Zero.
Here’s what makes that number worse. About 70% of those firms were actively using AI. They’d adopted it. They’d deployed it. They’d spent money on it. And over 80% saw no measurable impact on either employment or productivity.
Two-thirds of CEOs personally use AI — but their average use is only 1.5 hours per week. That’s roughly 18 minutes a day. Less than most people spend on email before lunch.
The firms do expect AI will eventually deliver. Their forward-looking predictions: 1.4% productivity boost, 0.7% headcount reduction, and 0.8% output increase over the next three years. But those are predictions, not results. And predictions from the same executives who’ve already been using AI for three years with nothing to show for it.
It’s Not Just One Survey
PwC’s 29th Global CEO Survey — 4,454 CEOs across 95 countries — found 56% had seen no significant financial benefit from AI to date. Only 30% reported increased revenue. Only 26% saw lower costs. And just 12% experienced both.
Mohamed Kande, PwC’s Global Chairman, summarized it cleanly: “A small group of companies are already turning AI into measurable financial returns, while many others are still struggling to move beyond pilots.”
BCG’s September 2025 report put it more bluntly: 60% of companies generate no material value from AI despite investments. Only 5% create substantial value at scale.
RAND Corporation found that over 80% of AI projects fail — twice the failure rate of non-AI IT projects. Their interviews with 65 experienced data scientists pointed to the same top causes: misalignment on problem definition, bad data, technology-first thinking, and infrastructure gaps.
And Gartner predicted at least 30% of generative AI projects would be abandoned after proof of concept by end of 2025. Their analyst Rita Sallam: “Executives are impatient to see returns on GenAI investments, yet organizations are struggling to prove and realize value.”
The Solow Paradox, Round Two
Torsten Slok, Chief Economist at Apollo, has been saying this for months: “AI is everywhere except in the incoming macroeconomic data.”
No visible impact in employment data. No visible impact in productivity data. No visible impact in inflation data. For the S&P 493 — everyone except the Magnificent Seven — no signs of AI in profit margins or earnings expectations.
Meanwhile, 306 S&P 500 companies mentioned AI in their Q3 2025 earnings calls — the highest count in a decade. Everyone’s talking about AI. The financials aren’t listening.
Sound familiar? It should.
After Solow’s observation in 1987, economists spent decades arguing about why computers weren’t showing up in the macro data. The eventual answer: it took until the mid-1990s — roughly 25 years after widespread adoption of mainframes — for productivity to surge. When it finally came, the US saw a 1.5% increase in productivity growth from 1995 to 2005.
The optimists’ argument: we’re simply early. Give it time.
Why $405 Billion Isn’t Showing Up
Big Tech spent $405 billion in capital expenditure in 2025 — up 62% year-over-year. Amazon alone: $125 billion. Microsoft: $80 billion. Alphabet: $85 billion.
Goldman Sachs projects hyperscaler capex at $1.15 trillion from 2025-2027. More than double the 2022-2024 period.
So where’s the return?
A few patterns keep emerging from the research:
Most companies are using AI for basic tasks. EY’s 2025 survey of 15,000 employees found 88% use AI daily — but mostly for search and summarizing. Only 5% use AI in advanced ways that transform workflows. You don’t restructure an economy by having everyone summarize their emails slightly faster.
The deployment gap is real. McKinsey’s 2025 State of AI found 88% of organizations use AI in at least one function, but only 7% have fully scaled it. Two-thirds remain in experiment or pilot mode. And only about 6% qualify as “AI high performers” reporting more than 5% impact on EBIT.
Leadership barely uses the tools themselves. If CEOs average 1.5 hours a week with AI, they don’t understand what it can do. Gallup found a massive communication gap: 69% of C-suite executives believe their organization has communicated clearly about AI, but only 12% of entry-level staff agree.
The training gap cripples adoption. Only 14% of workers received any AI training from their employer last year, according to Google/Ipsos. 56% received no recent training of any kind, per ManpowerGroup. You can’t have an AI productivity revolution when 86% of your workforce is self-taught or untaught.
Where AI IS Working
Not everywhere is a wasteland. The counter-evidence matters.
Erik Brynjolfsson’s study of 5,179 customer support agents found AI increased productivity by 14% on average — and 34% for novice workers. Experienced workers saw minimal impact, but the floor came up dramatically. Published in the Quarterly Journal of Economics, this is the most cited positive evidence for AI’s workplace impact.
Deloitte’s 2026 enterprise survey found 66% of organizations reporting productivity or efficiency gains. But — and this is a big but — revenue growth from AI remains aspirational. 74% hope to grow revenue versus only 20% actually doing so.
And Brynjolfsson argues that 2025 productivity growth hit about 2.7% — roughly double the decade’s average. The BLS reported manufacturing productivity increased 3.7% in Q3 2025, the largest four-quarter gain since 2021. Some economists think this IS the early AI dividend showing up, even if CEOs aren’t attributing it to AI yet.
Penn Wharton models project AI will increase GDP by 1.5% by 2035, reaching 3.7% by 2075. The peak contribution? Just 0.2 percentage points per year in 2032.
The Electricity Lesson
When factories first adopted electricity in the 1880s, they literally bolted electric motors into the same layouts designed for steam engines. The productivity gain? Minimal. Factories needed 30-40 years to figure out that electricity enabled completely different architectures — single-floor factories, assembly lines, distributed power to individual machines.
The academics’ phrase for this: managers “overlaid one technical system upon a preexisting stratum.”
That’s exactly what most companies are doing with AI. They’re bolting chatbots onto existing workflows. Writing the same emails, just faster. Summarizing the same meetings they shouldn’t be having. Running the same reports with slightly different inputs.
Daron Acemoglu, the MIT economist who won the 2024 Nobel Prize, put it directly: “No worthwhile technology can just be sprinkled over a business like pepper. Think about what’s special about your company and your workforce — and use technology to amplify that, not to replace it.”
He estimates AI will increase GDP by 1.1% to 1.6% over the next decade. That’s roughly 0.05% per year. Not nothing. But not the revolution that $1.15 trillion in spending implies.
What This Means for You
If you’re a CEO who hasn’t seen results from AI yet, you’re in the majority. The 89% majority. That’s cold comfort, but it’s also context — you’re not uniquely bad at this.
If you’re investing in AI, the research suggests checking three things:
Are people actually using it? Not “do they have access” — are they using it weekly, across multiple tasks? The Google/Ipsos data shows that when organizations provide both tools AND guidance, workers are 4.5x more likely to become AI fluent.
Are you changing workflows or just adding AI on top? The electricity analogy holds. If your processes are the same as before AI, you’re bolting electric motors onto a steam factory.
Are leaders using it themselves? 1.5 hours a week isn’t leadership. It’s tourism. If your leadership team can’t demonstrate AI fluency, they can’t evaluate it, fund it, or set realistic expectations for it.
The Honest Take
Here’s what I think is happening. AI works — the evidence for individual task productivity is strong. But organizations haven’t figured out how to capture that productivity at scale. The gains evaporate into longer hours, scope creep, and meeting culture.
Solow’s paradox resolved itself eventually. The IT productivity boom of the late 1990s was real and massive. But it took reorganized companies, not just better computers.
Same thing here. AI will change the numbers. But not until companies change themselves.
The question isn’t whether AI can make work productive. It’s whether your organization is structured to let it.
And for most of the 6,000 executives in that NBER survey, the honest answer — three years in, hundreds of billions spent — is: not yet.