In 1987, Robert Solow observed that you could see the computer age everywhere but in the productivity statistics. In 2026, six thousand executives reported that AI had produced no measurable impact. The paradox is the same. The response is not. This time, instead of waiting for the metric to vindicate the investment, companies are replacing the metric.
In 1987, Robert Solow wrote a line that became the most cited sentence in technology economics: you can see the computer age everywhere but in the productivity statistics. He was observing a paradox. Companies were spending billions on computers. The spending was accelerating. The productivity gains were not showing up.
Nearly four decades later, researchers affiliated with the National Bureau of Economic Research surveyed six thousand executives across the United States, United Kingdom, Germany, and Australia. Ninety percent reported that AI had produced no measurable impact on their company's productivity or employment over the previous three years. The Solow Paradox is back. The numbers are worse.
The Ledger
Four companies — Alphabet, Amazon, Meta, and Microsoft — committed six hundred and fifty billion dollars to AI infrastructure in 2026. Including Oracle's announced spending, the figure approaches seven hundred billion. By mid-March, some analysts revised the five-company total upward past seven hundred billion based on updated company guidance.
On the other side: only five percent of enterprises report substantial returns on their AI investments, according to converging estimates from MIT research on generative AI pilots, CIO surveys, and multiple industry analyses. S&P Global found that forty-two percent of companies scrapped most of their AI initiatives in 2025 — up from seventeen percent the year before. PwC's latest AI Predictions survey found that just twelve percent of enterprises achieved both increased revenue and decreased costs from AI. Seventy-seven percent of organizations deploying AI cannot prove whether it delivers value at all.
Seven hundred billion dollars flowing in. Five percent seeing returns flowing out. Forty-two percent actively retreating. The gap is not new — The Demethylation documented it three weeks ago when fifty-six percent of CEOs reported zero financial returns. What is new is the institutional response.
The Move
In 1987, companies kept the metric and waited. The IT productivity paradox eventually resolved. Between 1995 and 2000, total factor productivity surged as organizations reorganized around networked computing. Erik Brynjolfsson tracked the numbers through the drought and documented the harvest when it arrived. The measurement framework survived intact. The patience was rewarded.
In 2026, companies are doing something different. They are changing the metric.
Gartner's Mary Mesaglio introduced a framework at the IT Symposium that explicitly moves enterprise AI measurement "beyond ROI." The framework has three components. ROI — traditional financial returns. ROE — Return on Employee, measuring engagement and personal productivity tools. And ROF — Return on the Future, measuring whether AI positions the company to rewrite its industry's rules.
Return on the Future cannot be falsified. The future has not arrived yet. A company that reports zero current productivity gains from AI can claim it is investing in strategic optionality — and no dataset available today will contradict the claim. If the future arrives and the investment paid off, the framework is vindicated. If the future arrives and the investment did not pay off, the company will already be measuring something else.
Fortune 500 leaders are adopting similar structures. Four-pillar measurement frameworks replace traditional return on investment with four simultaneous metrics: Efficiency, Revenue, Risk Mitigation, and Business Agility. Each additional pillar dilutes the denominator. When productivity alone is the metric, the verdict is uncomfortable — ninety percent report no impact. When you add strategic optionality, risk mitigation, and business agility to the scorecard, the same investment portfolio transforms into prudent long-term positioning.
This is not fraud. It is not necessarily wrong. It is the institutional reflex when a measurement regime produces answers the institution does not want to hear.
The Difference
The 1987 paradox and the 2026 paradox share the same observation: massive technology spending with no visible productivity return. They differ in one structural way that determines whether anyone will ever know if the investment paid off.
Solow's paradox resolved because someone was still measuring. Brynjolfsson and his colleagues kept the productivity metric intact through the 1990s. When the surge arrived — driven by companies that had finally reorganized their workflows around IT rather than layering IT on top of existing workflows — the metric detected it. The causal link between investment and return was temporarily obscured but eventually revealed.
When you change the metric during the paradox, you lose the ability to detect the resolution. If AI eventually produces the productivity gains its proponents promise, the companies that shifted to Return on the Future will not be able to distinguish between "our AI investment paid off" and "our business improved for other reasons while we happened to be spending on AI." The measurement framework that would reveal the causal link has been replaced by one that cannot.
The S&P Global number — forty-two percent scrapping AI initiatives, up from seventeen percent — represents companies still running the honest test. They tried AI, measured the results against the original success criteria, and stopped when the results were not there. That is the paradox operating as designed: a diagnostic that tells you which investments work and which do not.
The Gartner framework represents companies that opted out of the test. Not by stopping the investment. Not by reducing the spending. By redefining success so the investment cannot fail.
The most expensive question in technology right now is not whether AI works. It is whether the institutions spending seven hundred billion dollars on it would know if it did not.
The forty-two percent who scrapped their initiatives will learn. They ran an experiment, collected data, and acted on it. Some will try again with better approaches and succeed. Others will reallocate the capital. Either way, the measurement did its job.
The companies that shifted to Return on the Future will spend the same seven hundred billion and never learn — because the framework they adopted is structurally incapable of producing a negative result. Strategic optionality is always positive in prospect. Risk mitigation is always justified after the fact. Business agility is always better than business rigidity. None of these metrics can return a value that says stop.
Robert Solow's paradox taught us that transformative technologies take time — and that the patience to keep measuring through the drought is how you eventually detect the rain. The 2026 paradox may teach us something different: that the institutions funding the transformation have learned to make the question unanswerable before the answer arrives.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)