If AI Is Everywhere, Why Isn't Productivity?
AI now dominates earnings calls, capital spending plans, and boardroom strategy. And yet the surge isn't there. Not yet.
AI now dominates earnings calls, capital spending plans, and boardroom strategy. It is routinely described as a once-in-a-generation productivity breakthrough.
And yet, when you look at the data, something uncomfortable appears.
The surge isn't there.
Not yet.
That gap between narrative and measurement matters, because major decisions are already being made as if the gains have arrived.
The promise versus the numbers
At the macro level, productivity is not mysterious. It is measured, tracked, and historically responsive to real technological change.
According to data from the U.S. Bureau of Labor Statistics, nonfarm business labor productivity has grown modestly over the past two years, but nowhere near the kind of sustained acceleration that accompanied past general-purpose technologies like electrification or widespread computing.
A simple way to see this gap is through the BLS nonfarm business labor productivity chart, which tracks output per hour worked. Despite recent quarterly fluctuations, there is no sustained upward break from the pre-AI trend. The kind of step-change associated with past technological revolutions simply has not appeared.
Internationally, the picture is similar. OECD comparisons show productivity growth across advanced economies remains uneven and subdued, even as AI investment accelerates.
This does not mean AI lacks potential. It means the economy has not yet absorbed it in a way that shows up at scale.
Why firm-level success does not translate to macro gains
This is where the story often breaks down.
Individual companies are seeing localized benefits: faster code generation, quicker document drafting, more efficient internal workflows.
But productivity at the national level is not about isolated wins. It reflects broad diffusion, organizational redesign, and capital reallocation.
Research from the National Bureau of Economic Research has long shown that productivity gains from transformative technologies tend to arrive in phases: initial experimentation, organizational disruption, and delayed but meaningful macro payoff.
We are still early in phase two. Most firms are layering AI onto existing processes rather than restructuring work around it. That creates convenience, not transformation.
Measurement lags are real, but not infinite
Some argue the data simply hasn't caught up.
There is truth to that. Intangible investments, learning curves, and complementary changes take time to register. Economists have debated this issue since the early days of information technology, famously captured in Robert Solow's observation that computers were everywhere except in the productivity statistics.
But history also shows that when productivity turns, it does so visibly.
The post-1995 IT boom did not require interpretive generosity to spot. The data moved decisively.
We are not seeing that yet.
Incentives are pulling the story forward faster than reality
So why does the confidence sound so absolute?
Because the incentives reward optimism: executives benefit from forward-looking efficiency narratives, investors reward credible automation stories, vendors profit from urgency rather than patience, and policymakers want growth without inflation.
None of this requires dishonesty. It does, however, encourage premature certainty. AI becomes a future-tense justification rather than a present-tense outcome.
What a real productivity boom would actually look like
When AI productivity truly arrives at scale, we will not need to argue about it.
We will see sustained output growth without proportional labor increases. Measurable gains across multiple sectors, not just technology. Clear improvements in unit labor costs. Broad organizational redesign, not just tool adoption.
Until then, skepticism is not resistance. It is responsible interpretation.
The risk of declaring victory too early
Overstating productivity gains carries real consequences: workforce decisions made on expectations rather than results, policy errors driven by assumed growth, disillusionment when promised gains fail to materialize.
The history of technology adoption is not one of instant miracles. It is one of uneven progress, delayed payoff, and eventual — but earned — impact.
AI may well deliver on its promise.
The data simply has not confirmed it yet.
Not yet.