Why the Fed Says AI Productivity Is Still Missing-and Why That May Not Be Bad News

AI is visible in corporate talk, but not yet in the aggregate productivity number

The market is not debating whether AI is real. It is debating whether the economy has delivered the payoff investors expected by now. St. Louis Fed research scanning nearly 490,000 earnings calls found that AI-related productivity talk rose sharply, with roughly 15% of productivity discussion tied to AI by the end of 2025. But the same study found no measurable bump in aggregate productivity tied to AI so far. That gap is the pressure point.

Productivity is rising in some sectors, just not uniformly

The broader macro backdrop is not flat. U.S. labor productivity has grown at a 2.4 percent annualized rate since the beginning of 2024, above the 1.6 percent average in the five years before the pandemic. That strength is concentrated in information, finance and insurance, and professional and technical services-the three areas the Dallas Fed analysis identifies as most AI-exposed.

That does not prove AI has already lifted the whole economy. But it does support a more nuanced read than either "AI is working" or "AI is failing." The evidence is clearer on exposure and early sector gains than on a broad, measured productivity breakout.

The Fed's three-step framework fits a century-old pattern

The Federal Reserve's current framing helps explain why the macro data may lag the story investors have been buying.

Capabilities come first; productivity comes later

The Fed's monitoring roadmap starts with capabilities and costs, then moves to firm investment and adoption, and only later to productivity and labor outcomes. The note says that sequence maps onto the pattern often associated with general-purpose technologies. In that reading, investors may be watching a buildout phase with a scoreboard designed for the payoff phase.

Sector data shows where gains are appearing

AI-exposed industries have posted 3.7 percent average annualized productivity growth since the first quarter of 2024, versus 1.7 percent for the rest of the economy. The Dallas Fed study also finds a significant, positive relationship between AI exposure and productivity growth at the industry level. That is not the same as proof that AI alone caused every gain, but it is evidence that the biggest productivity beats are clustering where AI exposure is highest.

The key risk to the optimism is straightforward: if AI use keeps rising but sector productivity outperformance does not persist, the payoff case is weaker than the buildout narrative implies.

Why a real gain can still leave no mark in the data

One of the St. Louis Fed study's more important points is that some AI benefits may be hard to measure. The researchers warn that AI gains may be structurally invisible if AI makes certain output cheaper to produce, which can reduce the economic value of that output and offset the measured productivity gain.

That helps explain why companies can feel real efficiency improvements before the national accounts do. In the St. Louis Fed sample, about 95% of AI-related productivity sentences describe gains executives expect in the future, not gains already realized. So much of the current corporate evidence points to expected process gains-faster drafts, less rework, more automation-not yet to widely reported results.

Organization change also takes time. AI can speed the first draft, but many workflows still depend on human review, coordination, and decision-making. That friction can delay the translation of tool-level gains into economy-wide productivity.

What matters more for investors: who gets paid, and who keeps the value

The near-term edge is not spotting AI mentions. It is identifying who benefits first during the buildout and who can retain more of that benefit once adoption spreads.

Shovel sellers may win early; adopters may win later

The Fed's framework moves from capabilities and costs to firm investment and adoption, and only later to productivity and labor outcomes. That suggests compute, software, and infrastructure providers may show results before the broader economy does. But strong demand for AI tools does not, by itself, prove that downstream businesses are capturing lasting economic gains.

The harder question is whether a company can use AI to improve cycles or margins without making its output so abundant that pricing power erodes. That is why the possibility that some AI gains are structurally invisible matters for investors: a task can become easier while the market values the result less.

The watchpoint

The main check on the optimistic case is simple. The thesis weakens if AI use keeps rising but sector productivity outperformance does not persist. That would not end the AI story, but it would mean the economic payoff is weaker than the current buildout narrative suggests.