The Fed's 95% AI Warning: Gains May Be Real, but Investors Shouldn't Expect a Quick Payoff

The data still shows a waiting game

The market is trading the promise, not the receipt. A recent St. Louis Fed study found no measurable aggregate productivity increase attributable to AI after reviewing about 490,000 earnings call transcripts from 5,198 firms. More importantly, about 95% of AI-related productivity discussion describes gains executives expect in the future, not gains companies have already secured. That leaves investors pricing a payoff before the evidence is fully there.

That does not mean the excitement is empty. Companies discussing AI positively have also increased R&D, capital spending, and broader investment, suggesting the enthusiasm is backed by real financial commitments. But investment is the buildout phase, not the harvest. History and recent Fed research suggest investors should expect a long lag between adoption and broad productivity results.

Why AI gains may stay invisible longer than expected

Cheaper output is not the same as richer output

There is a deeper reason the AI payoff may remain hard to see: when technology makes output cheaper to produce, that output can also become less valuable because it is more abundant. AI is making some output radically cheaper to produce, and the research warns that productivity gains can be offset when abundance reduces pricing power. In practical terms, a company may generate much more copy, code, design work, or support content, but if that output becomes the new baseline, revenue and profits do not automatically rise with it.

This is the key risk for investors. Productivity can improve in the spreadsheet without improving the business model. If AI mainly lowers the cost of producing commoditized work, the efficiency gain may be real but still too small to justify today's expectations.

Investment is visible; broad results are not

That helps explain the gap between market excitement and macro data. Recent Federal Reserve analysis frames the AI rollout as a sequence: capability improvements and cost declines come first, then firm adoption and investment, and only later do measurable aggregate productivity gains become obvious. Under that framework, the absence of an economy-wide breakout is not proof that AI is ineffective. It may simply mean the economy is still early in the rollout.

Where the signal is strongest-and where it is still weak

If aggregate data are too blunt, a better starting point is the sectors most exposed to AI. Since the start of 2024, labor productivity in the three most AI-exposed sectors has grown at a 3.7 percent annualized rate, versus 1.7 percent for the rest of the economy. That suggests benefits are showing up first where AI exposure is highest.

But the evidence still has limits. The same analysis finds a positive relationship between AI exposure and productivity growth, not definitive proof that AI alone caused the improvement. The researchers explicitly raise the possibility that other advantages already present in those sectors are also at work.

For investors, that distinction matters. AI may be real and still not yet broad enough, fast enough, or profitable enough to settle the debate.

What to watch as the AI buildout continues

The next few quarters are best viewed as execution, not prophecy. The more useful framework is infrastructure-style staging: first capability improvements and cost declines, then adoption and investment, and only later measurable productivity gains.

Signals worth monitoring

• Profit margins: Are AI gains showing up as better profitability, or mainly as cost control that helps firms stay competitive?
• Sector leadership: Are the most AI-exposed sectors still pulling ahead in productivity versus the rest of the economy?
• Pricing power: Can companies monetize AI output without the market treating it as cheap and abundant? That is the risk behind the idea that AI can make output less valuable even as it becomes cheaper to produce.
• Business quality: Is AI helping firms deepen customer workflows, improve outcomes, or raise output quality rather than simply speed up routine tasks?

What would strengthen the bull case

Two changes would make the optimistic case more credible.

First, management commentary would need to shift. Right now, about 95% of AI-related productivity discussion describes gains executives expect in the future. The cautious view weakens if companies begin reporting more present-tense results they have already achieved.

Second, the gains would need to spread. The clearest productivity signal is still concentrated in AI-heavy sectors. A stronger bull case needs evidence that those benefits are broadening beyond them and showing up in the wider economy.