AI Productivity Gap Widens as Layoffs Climb

Corporate AI spending is at record levels, yet roughly 90% of business leaders say the technology has not lifted productivity at their companies. That finding comes from a Federal Reserve Bank of Atlanta survey released this month. It sits awkwardly beside a wave of AI-cited job cuts at Salesforce, Amazon, and Klarna. The result is an AI productivity gap that capital spending has not closed. New research suggests the layoffs themselves may be widening the AI productivity gap.

That disconnect is now being acknowledged at the top of companies. A National Bureau of Economic Research working paper built on responses from nearly 6,000 CEOs and CFOs lands on the same number: nine in ten report no productivity gains from AI. In the same Atlanta Fed survey, 80% of executives said AI investment has yet to change their headcount or improve productivity. Related survey work found roughly two-thirds of executives saying their organizations are not using AI in the workplace at all.

The AI productivity gap, measured by executive surveys

The 90% figure measures executive perception, and perception is not the same as output. The surrounding evidence points the same way: productivity gains recorded in the U.S. since 2021 may owe more to remote work and to downsizing in technology and other sectors than to AI adoption. Follow-up surveys put the share of companies reporting no AI effect on jobs or output over a three-year period at close to 90%. The same research program examined how stock markets react to AI-linked layoff announcements, a test of whether investors reward the trade-off before any productivity data arrives.

The layoffs themselves continue at pace. Challenger, Gray & Christmas data shows AI cited in roughly 17% to 26% of layoffs, while tech-sector reductions have averaged more than 1,100 workers per day in 2026. Heavy AI investment, frequent AI-cited cuts, and no measured productivity return together capture the AI productivity gap that executives now concede in public.

The AI productivity gap in corporate records

The gap also appears in corporate records, alongside survey responses. An analysis of five years of U.S. public company records finds a consistent correlation: as AI investment announcements rise in frequency, so do layoff announcements attributed to AI. The dataset behind that finding spans millions of employee job-satisfaction reviews, thousands of financial reports, and hundreds of AI investment and layoff announcements. The researchers read the pattern as a strategy that treats workforce reduction as an integral part of the AI program itself. The correlation offers an early read on the AI productivity gap before earnings reports or productivity statistics arrive.

Why layoffs can widen the AI productivity gap

The strongest counterargument to AI-driven cuts concerns employee sentiment. Job cuts damage how remaining workers view AI, and attitude toward AI is one of the strongest predictors of firm productivity once the technology is actually in use. If that relationship holds, cutting jobs to fund AI may erode the trust and engagement the technology needs to deliver measurable gains. That is a direct route from headcount decisions to a wider AI productivity gap.

The mechanism runs through daily work. Employees who fear replacement are less likely to integrate AI into their workflows, report problems early, or push tools to their limits. Those behaviors determine whether the technology changes measured output. The researchers argue that job insecurity converts a tools rollout into a survival contest, which is why the promised gains stay out of reach.

The logic of the layoff case rests on a specific assumption: that AI makes remaining teams productive enough to justify smaller headcounts while lowering costs. The evidence for that assumption is mixed. Mercer's research finds executives regard AI as central to future performance, yet many organizations are struggling to redesign work around it. A large Danish study linking individual AI adoption to firm outcomes found that adoption restructures work, and restructuring does not automatically produce the productivity lift leaders say they are waiting for.

The trade-off executives are making

On paper, the strategy is coherent: fund AI aggressively, cut costs to pay for it, and let efficiency gains arrive later. The surveys suggest the sequence is inverted. The cost cuts arrive now, while the gains remain unmeasured at nine in ten companies. For boards and investors, the AI productivity gap frames the open question: will AI-linked layoffs be judged by the savings they produce today or by the productivity they foreclose tomorrow?

Executives effectively have three options. They can cut first and measure later, which is the current pattern. They can hold headcount and redesign work around AI, which Mercer's data suggests most organizations are not equipped to do yet. Or they can wait for evidence before committing headcount decisions, which risks falling behind competitors on cost. Each option carries a real cost, and none of them is validated by the productivity data executives themselves report.

Two readings of the same data are available, and the difference matters for strategy. The optimistic reading borrows from the IT era: the gains are real but delayed, and companies that cut now will look disciplined when the payoff arrives. The skeptical reading is that the current cycle differs in one structural way. The cuts are happening before the gains, and the sentiment evidence says that sequence destroys the conditions for the gains. Executives can believe either version; the surveys say they cannot yet prove either.

The current moment has also drawn comparisons to the information-technology paradox of the 1980s. Years of heavy IT investment produced no visible productivity payoff before eventually showing up in aggregate statistics. The uncomfortable difference is that the earlier cycle did not require cutting the workers who would use the technology as a precondition for funding it. That difference is where the sentiment research becomes strategically important: the mechanism that is supposed to produce the payoff may be the mechanism being dismantled.

The accountability question follows directly. Layoff savings are visible within a quarter, while the productivity claim behind them is not testable for quarters, and the sentiment damage appears only in internal surveys. That asymmetry creates an incentive to announce AI-driven cuts and move on. That is why the correlation between investment announcements and layoff announcements matters as a leading indicator of the AI productivity gap rather than as evidence of results.

The verdict: measure the AI productivity gap

The evidence supports treating AI-linked layoffs as a testable hypothesis rather than a settled strategy. Concretely, that means tracking employee sentiment alongside headcount, tying AI rollout to work redesign instead of workforce reduction, and treating the absence of measured gains as a signal to change course rather than to double down. Companies that cite AI in layoff decisions should expect the same evidentiary scrutiny investors apply to any cost-saving claim. The AI productivity gap will be settled by results rather than announcements.

The metrics to watch are specific: productivity figures in earnings reports, sentiment measures in engagement surveys, and the pace of AI-cited layoffs relative to measured gains. Until those numbers move, the AI productivity gap will remain open, and the burden of proof sits with the companies making the cuts rather than with the workers being asked to compensate for them.

Why this matters

The AI productivity gap is now a documented executive consensus rather than a fringe claim. For the companies funding this cycle, the stakes are concrete: if sentiment drives AI-era productivity, each layoff justified by AI carries a cost in the very gains it is meant to secure. Boards, investors, and workers all have reason to demand productivity numbers alongside the next round of AI announcements.

✔Human Verified

Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.