AI Job Cuts Are Booking a Cost That Won’t Appear Until 2029

By 2029, up to 30% of workers laid off because AI replaced them will need to be rehired, and often at a premium over what they were paid before. That is the projection Gartner released this month, and it turns a story usually told as a warning into something more precise: an accounting problem.

The savings companies book today when they cut headcount for AI already carry a liability attached. Most finance teams have not modelled it. Gartner’s September 2026 guidance, presented by VP Analyst Tori Paulman, names four strategic shifts that separate firms building durable AI advantage from those quietly accumulating workforce debt. The timing is not incidental. AI-attributed job cuts have accelerated sharply, with 55,000 U.S. jobs explicitly linked to AI in 2025 alone, more than twelve times the figure from two years earlier.

Here is the practical lens this analysis gives you. After reading it, you will be able to tell which management behaviours signal a company is using AI to compound competitive advantage, and which signal it is trading long-term capability for a single near-term earnings line.

The real cost of AI downsizing that most earnings calls do not mention

An earnings call that announces AI-driven payroll savings almost never announces the cost that comes with them. The saving is immediate and quantifiable. The cost is delayed, diffuse, and lands in a future quarter that no one is discussing yet.

Start with the rehiring figure. Gartner projects that by 2029, up to 30% of AI-displaced workers will be brought back, frequently at higher cost than their original employment. For large enterprises specifically, that rate climbs to roughly 40%. These are not workers returning to the same salary; they return to a tighter, more expensive market for the exact skills the company let go.

Then add what it costs to rebuild a role. Workforce research puts the replacement cost of a departed knowledge worker at 50% to 200% of their annual salary, with 8 to 12 months required for a new hire to reach full productivity. The knowledge that leaves is harder to price. Research by Atlan estimates institutional knowledge loss costs U.S. companies $1.3 trillion annually.

Now the arithmetic starts to assemble itself.

The regret signal: Data analysis by Digital Applied in 2026 found that 55% of companies that cut jobs for AI reported regretting the decision, citing quality degradation, morale damage, and knowledge loss that outweighed the initial savings.

Here is what this means for you as an investor. A company reporting AI-driven savings in 2025 may be setting up a visible margin drag in 2027 or 2028. When you evaluate an AI-heavy business model, the question is whether those future costs have been provisioned for, or whether a headline saving is being sold to you as a durable margin improvement when the institutional knowledge needed to execute at quality has already walked out the door.

Why Gartner’s four shifts reframe the AI-workforce question entirely

Gartner’s four shifts are usually read as a best-practice checklist. Read that way, they lose their force. Each shift is actually a corrective to a specific way the cost-cutting approach fails, and the logic accumulates when you take them in sequence.

1. Expanding human capability through human-AI collaboration. This corrects the assumption that AI can fully substitute for human judgment, accountability, and relational work. The remedy is teamwork that augments employees while keeping humans answerable for outcomes.
2. Building an AI-ready, adaptive workforce. This addresses the skills gap that opens when a company automates before it upskills, leaving the staff who remain unable to operate effectively in the changed environment.
3. Preserving context, judgment, and institutional knowledge. This targets organisational amnesia directly, where decision quality degrades even though the AI works exactly as designed.
4. Creating compounding long-term value from AI. This reframes AI return from a one-time cost event into a capability that gets faster, cheaper, and lower-risk with each deployment cycle.

The stakes attached to getting this wrong are not distant. Gartner forecasts that by 2027, roughly 75% of organisations focused on capturing AI productivity gains as cost savings will fall behind competitors that channel those same gains into innovation, modernisation, and upskilling instead. 2027 is barely fifteen months from this research. The divergence Gartner describes is not a forecast to monitor; it is already underway and visible in how companies are positioned today.

Preserving what AI cannot replicate

Shift three is the subtle one, because it describes a failure that occurs even when the technology succeeds. The AI can complete its assigned task correctly and the organisation can still make worse decisions, because the human context needed to interpret the AI’s output has been removed.

Workers need to understand not just how a process runs but why it runs that way. Gartner points to conversational user interfaces, decision intelligence platforms, and generative UI as enablers here. Each attacks the same problem from a different angle: keeping human oversight and contextual judgment inside the loop rather than automating them out of it.

Gartner’s emphasis on preserving human oversight and contextual judgment connects directly to the domains AI cannot yet absorb that Howard Marks identified after reversing his AI scepticism: genuine novelty, qualitative assessment of people and incentives, and decisions shaped by having real capital at risk, precisely the functions that disappear when automation removes humans from the loop entirely.

Building compounding returns rather than one-time savings

Shift four is where the returns profiles of two strategies visibly separate. In the cost-event model, AI delivers a single payroll saving and the benefit stops there. In the compounding model, each successive deployment becomes progressively cheaper and lower-risk because the organisation has built the foundation to reuse what it has already learned.

Gartner names AI-powered wearables, domain-specific generative AI models, embodied AI, and vibe coding as technologies driving this shift. The point is not the tools themselves. It is that firms fixated on shift one as a cost event are accumulating the liabilities described earlier, while those articulating strategy around shifts two through four are building advantage that widens over time. That distinction is your audit framework for reading management commentary.

Klarna and Accenture as the two templates every investor should know

Two companies started from the same place: using AI to cut a workforce. They ended somewhere very different, which makes the pair close to a natural experiment.

Take Klarna first, the cautionary case. The logic was internally coherent and the early metrics looked good. The Swedish fintech replaced roughly 700 customer support workers with AI, froze hiring, and contracted headcount by around 24% within a year, approaching a 40% reduction from its 2022 peak.

Then service quality and customer satisfaction declined. Klarna was forced into a partial reversal, rebuilding a hybrid human-AI support function to recover what the automation had cost. The short-term numbers had been favourable; the outcome still demanded an expensive walk-back.

Accenture began from an identical starting point and made a different capital allocation decision. The firm executed an $865 million restructuring involving 11,000 workforce reductions that generated $615 million in savings.

The difference sits in where the money went. Accenture redirected $250 million of those savings straight into AI growth, expanding its specialised data and AI talent pool to 75,000 professionals with a target of 80,000 by FY26. The reduction was not the strategy. It was the funding source for the strategy.

AI wage premium dynamics at the firm level help explain why Accenture’s reinvestment logic is defensible on financial grounds: PwC’s analysis of over one billion job postings finds that high-amplification firms, those using AI to raise worker output rather than cut headcount, achieved 163% productivity growth since 2018 and grew employment 52% versus 36% at lower-exposure peers.

The lesson is close to obvious a beat before it is stated: one firm extracted the gain, the other invested it.

The scale of the prize: Morgan Stanley Research estimates full AI adoption across S&P 500 companies could add up to $920 billion in annual net benefit and lift market capitalisation by 24% to 29%, but only if AI shifts work toward higher-value tasks rather than merely replacing labour.

That condition is the whole game. The gap between Klarna’s model and Accenture’s is not just operational; if the reinvestment approach is the one that scales, the difference is potentially worth trillions in aggregate market value. What you want to test in any management commentary is which of the two you are looking at.

What the macro picture tells you about where this is heading

The strategic divergence Gartner describes is not a firm-level quirk. It shows up in the macro data, where several major institutions have arrived at broadly the same picture from different starting points.

The macro picture sharpens considerably when global AI job exposure is disaggregated by income tier: Bank of America’s research finds high-income economies face roughly 33.5% exposure versus 11% in low-income countries, meaning the firms and sectors where AI-driven headcount decisions are most frequent are also the ones where rehiring costs will be highest.

• World Economic Forum: Over five years, 83 million jobs lost and 69 million created, a net reduction of 14 million, with 23% of all roles disrupted and nearly 75% of surveyed companies adopting AI.
• Goldman Sachs: A monthly net loss of 5,000 to 10,000 jobs in the most affected U.S. industries, and roughly 9% global worker displacement over a decade horizon.
• Federal Reserve (New York Fed): Only 4% of surveyed service firms reported AI-driven layoffs, with the majority retraining staff and adjusting hiring instead.
• Bureau of Labor Statistics: Occupations heavily exposed to AI, about 10 million jobs, saw a 0.2% employment drop between May 2024 and May 2025, concentrated in customer service, secretarial, and some sales roles.

The Federal Reserve figure is worth holding onto as a counterweight to panic. Mass AI-driven layoffs remain rare at the aggregate level; most firms are adjusting rather than cutting. But the firms that do cut aggressively are the ones taking on the liability described throughout this piece.

The talent market mechanics driving Gartner’s 2029 forecast

Goldman Sachs research explains why rehiring will be expensive rather than merely inconvenient. Technology-displaced workers experience real earnings growth nearly 10 percentage points lower over the decade after displacement than non-displaced peers, and about five points lower than workers displaced for non-tech reasons.

Goldman Sachs research on the AI displacement earnings penalty clarifies why rehiring will cost more than simply reposting the same role: technology-displaced workers carry a decade-long earnings growth deficit relative to peers, shrinking the experienced talent pipeline at exactly the moment companies need to rebuild.

That scarring does more than harm individuals. It delays their homeownership and wealth accumulation, and it shrinks the pipeline of experienced talent that firms will later need to rebuild. A thinner pipeline means scarcer, pricier talent.

Connect that to Gartner’s 30% rehiring forecast and the numbers stop looking speculative. For you as an investor assessing AI-heavy models, the scarring data is the critical variable: a company that has cut deeply in high-AI-exposure occupations faces a structurally thinner and more expensive market when it needs to rebuild. That is the mechanism behind the 2029 projection, and it gives you an external data source to cross-check against any single company’s workforce strategy.

Asking the right questions before the next AI efficiency announcement

The next time a company announces AI-driven workforce savings, you have a way to test whether you are looking at compounding advantage or one-time extraction. Three questions do most of the work.

1. Reinvestment commitment. Is the saving being redeployed into capability? A positive answer looks like Accenture directing money into AI talent. A negative answer is a saving that simply drops to the bottom line with no destination named.
2. Knowledge-preservation infrastructure. Is Gartner’s shift three visible in how the company describes its deployment? A positive answer keeps human oversight and context inside the process. A negative answer treats automation as removal, with no mention of preserving judgment.
3. Workforce adaptability investment. Is upskilling explicit? A positive answer names continuous learning and AI literacy. A negative answer talks only about headcount efficiency.

Not all AI-driven reduction is a mistake. It can be sound when it follows the recommended sequence: prove the AI at scale, redesign the work, then reduce only where the capability has been demonstrated reliably at edge cases. What makes the difference is order, not appetite. With stagnant global labour force growth, competition for skilled workers will only intensify.

The clock is the point. Gartner’s 2027 overtake projection means this divergence will be visible in market and operational outcomes within roughly the next eighteen months. Apply these three questions now, and you can identify which side a company is on before it is priced into consensus.

This article is for informational purposes only and should not be considered financial advice. Investors should conduct their own research and consult with financial professionals before making investment decisions. Past performance does not guarantee future results. Financial projections are subject to market conditions and various risk factors, and forward-looking statements are speculative and subject to change based on market developments.