How Slowing AI Investment Could Reshape U.S. Employment: Fears of Weaker Hiring vs. Hopes for Rehiring Skilled Workers
Authored On
Modified
AI investment returns below expectations could put downward pressure on the economy Companies that cut staff are a wild card in a chilly labor market Disappointing automation results could bring renewed demand for workers

Warnings have emerged that cooling enthusiasm for artificial intelligence (AI) investment could send the effects of falling technology stocks and shrinking capital spending rippling through the labor market. The concern is that if returns on AI investment fall short of expectations, declining stock prices and a slowing economy could drive a larger increase in unemployment. Although pessimism that an economic slowdown will weaken employment dominates market sentiment, a substantial number of optimists argue that hiring could actually recover at companies that cut staff in anticipation of automation. They foresee an “AI boomerang”: as work that AI cannot handle piles up and the cost of correcting errors rises, companies may again seek employees familiar with their operations.
AI Employment Outlook Through the Fed’s Economic Models
According to Reuters on the 16th (all dates hereafter are local time), U.S. Federal Reserve Chair Kevin Warsh warned that if AI investment fails to deliver the expected returns, falling stock prices could be followed by an economic slowdown and deteriorating employment. At the “Productivity and Jobs” conference held that day at the Federal Reserve Bank of San Francisco, Warsh presented three scenarios for how the spread of AI could affect the U.S. economy over the coming years. Federal Reserve Board staff are using economic models to analyze how growth and unemployment might evolve depending on the extent to which AI raises productivity and how companies manage their workforces. Warsh explained that the three scenarios were hypothetical cases intended to explore a range of possibilities in preparation for future monetary policy discussions.
The first scenario envisions AI raising productivity, with the benefits shared broadly by workers and owners of capital. Warsh paid particular attention to AI’s ability to accelerate the process of discovering new ideas itself. According to research he cited, the United States has increased its research effort more than 20-fold since the 1930s, yet research productivity has declined. As technological innovation requires ever more effort, AI could lower barriers to expertise, allowing more people to participate in developing new technologies or improving existing ones. Warsh said that if AI-driven technological innovation steadily raises productivity, U.S. real gross domestic product (GDP) growth would exceed its long-run average. Assuming the gains from growth are shared broadly with workers, he also expected employment to remain close to full employment.
Table 1. Fed Chair Kevin Warsh’s Economic Outlook Under Different AI Adoption Scenarios
| Scenario | Assumptions and Transmission Channels | Growth Outlook | Employment Outlook |
|---|---|---|---|
| Productivity Gains and Shared Benefits | AI accelerates technological innovation, raising productivity, with the gains shared broadly by workers and owners of capital | Real GDP growth exceeds its long-run average | Employment remains close to full employment |
| Expanding Automation and Job Displacement | Automation increases output, but labor demand falls in occupations where AI replaces human tasks | Real GDP growth exceeds its long-run average | If reemployment is delayed, unemployment exceeds the level consistent with full employment |
| Weakening Expectations for AI Investment | Productivity improvements fall short of expectations, delaying corporate investment, depressing stock prices, and curbing household consumption | Potential economic slowdown as investment and consumption contract | An economic slowdown delays hiring and leads to a larger rise in unemployment |
Unemployment Could Rise Even as AI Lifts Growth
Even if AI adoption boosts economic growth, employment conditions could still deteriorate. In Warsh’s second scenario, companies increase output through automation, pushing U.S. real GDP growth above its long-run average. But as AI takes over tasks previously performed by people, demand for workers in some occupations declines. Indeed, an analysis by Stanford University researchers of payroll data through June of this year found that employment declines following the spread of generative AI were concentrated in occupations where AI replaces human tasks. Workers who lose their jobs in these occupations must acquire the skills needed for new work and find hiring opportunities before moving into other occupations. Warsh said such delays in reemployment could leave unemployment above the level consistent with full employment even as the economy grows.
The third scenario involves fading expectations for AI investment causing stock market turmoil that spills over into the real economy. U.S. companies have already poured enormous sums into AI infrastructure, including data centers, and their share prices reflect expectations for future profits. In this setting, if productivity improvements from AI adoption are slow, companies will postpone further capital spending, while investors lower their earnings forecasts, putting downward pressure on stock prices. If households whose wealth has fallen with stock prices also cut consumption, companies become more likely to delay hiring in response to weaker demand. Warsh expected unemployment to rise even more sharply if a stock market correction were followed by an economic slowdown. In other words, a shock originating in AI-related companies could threaten jobs in other industries through shrinking investment and consumption.
Uncertain Returns on Data Center Investment Add an Employment Wild Card
An AI bubble bursting is also the risk investors in financial markets are most wary of. In Bank of America’s (BofA) November Global Fund Manager Survey, released on the 17th, respondents most frequently identified the bursting of an AI bubble as a low-probability risk that could deliver a major shock to markets. The gap between the share of respondents holding overweight positions in technology stocks and those holding underweight positions also narrowed from 37 percentage points in October to 12 percentage points in November. This was the largest investor shift toward reducing technology exposure since June 2006.
Underlying this caution is the view that even rising AI demand does not guarantee a return on investment. Consultancy Bain & Company estimated that the AI industry would need to generate USD 2 trillion in annual revenue to expand data centers in line with projected demand in 2030. Yet even if companies shifted their entire in-house computing budgets to the cloud and reinvested all the savings from AI adoption, annual revenue would still fall USD 800 billion short of the required level. Concerns about excessive investment also appear in the outlook for corporate capital spending. In the BofA survey, a record 38% of respondents said companies were overinvesting. Another 52% expected at least one major cloud provider to announce capital spending cuts within the next 6 months.
If major cloud providers delay new data center construction as investors expect, construction companies and equipment installers will be the first to see their workloads shrink. The U.S. telecommunications industry in the early 2000s offers a precedent for an optimistic view of technology adoption driving a massive capacity expansion that ultimately hurt employment. As internet usage grew, telecommunications companies raced to expand their networks, increasing demand for workers to lay lines and install equipment. According to the U.S. Bureau of Labor Statistics (BLS), the U.S. telecommunications industry added 352,000 jobs between January 1996 and March 2001. But after the dot-com bubble burst and telecommunications network investment contracted, 337,000 jobs disappeared by the end of 2005. As construction work dwindled, employment in infrastructure construction, which includes telecommunications line installation, also fell by 30,000 between the second quarter of 2001 and mid-2003.
Table 2. Cases Requiring Rehiring After AI Automation
| Category | Factors That Can Undermine Layoff Decisions | Reasons Rehiring Is Needed |
|---|---|---|
| Delays in Next-Generation Model Development | Plunging AI stock prices and weaker new investment could delay the development and launch of next-generation models by developers dependent on external funding | Companies that cut staff preemptively in anticipation of automation will need to replenish their workforces if order and customer inquiry volumes hold steady |
| Differences in Automation Benefits Across Tasks | Applying early research findings focused on tasks with easily measurable outcomes to entire jobs can overstate the benefits of automation | Even if drafting contracts takes less time, adjusting transaction terms and resolving disputes still require staff familiar with the negotiation history and counterparties’ circumstances |
| Rising External Advisory Costs | Seeking outside experts whenever a dispute arises after contract staff have been cut reduces labor cost savings through advisory fees and work delays | As external advisory fees and losses from delays grow, rehiring employees familiar with the transactions becomes more cost-effective |
The Possibility of Rehiring as AI Investment Contracts
Some optimistic analyses, however, suggest that cooling enthusiasm for AI investment could prompt companies to restore some of the jobs they cut on the assumption of automation. Investors who borrowed against AI stocks must provide additional collateral or sell shares to repay their debts when stock prices plunge. If new investment also weakens amid mounting selling pressure, developers that have depended on external funding will have to adjust spending on next-generation models and their launch schedules. Even if technological development slows, orders and customer inquiries at client companies do not necessarily decline at the same pace. Companies that dismissed employees in advance, relying on the expected capabilities of next-generation models, may therefore find themselves needing workers again to handle the remaining workload.
One variable companies can easily overlook when deciding how many jobs to cut is that the benefits of automation vary substantially by task, even within the same employee’s job. In a paper published in the journal Economic Policy, MIT Professor Daron Acemoglu noted that early studies measuring AI’s productivity effects focused on tasks whose outcomes are easy to assess objectively. For example, a draft contract can be prepared using a prescribed template, but adjusting transaction terms or resolving disputes requires taking into account the history of negotiations and the counterparty’s circumstances. If a company dismisses contract staff because drafting takes less time, only to turn to outside experts whenever a dispute arises, the labor cost savings originally anticipated will shrink. Once external advisory fees and losses from work delays are included, rehiring employees familiar with the transactions may be more cost-effective.
Rehiring Skilled Workers Could Cushion the Employment Shock
One factor determining whether automation pays off is workload, which varies from company to company. In 2024, a team led by Neil Thompson, a research scientist at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), compared labor costs with the cost of building and operating computer vision systems that perform visual assessment tasks at U.S. companies with human-level accuracy. Using costs at the time, they found that tasks for which automation would save money accounted for just 23% of the total wages paid for such work. Even for the same inspection task, workplaces with low volumes struggle to recoup the fixed costs of acquiring data and building systems through labor cost savings. If the cost of adopting AI does not fall as much as expected, companies with modest inspection volumes may find it more economical to retain skilled workers who handle multiple tasks.
In office work, the cost of reviewing AI-generated work and correcting errors also shapes staffing decisions. In an experiment involving 758 management consultants, published in the journal Organization Science, participants who received AI assistance performed better on some tasks. But on tasks that AI could not solve properly, those using AI actually had lower rates of correct answers. Even companies that reduce drafting time face a greater burden of discovering errors late and redoing work if they also cut the employees responsible for identifying flawed judgments. Jobs with heavy verification requirements are thus a potential source of the AI boomerang, in which companies rehire experienced workers for positions they previously eliminated.
The AI boomerang could also emerge in new tasks created after automation. Researchers Anders Humlum and Emilie Vestergaard linked Danish AI adoption data with administrative labor market records and found no clear changes in workers’ average earnings or working hours during the early stages of AI adoption. What changed first was the content of their work. As companies adopted AI, they assigned employees new tasks, such as reviewing outputs and integrating AI with internal systems. In particular, identifying parts of AI-generated responses that conflict with customer contracts or established procedures requires staff with experience in the relevant work. The analysis suggests that at companies where such review work accumulates, former employees familiar with internal procedures and customers are likely to become priority candidates for hiring.
- Previous “Domestic Demand Freezes, Export Opportunities Narrow”: China Faces a Structural Economic Crisis as Consumption Recovery Remains Elusive Despite Massive Stimulus and High-Tech Investment
- Next “Neither Rate Hikes nor Currency Intervention Is Enough”: Prolonged Yen Weakness Pressures Japan’s Finances and Real Economy, Shaking Investor Confidence