The U.S. Bureau of Labor Statistics projects total American employment to rise from 2025 to 2035, with healthcare and social assistance accounting for much of the expected growth. At the same time, some occupations that depend heavily on routine information processing are expected to shrink. AI is therefore more likely to rearrange the labor market than eliminate work altogether. The first workers exposed are likely to be those whose jobs consist largely of predictable digital tasks.
Customer-service representatives, administrative workers, data-entry employees, transcriptionists, routine financial processors and some forms of legal and analytical support can increasingly be assisted — or in some cases replaced — by software that can read documents, classify information, draft responses and complete transactions.
But even here, the distinction between job exposure and job destruction matters.
BLS projects software-developer employment to grow 15.8% between 2024 and 2034, adding more than 267,000 jobs. Information-security analysts and several other computer occupations are also projected to grow rapidly.
The paradox is straightforward: AI can make an individual programmer far more productive while simultaneously allowing a company to produce more software with fewer programmers.
That is already beginning to influence corporate decisions.
Amazon has said that generative AI and AI agents will allow it to operate with fewer people in some corporate functions. The company announced approximately 16,000 job cuts in January 2026. Salesforce has also reduced parts of its workforce while expanding AI agents designed to handle customer interactions. Block announced plans in February to eliminate more than 4,000 positions, nearly half of its workforce, as it reorganized around AI and automation.
Those examples should not be treated as proof that AI will cause comparable reductions across the entire American economy. Corporate layoffs have multiple causes, including restructuring, changing demand and previous over-hiring.
But they reveal the economic logic that is beginning to spread through corporate America.
A company does not necessarily have to fire an employee because of AI.
It can simply stop replacing employees who leave. It can hire fewer graduates, ask one manager to supervise several AI systems or give one analyst the workload previously handled by three.
For younger Americans, that may be one of the most consequential effects of the technology.
The most vulnerable part of the labor market may not be the entire profession. It may be the entry-level rung.
Junior employees traditionally perform repetitive work while learning how their industries operate. A young lawyer reviews documents. A junior analyst prepares spreadsheets. A new programmer fixes relatively simple problems. An entry-level marketing employee writes first drafts.
Those are precisely the tasks that generative AI is becoming good at.
If companies need fewer people to perform them, the immediate result may be fewer entry-level jobs rather than mass unemployment.
The next two years
It would be irresponsible to name companies and predict that they will make specific numbers of layoffs in 2027 or 2028.
There is, however, a clear group to watch: companies with large customer-service, administrative, software-development, advertising, financial-processing and other highly digitized workforces.
Amazon, Salesforce, Microsoft, Meta, Oracle and other large technology companies have already been restructuring while simultaneously increasing their investment in AI infrastructure. Recent reporting has documented a broader wave of companies explicitly linking some workforce reductions to AI.
The bigger change may occur outside Silicon Valley.
Banks, insurers, law firms, accounting firms, retailers, consulting companies and media organizations all have large pools of repetitive knowledge work. Once AI systems become reliable enough to perform complete workflows rather than individual tasks, those industries have a financial incentive to reduce labor requirements.
The result could be a labor market in which productivity rises faster than employment.
Americans may feel AI outside the workplace
The AI economy also has a physical footprint.
The servers running large AI systems require enormous amounts of electricity and cooling.
The Department of Energy says U.S. data centers consumed about 4.4% of the country’s electricity in 2023 and could account for roughly 6.7% to 12% by 2028. A newer Lawrence Berkeley National Laboratory analysis puts potential data-center electricity use at about 11.8% of U.S. electricity consumption by 2030.
Water is another issue.
Lawrence Berkeley National Laboratory estimates that U.S. data centers could consume between 0.14 billion and 0.28 billion cubic meters of water annually by 2028, depending on the technology and assumptions used.
That does not mean AI is about to make everyone’s electricity or water bills soar.
Research published by USAFacts found no clear national relationship through 2025 between the concentration of data centers in a state and faster residential electricity-price increases. It also noted that local effects can be different from national averages.
The issue is increasingly local.
A data center can bring investment and construction activity to a community while simultaneously creating demand for new power generation, transmission capacity and water infrastructure.
The Federal Energy Regulatory Commission and the Energy Department are already addressing the problem because large new data-center loads are changing electricity-demand forecasts and creating pressure on the grid.
The social cost may not appear in unemployment statistics
Work provides more than income. It provides routines, professional identity, social relationships and a path into adulthood.
If AI reduces the number of entry-level employees in offices, banks, law firms, technology companies and other professional organizations, the effects could extend beyond payrolls.
A young person who cannot obtain a first job may delay buying a home. A family facing reduced income may move. A worker whose profession disappears may retrain for an entirely different occupation.
A smaller company workforce may also mean fewer colleagues, fewer workplace relationships and fewer opportunities for informal mentoring. The economic statistics may record this as productivity.
The household may experience it as uncertainty.
The dividing line
The American labor market is unlikely to split neatly between jobs that AI can do and jobs that it cannot.
A more useful distinction is between work that AI can perform independently and work in which AI remains a tool controlled by a human.
A machine can summarize a contract. A lawyer remains responsible for the legal decision.
An AI system can write software. A company still needs people to decide what software should be built.
An AI agent can answer a customer. A human may still be required when the problem becomes unusual, expensive or sensitive.
A robot can automate some physical tasks. A plumber, electrician, nurse or construction worker still has to operate in an unpredictable physical environment.
That is why BLS continues to project strong employment growth in healthcare and other occupations that combine human judgment, physical presence and interpersonal interaction.
The biggest change may therefore be neither mass unemployment nor business as usual.
It may be fewer people doing more work.
That would make AI one of the most powerful productivity technologies in American history.
It would also force the country to answer a much older economic question in a new way:
When machines make workers dramatically more productive, who receives the gains — the workers, the companies, the owners of the technology, consumers through lower prices, or some combination of all four?