How AI Is Reshaping the Job Market: The Blue-Collar Boom and White-Collar Squeeze

Generative AI is not causing an immediate unemployment crisis, but it is radically transforming where new jobs are created. While administrative and office positions face quiet contractions, the AI boom is driving unexpected hiring and wage growth in physical infrastructure, from data centres to energy grids. Latest labor market figures reveal how automation is repricing work - creating a surge in demand for technical and trade skills while squeezing routine white-collar roles.
The first phase of the AI revolution looks less like a software upgrade than an industrial construction boom.
The hidden economy behind the AI boom
Generative AI is usually discussed as a technology capable of replacing people who work with information. Yet its most immediate economic footprint is remarkably physical.
Annual spending on semiconductors, servers, data centres, cooling systems and electricity infrastructure is now about $500bn higher than it was in 2022, before ChatGPT became a mass-market phenomenon. US data-centre construction alone has passed an annualised $75bn, almost 60% above the previous year.
That spending requires people to build and operate the infrastructure.
Five industries closely connected to data-centre construction added about 320,000 more workers between 2023 and 2026 than underlying construction and manufacturing trends would otherwise have predicted. LinkedIn recorded almost 500,000 new physical data-centre positions between 2023 and 2025.
The labour market consequences are becoming visible in pay. Job postings for data-centre installation and maintenance offer wages about 40% above comparable positions elsewhere. Hourly pay in electrical equipment manufacturing has risen by more than 13% year on year, while electricians' pay has increased by almost 8%.
This is an important reversal of the conventional AI narrative. The technology that appears to threaten office workers is creating a shortage of workers who rarely appear in discussions about the future of AI: electricians, HVAC technicians, server technicians, grid engineers and workers manufacturing the equipment needed to keep computing infrastructure running.
The bottleneck is not only intelligence. It is power, buildings, cooling and hardware.
The office is where the pressure is concentrated
There is another labour market emerging alongside this infrastructure boom.
LinkedIn tracked roughly 640,000 new specialised AI job postings between 2023 and 2025. AI-specific positions now account for about 1% of professional jobs in the US, according to Burning Glass Institute chief economist Gad Levanon, although the share reaches 4-5% in computer science and biotechnology. Postings for AI engineers, AI directors and heads of AI roughly doubled between 2023 and 2024.
Those jobs sit at the opposite end of the labour market from the workers whose tasks can be standardised and automated.
Customer support employment has fallen by about 10% since January 2023. Secretary and administrative assistant positions have declined by roughly 15%. The BLS projects a net loss of 752,000 office and administrative support jobs by 2035.
The distinction matters. The current data do not show an economy in which AI simply eliminates jobs and leaves fewer jobs behind. They show an economy in which the value of certain forms of labour is being repriced.
Routine information processing is becoming cheaper. Physical implementation of computing is becoming more valuable. So is work that requires specialised expertise, judgement or responsibility for the output of increasingly powerful systems.
That leaves a difficult question for policymakers: whether workers leaving shrinking administrative occupations can realistically move into the sectors benefiting from the AI boom.
An office worker whose job disappears is not automatically an electrician, power engineer or data-centre technician. Retraining can bridge some gaps, but the skills, geographical distribution and working conditions are different. The wage premium in infrastructure therefore cannot simply be treated as a compensation mechanism for everyone displaced by automation.
Why productivity has not meant fewer workers everywhere
The employment effects of AI also depend on what happens to demand after productivity improves.
If technology allows a company to produce the same service with fewer workers but demand remains unchanged, employment can fall. But if lower costs make the service affordable to more customers, businesses can expand enough to employ more people overall.
There are early signs of precisely this dynamic.
Paralegal employment increased by about 11% between 2023 and 2025, while employment of market research analysts rose 6%. Both figures exceeded the 2.5% growth rate of the broader national economy over the same period.
AI can reduce the cost of producing professional services without eliminating demand for those services. A legal firm may be able to process more material, or a research business may be able to analyse more markets, because routine tasks become faster and cheaper. The productivity gain then creates room for expansion.
This is why the distinction between automation and economic contraction is crucial. A machine that performs a task previously done by a person does not necessarily eliminate the economic activity surrounding that task. It can reduce the price of the service, increase consumption and shift labour towards tasks that remain scarce.
For now, that mechanism appears to be outweighing direct AI displacement in several sectors.
The youth jobs test has not failed - yet
One of the most closely watched indicators has been employment among younger workers. Early warnings about generative AI often focused on entry-level positions because junior employees tend to perform precisely the repetitive, standardised tasks that new systems can handle.
Yet the unemployment gap between workers aged 20–24 and the wider labour force remains near a multi-decade low.
That does not mean young workers are insulated from AI. It means the expected first-round shock has not materialised at the macroeconomic level.
The explanation may lie partly in the sheer scale of the investment cycle. An economy spending hundreds of billions of additional dollars on computing and energy infrastructure generates demand across a wide range of occupations at the same time as AI reduces demand for some office tasks.
The danger is that this balance may not last.
The numbers have a blind spot
There is also a measurement problem.
US employers carry out roughly 1.7m routine job separations every month through turnover, restructuring, closures and other ordinary labour-market processes. Against that background, the roughly 16,000 monthly job cuts explicitly attributed to AI are small.
In August, Challenger, Gray & Christmas recorded only 3,462 announced cuts explicitly linked to AI, the lowest such monthly figure since late 2025.
That is useful evidence, but it should not be mistaken for a complete measure of AI's impact. Corporate restructuring rarely has a single cause, and employers do not necessarily describe every technology-driven reduction as an AI layoff. A company can automate tasks, change hiring plans or decline to replace departing employees without announcing a job cut labelled “AI”.
The official figures are therefore better at capturing explicit displacement than the slower changes in workforce composition.
For now, though, the broader labour-market evidence points in the same direction: the US is not experiencing an AI unemployment crisis.
It is experiencing an AI reallocation.
The next phase will be harder to read
The current balance is partly a consequence of timing. Building the infrastructure required for generative AI is labour-intensive. Once the data centres, networks, power systems and hardware supply chains are established, that source of employment growth should not be assumed to continue at today's pace.
That creates a question the present jobs data cannot answer.
If the infrastructure boom eventually matures, the economy may be left with AI's productivity gains but without the extraordinary construction and engineering demand that currently offsets displacement elsewhere. The specialised AI workforce may continue to grow, and cheaper professional services may generate additional demand. Still, the labour market will have fewer physical projects absorbing workers pushed out of routine office roles.
The first stage of the AI economy has therefore produced an unusual coalition of beneficiaries: software specialists and electricians, data-centre technicians and highly productive professional services. Its losers are more concentrated in the middle of the administrative economy.
The decisive test comes later, when the cranes leave the data-centre sites, and the capital spending boom becomes a mature installed base. At that point, the question will no longer be whether AI can create jobs. It will be whether the economy can keep creating enough new forms of work after the infrastructure needed to build the machines is already in place.
Sources: Economist, The Podcast Summary, Newsweek, Singularity.kiwi \\
