AI capital expenditures are booming. Jobs aren't. Here's why

Amazon $AMZN, Microsoft $MSFT, Meta $META, and Google $GOOGL are planning a combined, unprecedented $650 billion worth of capital expenditures this year. In any prior era, a corporate spending spree of this magnitude would have been an unambiguous jobs engine. Factories would have gone up. Hiring would have followed.

This time, the money is going into a different kind of infrastructure — and the labor market is not following the capital. Here's what to know.

Where the money is going

The mechanics of this capex cycle are distinct from any recent predecessor. Current estimates suggest that 80% of the growth in final private domestic demand in the first half of 2025 came from AI data centers and high-tech-related spending, according to S&P Global $SPGI research.

Strip out AI-related investment, and U.S. business spending tells a different story. Pantheon Macroeconomics found that private fixed investment "is rising only due to AI-related spending," with analyst Oliver Allen showing that all other categories are in decline, according to Fortune.

The composition of the spending explains the disconnect. When companies build a factory, the capital investment creates demand for workers who will operate that factory for decades. When companies build a data center, the calculus is different. According to McKinsey, global data center investments are projected to reach $7 trillion by 2030, with more than $4 trillion allocated to computing hardware — chips, servers, and storage, much of it manufactured overseas. McKinsey's analysis of the value chain shows that of the $6.7 trillion total, about $4.3 trillion flows to technology developers and manufacturers (servers, semiconductors, storage), while only about $600 billion goes to construction labor.

The 500-to-50 problem

Data centers are, by design, built to run with minimal human labor. As Data Center Knowledge has noted, data center projects generate large numbers of temporary jobs during construction.

Microsoft's data center in Quincy, Wash., for example, had as many as 500 workers on-site during the construction process but now employs 50 full-time staff. That pattern — a spike of construction employment followed by a permanent workforce that fits in a large conference room — repeats across the industry. A standard 12-megawatt data center typically requires between 20 and 22 operational staff, according to industry staffing estimates. Compare that to a manufacturing plant of comparable capital investment, which might employ hundreds or thousands.

Good Jobs First, a nonprofit that tracks corporate subsidies, has documented this pattern in detail. The organization found that almost half of state data center subsidies do not require job creation. Those that do usually require 50 or fewer jobs per project, compared to manufacturing projects that can create thousands. In one case, a data center in New York promised 125 jobs in exchange for $1.4 billion in public subsidies — $11 million per job. Data center employment accounts for only about 0.01% of all jobs in the U.S., yet the industry consumes about 4.4% of the nation's electricity.

That ratio captures the core of the problem: This is an investment cycle that consumes enormous resources while generating thin employment.

Capex without hiring is new

The historical relationship between capital expenditures and employment was straightforward: When businesses invested more, they hired more. JPMorgan $JPM's analysts said the kind of economic decoupling they're tracking is unusual.

"Accelerating capex amid a stall in job growth is hard to incorporate into the outlook. Such a juxtaposition is not evident over any US expansion in the past 60 years," the bank said in its 2026 outlook, according to Business Insider.

S&P Global observed that historically, the tech industry has been viewed as human capital-heavy with a light capex footprint, in contrast to manufacturing, which was increasingly capex-heavy and light on employment. AI is collapsing those categories.

Tech companies are now the ones driving capital-intensive spending, but the assets they are building — data centers, GPU clusters, power infrastructure — are designed to replace human labor, not employ it.

As Quartz has reported, the hundreds of billions that Amazon, Microsoft, Google, and Meta are pouring into AI infrastructure are not translating into hiring, because they are going into data centers, not people. Columbia Business School professor Daniel Keum described the spending as a "technological shock" in which AI is targeted at replacing workers and reducing headcounts, not augmenting the workforce.

What it means

S&P Global noted that the combination of positive GDP growth and flat employment points to potential early productivity gains from AI, powered by data centers. That is the optimistic interpretation: The economy is getting more productive.

The less optimistic interpretation is that the gains from this investment cycle are accruing to capital, not labor, in a way that prior cycles did not. Numerous commentators and economists note that AI spending is now large enough to be visible in total GDP numbers, and in some recent quarters, this growth even outpaced the contributions of consumer spending.

The U.S. economy is, in a sense, being kept afloat by the AI buildout. But unlike the postwar industrial expansion or the internet-era tech boom, this buildout does not pull millions of workers into well-paid employment. It pulls silicon, electricity, and concrete.

The question is what happens when the spending peaks. Construction workers will move to the next site. The facilities themselves will hum along with skeleton crews. And the technology inside those buildings will continue doing what it was designed to do: work that humans used to perform.