Where AI is growing labor demand

The AI boom's fastest-growing jobs aren't at a desk, they're in the physical plants that house, power, and cool its servers, and in the factories that make its chips.

• HVAC: listings have roughly quintupled since 2023, as AI hardware runs too hot for conventional cooling.

• Data-center roles: at a record high, on a construction wave so large the Census now tracks data centers as their own category — spending topped a $50 billion annual rate in April 2026, bigger than general office construction.

• Electricians: data-center electrician listings are up more than tenfold, tracking the industry's most-cited bottleneck — a shortage so acute that Google, Meta, and BlackRock have pledged over $265 million to train workers.

• Chipmaking: TSMC's Arizona hiring went from near zero to 250-plus in about a year, and the fab draws power like a data center.

• The through-line: power is the common constraint at every stage.

Where the AI jobs are growing

Our latest reportage from the trenches of the AI revolution focused on the jobs AI is eliminating: software, sales, support. Today we're investigating the labor demand this same boom is ginning up. The fastest-growing demand is in the physical plant of artificial intelligence—the buildings that house it, the power that feeds it, the cooling that keeps it from melting down, and the factories that make its chips. In four Compass charts, we detail the labor story of AI's physical buildout.

HVAC listings have roughly quintupled since 2023.

Job Listings for HVAC technicians since 1/1/2023

Job listings for HVAC roles rose from about 1,100 at the start of 2023 to around 5,500 by mid-2026, with the steepest climb in the past year. AI accelerators run far hotter than ordinary servers, packing more power into each rack than air cooling can remove, which is why AI-dedicated facilities increasingly depend on liquid cooling. Goldman Sachs describes that shift, along with higher rack density, as a defining feature of this new class of data center, and projects that global data-center power demand will rise 165% by 2030 compared with 2023. Every facility on that curve needs cooling systems installed and maintained.

Data-Center-related roles at record high. 

Listings for data-center-related roles climbed from roughly 7,000 in early 2023 to about 10,700 in early 2026, dipping late in 2025 before setting new highs, with heavy turnover throughout. Behind the hiring is a construction wave the government now tracks on its own. The U.S. Census Bureau only recently began breaking out data centers as a separate construction category, and the figures show why: spending on data-center construction topped a $50 billion annual rate for the first time in April 2026 — enough to make data centers about 2.3% of all U.S. construction and, for the first time, a bigger category than general office buildings. (The money behind it keeps climbing: by a Financial Times tally of first-quarter earnings, Amazon, Microsoft, Alphabet, and Meta plan roughly $725 billion in combined capital expenditure in 2026, up about 77% from the prior year, most of it aimed at AI data centers. Each building, once up, needs technicians to run it.

Data-center electrician listings have jumped more than tenfold.

Listings for electricians with “data center” in job description since 2023

This is the steepest line in the set: listings for data-center electricians went from around 15 in 2023 to roughly 200 by 2026, with the sharp move beginning in late 2025. The numbers are small, but they track the industry's most-cited bottleneck. Microsoft president Brad Smith has called the electrician shortage the single biggest constraint on U.S. data-center expansion, and Fortune reports that electrical work accounts for 45% to 70% of a data center's construction cost. Companies are paying to close the gap: Google has committed $50 million to the electrical workers' union to lift annual apprentice intake from 19,500 to 30,000, part of a broader effort the New York Times puts at more than $265 million across Google, Meta, and BlackRock. The constraint is physical, not financial: Goldman estimates only about 50% to 60% of planned data-center capacity will come online on schedule, held back largely by power and permitting. It is also the job on this list least exposed to AI itself: someone has to wire the building, and there aren't enough trained people to do it.

TSMC's Arizona hiring went from near zero to more than 250 in about a year.

TSMC Job Openings in Arizona, since Jan 2025

At the front of the chain, the chips, the same pattern holds, and here too it runs on power. TSMC's listings for its new Arizona plant sat near zero until early 2025, then rose in a nearly straight line to more than 250 by mid-2026. The plant is a $165 billion project, and it draws electricity like a data center. Arizona Public Service says the first fab alone needs about 200 megawatts, which the utility equates to roughly 30,000 homes, and placed TSMC on a rate plan first designed for data centers. To serve it, APS built a new substation and relocated three sets of transmission lines, at one point using a helicopter to string wire along four miles of steel poles. The site keeps its own backup generators, but its reliance on outside power showed in September 2025, when a fault at a TSMC supplier briefly halted the fab and forced the company to scrap some wafers. What comes off the line justifies the effort: Commerce Secretary Gina Raimondo told Reuters the plant is producing leading-edge 4-nanometer chips on U.S. soil for the first time, the class of processor the whole shebang is built on.

In the story of the AI buildout, the constraint of electrical power is the common thread. It's what the cooling systems manage, what the electricians install, what delays the buildings, and what a fab can't run a shift without. The demand AI is generating isn't abstract. It has to be built, wired, and cooled, and that work is landing on trades and technicians.