The trade the grid needs did not grow

There is a good story going round that the AI build-out has produced a blue-collar employment boom: welders, pipefitters, electricians and HVAC technicians in demand, apprentices on $40,000 to $60,000, experienced electricians past $100,000, postings for welders and pipefitters up 164 per cent year on year. The job postings are real. The payroll survey is a different document, and it is the one that decides whether this is a boom or a rounding error.

Total construction employment in the United States was 8,359,000 in August, up 120,000 on the year — 1.5 per cent. Inside that number is the thing worth looking at. Nonresidential building construction reached 947,300, up 26,000 on the year and 29,100 on two years. Residential building construction was 923,700, up 900 on the year and still 14,200 below August 2024. Two years ago residential led nonresidential by nearly 20,000 people; it now trails by 23,600. The crossover is the build-out in one line of a federal survey: America has not hired more builders, it has moved them from houses to sheds.

The trades break down the way the anecdotes suggest, at a fraction of the implied scale. Electrical contractors employed 1,157,100 in July, up 42,000 on the year — 3.8 per cent, the fastest growth of any large trade. Plumbing, heating and air-conditioning contractors were at 1,338,200, up 23,200. Electrical equipment manufacturing, which makes transformers and switchgear, employed 443,400 in August, up 14,900. Good years for a mature trade. Not a national labour-market event.

Then the series nobody quotes. Power and communication line and related structures construction — the line crews and substation connections, the people who physically attach a campus to the grid — employed 248,900 in July, down 1,500 on the year. The constraint on this build-out is the interconnection queue and the lead time on heavy electrical plant, and the one occupational series that would shorten either is shrinking. Whatever the hyperscalers are bidding for, it is not line crews.

That matters for the multiplier now being put in front of county boards. Cushman & Wakefield's research has every 100 megawatts of new development creating nearly 1,300 jobs locally, $110m of annual wages and $344m of gross output. Hold it against the stock. The Brookings paper this desk examined on Thursday counts 57 gigawatts of operating American capacity and assumes 183 by 2032. At 1,300 per 100 megawatts, today's fleet would already account for 741,000 jobs — against 947,300 people employed in all nonresidential building construction in the country. The 2032 figure would imply 2.4 million, a 28 per cent rise in total construction employment. That number can only be temporary job-years counted with indirect effects, and it is used in rooms where people hear permanent jobs.

Michael Hicks, who has studied the local economics of data centres since the early 1990s, puts it plainly: "Permanent labor market effects are muted." He blames the backlash on the promises rather than the buildings — "developers have woefully overpromised jobs and given tax breaks that are beyond obscene" — while arguing the real local benefit is the tax base, not the payroll. Brookings found the debate moving faster than the research: the "evidence base has not kept pace".

Which brings the argument to the counties, where the circulating number is wrong. Data Center Watch, the research project at 10a Labs that tracks this, is being cited for "at least 75 data center projects worth roughly $130 billion" blocked or delayed this year. That is its figure for the first quarter of 2026 alone, which its own report calls the largest single-quarter concentration on record. Its second-quarter report adds at least 45 more projects worth nearly $68bn. The published running total for the first half is therefore at least 120 projects and around $198bn — not 75 and $130bn. The same reports count more than 300 state data-centre bills filed in the first six weeks of 2026, moratorium proposals in 14 states, and Maine one House vote short of the first statewide ban.

Texas then stopped being hypothetical. On 21 September the governor directed the Texas Commission on Environmental Quality to halt all permits sought by data centres until the grid operator completes an audit of every project in its interconnection queue. "Data centers must pay their own way, protect our grid and water," the directive reads. "Until they do, TCEQ will issue no permits sought by data center projects." A week earlier the water board had been told to compel water-use reporting. The commission must report on its compliance by 19 October. And at the end, the sentence that should worry a developer more than the freeze: next session, the governor will work with the legislature "to eliminate any financial incentives for data centers".

Public opinion is behind all of it — around seven in ten Americans opposed to a local data centre on Gallup's measure, roughly two-thirds across all three party identifications in a New York Times/Siena poll this month, 57 per cent in Texas itself — though this desk has not read those crosstabs. New York is reported to have gone further with a statewide ban; the governor's press office did not respond here and the claim is unconfirmed at source.

So the ground-level picture is coherent, and it is not the one either camp is selling. The jobs are real, concentrated in one trade, a fraction of what the multiplier implies, and almost entirely inside the fence. The crews who would clear the queue outside it are not being hired. And the political risk now attaches to the permit rather than the ballot — which, as the credit documents behind these campuses show, is where delay becomes a payment date.