The AI labor story that is not “learn to code” — it is pull wire, frame steel, and pour forms.
On July 29, 2026, the New York Times reported that AI companies are recruiting electricians and carpenters by the thousands to staff the physical half of CapEx: data centers. HN’s thread filled with IBEW wage notes, Abilene hotel rates, residential quotes that evaporated, and boom-bust warnings from oil-sands veterans.
This is the same infrastructure story we track in data-center environmental impact and orbital compute dreams — except the bottleneck is people with licenses, not just GPUs.
TL;DR
| Question | Answer |
|---|---|
| What’s hiring? | Trades for data-center construction |
| Why now? | Multi-year AI CapEx / power / shell buildout |
| Career risk? | Boom-bust — ops need fewer bodies than build |
| Carpenters? | Metal studs, forms, doors — not log-cabin DCs |
| Housing link? | Regional labor competition; not the only cause |
| Union angle? | IBEW / commercial rates strong in some metros |
| Next trade? | More process plumbing / liquid cooling |
| Robots replacing them? | Not on arbitrary jobsites this cycle |
What people are asking
“Should I quit software for the trades because of AI?”
Only if you want the work — not because a headline guarantees $300k forever. Commenters who lived oil booms: milk the premium while CapEx runs, save like the job ends, keep residential/commercial networks for the trough. Construction always moves sites; industrial electrical was boom-bust before ChatGPT.
Software is also cyclical. Relative stability is the question. Teen choosing a path: trades with electrification demand (EVs, heat pumps, grid) can be rational without betting your life on Stargate-scale campuses.
“Are data centers wooden now?”
No. Typical shells are tilt-up concrete, steel joists, metal roofs. Interior partitions use steel studs. Carpenters still frame those, hang doors, build concrete forms, and fit offices. NYT’s “carpenters” is craft classification, not a timber meme.
“Will this permanently raise my bathroom remodel quote?”
Sometimes regionally. HN: near large DC builds, residential electricians/HVAC book out-of-state plates and “fuck off” prices. Counterpoint from commercial electricians: residential and commercial pools barely overlap — homeowners are hard customers and pay less, so the shortage is also a preference problem. Zoning and permitting still dominate housing supply; cheaper wiremen alone will not fix California.
“Isn’t this just temporary construction employment?”
Yes for headcount-at-peak. Finished facilities need maintenance (UPS, HVAC, now denser liquid loops) but not armies of apprentices. Prefab/modular reduces on-site hours if factories exist — many owners are building before the prefab market is mature.
The incentive map
| Actor | Want |
|---|---|
| Labs / hyperscalers | Megawatts online before rivals |
| EPCs / GCs | Bodies on site, any state plates |
| Union halls | Hours + dues in a hot market |
| Apprentices | Entry that is waitlisted in normal years |
| Homeowners | Someone who still answers the phone |
| Local towns | Temporary hotel/RV boom; thin permanent ops jobs |
Bernie/Trump-era public money into AI infrastructure debates (HN) raise a separate fairness question: if taxpayers fund CapEx, what commons do they get back — open research, grid upgrades, or private shells?
Wages, unions, and the apprenticeship wall
Texas certificate anecdote on HN: helpers ~$16/hr, journeymen $30–40/hr, while Minneapolis/St. Paul IBEW journeyman base cited >$60/hr plus fringes toward **$110** total package. Masters often make money by running shops, not a huge solo wage premium.
Getting in is the bottleneck: short enrollment windows, family networks, decade-long waitlists in some eras. An ME degree does not teleport you past the queue. That inelastic supply is why a CapEx spike moves prices so hard.
Liquid cooling and the next bottleneck
As racks crawl toward hundreds of kW and vendor talk of ~1 MW / 800 VDC class designs, process plumbing and glycol loops matter more than drywall. Electrical arc-flash and DC incident-energy PPE gaps make energized work on dense racks a safety nightmare — de-energize or don’t touch. Expect more licensed pipefitters and fewer “just add fans” retrofits. Pair with our energy/math model selection and Stanford AI Index footprint notes.
Who “wins” and who pays
| Outcome | Who feels it |
|---|---|
| Higher trade wages on DC sites | Journeymen with hours; GCs pass cost to hyperscalers |
| Scarce residential service | Homeowners near build corridors |
| Temporary hotel/RV economies | Abilene-class towns; fades when shells finish |
| Permanent ops jobs | Thin — a few technicians per campus |
| Delayed non-AI industrial projects | Everyone bidding the same crews |
| Electrification backlog (heat pumps, EVSE) | Households competing with AI CapEx calendars |
Critics call this resource misallocation: concrete and copper into facilities many locals dislike, while housing and schools wait. Defenders: demand is demand; without AI CapEx those hours might not exist at premium rates. Both describe real distributional fights — same pattern as AI race protests meeting physical plant.
Software workers discovering construction labor costs after years of celebrating automation is not lost on trades commenters. Fair. Still: do not romanticize the work. Family electricians aging out with broken bodies are a data point beside viral $300k years.
Prefab, travel, and “enough” data centers
Owners prefer modular / prefab shells when factories exist; many are building anyway before that market matures. Sites fill with out-of-state plates. Travel rules matter: sometimes hotels; sometimes you are under the hotel cutoff and sleep in the truck. Older workers and parents self-select out — a younger-worker game at the margin.
Will we ever have “enough” data centers? CapEx guidance, power interconnect queues, and model-efficiency jumps (fewer watts per token) can all pause builds. That is exactly when overtime vanishes and crews flood other markets. Plan for enough as a moving target, not a destination.
Humanoid robots wiring arbitrary jobsites remain science fiction relative to chatbots taking drive-thru orders — which already struggled. Autocomplete for conduit in a settled, dusty shell is a harder problem than next-token prediction. Do not skip learning the trade because of demos; do not skip saving because of demos either.
Compare to other CapEx labor booms
Oil sands, shale man-camps, and semiconductor fab builds all pulled trades at premiums and then shed them. AI campuses rhyme: out-of-state plates, hotel scarcity, thin permanent ops. The difference is narrative — “intelligence infrastructure” vs “energy extraction” — not the labor math. If chip valuations wobble or interconnects slip, overtime disappears faster than LinkedIn posts about trades careers.
For software readers: empathy for construction cycles is useful; career cosplay is not. Pulling wire in July heat is not a vibe. If you go, go for the craft and the optionality across electrification — EVs, heat pumps, grid — not for a single tenant logo on a hard hat.
Career checklist if you are considering the jump
- Talk to local IBEW / non-union shops about current hours, not Twitter screenshots.
- Ask whether the work is travel / per diem / man-camp distance.
- Model taxes on spike years and savings rate as if year three is quiet.
- Keep a path back to residential/commercial when DC work thins.
- Do not confuse construction employment with permanent AI company employment.
- Ignore “robots next year” as a reason to avoid learning the trade — and ignore “robots never” as a reason to skip savings.
- If you only want “AI money,” price the hours and travel honestly — 70-hour site weeks are not laptop weeks.
- Watch interconnect queues and GPU CapEx guidance as leading indicators that build schedules (and overtime) will slip.
Boom hygiene
- 50%+ of premium wages → cash buffer
- Maintain license CEUs even on DC sites
- Network outside the single mega-project
- Watch GPU CapEx / power interconnect delays as leading indicators
- Separate “learns the trade” from “bets on Stargate forever”
Policy angle for readers who fund or regulate: public money into AI campuses without grid and apprenticeship expansion just bids up scarce licenses. Pair CapEx incentives with training seats if the goal is capacity, not only rent for existing journeymen.
Honest limitations
- NYT framing can over-attribute nonresidential construction trends to AI alone — check multi-decade FRED series before causal claims.
- Anecdotes ≠ national BLS medians (~$62k electrician median May 2024 in one cited snapshot — far from viral $200k Reddit).
- War / industrial policy shocks can redirect the same labor pool overnight.
- “Carpenter” and “electrician” cover wide skill distributions.
- Robot timelines are speculative; CapEx busts are historical.
Closing
AI’s scarcest input is not always FLOPs — sometimes it is a journeyman who will work nights in a half-finished shell. That is good for trades leverage and bad for everyone competing for the same crews. Treat the NYT story as a CapEx labor market signal, not career destiny, and keep reading power and water constraints beside the wage screenshots.
Follow @explainx_ai for infrastructure and energy follow-ups.
Related on explainx.ai
- AI unicorns barely publishing research
- Data center real environmental impact
- Model selection — energy math
- SpaceX orbital data center concepts
- Stanford AI Index 2026
- Can AI solve global warming?
- Stop AI race protest SF
Sources
- NYT — A.I. companies are recruiting electricians and carpenters by the thousands (Jul 29, 2026)
- HN discussion (search title; gift/archive links circulate in-thread)
- BLS — Electricians
Labor-market anecdotes and NYT framing as of July 29–30, 2026. Verify local wage scales, apprenticeship openings, and project timelines before changing careers.
