The data center bottleneck nobody budgets for
Hyperscale sites are chosen for power and fiber, not for population, and the construction peak routinely exceeds everything available within a reasonable commute. Accommodation has a short lead time, which is exactly why it gets deferred — and why it becomes the thing that slips. The fix is to model beds against the labor curve at the same time the long-lead equipment is ordered.
The arithmetic error that causes most of it
Peak headcount is not a plateau. It is a spike, and it lands late — during mechanical and electrical work, typically between months twelve and eighteen of a two-year build. On a realistic curve, peak runs at roughly twice the project average.
Plan accommodation to the average and you are short by half at the exact moment the schedule is least forgiving. And the shortfall falls disproportionately on the scarcest trade on the site.
Getting it wrong the other way is expensive too. Oversize the camp and you carry months of empty beds on either side of the peak. Neither error is a procurement failure. Both are modeling failures.
Why short lead time makes it more dangerous, not less
Transformers get ordered thirty-six months out because everyone accepts the lead time is real. Accommodation lead time is measured in weeks and months, so it stays on the list of things that can be dealt with later.
It cannot, because the constraint is not manufacturing. It is site selection, soils, access, utility strategy, foundation approach, permitting and commissioning — none of which compress just because someone has noticed the problem. By the time the shortfall is visible on a headcount report, every remaining option is expensive.
What it costs when it slips
Commissioning delay on a typical 60 megawatt facility is widely put at around $14.2 million a month in lost owner revenue. Set against that, the accommodation budget is a rounding error that controls a critical path.
The labor market is the other half of the equation. Roughly 41 percent of the current construction workforce is expected to retire by 2031, and the electrical trades — which account for 45 to 70 percent of total data center construction cost — are the scarcest of all, with more than 300,000 new electricians needed over the next decade.
In that market, accommodation quality is a schedule instrument rather than a welfare line item. Short commutes, private rooms, real food and reliable connectivity are the reason a scarce electrician stays on your site rather than moving to the project four states away that has a bed and a shorter drive.
The other question worth asking
Rented trailers and hotel blocks are pure expense and produce nothing at the end. Factory-built units are capital equipment — they demobilize, move and redeploy to the next build.
For anyone running a pipeline rather than a single project, that moves accommodation out of the project cost line and into the asset base, where it can be financed and depreciated across multiple deployments. Whether it should sit on your balance sheet or someone else's is a real question with defensible answers either way. What is not defensible is never asking it, and defaulting to rental because that was faster to approve.
Sources
- Associated Builders and Contractors, construction workforce model, February 2026 — 349,000 net new workers needed in 2026, 456,000 in 2027.
- NCCER — approximately 41 percent of the current construction workforce expected to retire by 2031.
- Industry analysis, 2026 — commissioning delay on a 60 MW facility estimated at approximately $14.2 million per month in lost owner revenue. Widely cited; secondary source.
- Industry analysis, 2026 — electrical systems at 45 to 70 percent of total data center construction cost; more than 300,000 additional electricians required over the next decade.
Reviewed September 2026. Published figures on this site are re-checked every six months and corrected in public when they move.
Know your peak headcount?
Tell us the site, the programme and the number. We will tell you what we would check first.