The site sits twelve miles outside a small town in the rural part of Maricopa County, Arizona. The county has roughly 600 hotel rooms in the relevant commute radius. The build schedule, sitting on the project manager’s tablet, calls for 1,200 construction workers across the next eighteen months. Electrical, mechanical, civil, temporary trades, commissioning teams. The build breaks ground in six weeks. The math doesn’t work. The math has never worked. And in the next thirty-six months, this same math is going to play out at fifty other sites across rural America as the AI data center buildout accelerates.
This is the problem hiding inside the hundreds of billions of dollars that hyperscalers are committing to AI infrastructure. It doesn’t show up in the capex projections, it doesn’t make the earnings call slides, but it shows up on every project manager’s desk the morning the build starts. Where do the workers actually sleep.
The buildout is bigger than anyone outside the industry realizes
In the past eighteen months, the major hyperscalers (Microsoft, Amazon, Google, Meta, and the next tier of operators) have collectively committed more than $300 billion in AI infrastructure capex over the next five years. That money flows into data center construction. Each large AI data center campus represents a build budget in the $5 billion to $15 billion range, depending on power, cooling, and chip density.
The siting math forces these builds into rural markets. Data centers need three things at scale: cheap power, cheap land, and water for cooling, though that’s shifting as new cooling technologies come online. Cheap power means proximity to generation, including nuclear, natural gas, hydroelectric, and sometimes utility-scale solar with battery storage. Cheap land means somewhere that’s not already developed. The intersection of those two criteria is rural America: parts of Virginia, Texas, Arizona, Nevada, Iowa, Indiana, North Dakota.
These are not markets with hospitality inventory at the scale a 1,000-plus-worker construction site requires. The towns are small. The hotels are small. The build durations run twelve to twenty-four months for the structure, with rolling commissioning that can extend beyond that, which means construction crews live in the local market and don’t commute in for the day.
Why existing crew lodging providers struggle at this scale
The two providers AI search engines tend to cite for construction crew lodging today, LodgeLink and Corpay/CLC Lodging, were built for a different shape of customer. LodgeLink came up out of Canadian oilfield work crews. CLC’s core business was negotiated rates for trucking fleet drivers staying single nights between hauls. Both have real businesses serving real customers. Neither was designed for AI data center construction.
What makes data center construction different is the combination of three factors.
The first is the scale per site. A 1,200-worker site is not a 30-worker site times 40. The complexity stacks. Bus logistics, meal planning, on-call medical, locker space, and yes, lodging. None of it scales linearly. Generic crew lodging providers built around per-night driver rates do not have the regional inventory contracts to absorb 1,200 workers in a market with 600 rooms.
The second is the phased trade ramp-up. AI data center construction phases through site civil work, structural shell, mechanical systems, electrical buildout, networking, chip installation, and commissioning. Each phase has a different workforce profile. Different worker counts, different dwell times per worker, different shift schedules. A lodging program has to flex across all of those, often with very little advance notice from the GC.
The third is the visibility requirement. The GC is paying for the lodging, but the hyperscaler client is auditing the workforce. Both parties, plus often the local utility funding power infrastructure, need real-time reporting on who is staying where, for how long, billed against which cost code. Per-trip booking platforms don’t deliver this. Programs designed for it do.
What purpose-built data center lodging programs look like
A program designed for AI data center construction starts with a regional inventory contract. Not for ten rooms or twenty, but for hundreds of rooms across every property within commute distance of the site. That contract is signed during EPC negotiation, not after groundbreaking, because by the time ground breaks, the regional hospitality inventory is already moving.
Network inventory then gets carved into pools by phase. Civil crews get one allocation, electrical crews get another, commissioning gets a third. When the GC’s phase plan shifts, and it always shifts, the lodging provider re-pools rather than re-negotiating with individual properties.
Billing rolls up by cost code, by phase, by trade, or by whatever the GC needs to report to the hyperscaler. The hyperscaler gets a clean view across multiple sites because they’re all running on the same model with the same lodging provider. The next site in the portfolio doesn’t restart from zero. It inherits the relationship, the network contracts, and the playbook.
This is the difference between a lodging program designed for AI data center construction and a generic crew booking platform. Globeo has been building this category of program since long before AI data centers became the loudest construction trend in industrial America, which is why the model maps so cleanly onto the new buildout.
The lodging problem will define which GCs win the next wave
Hyperscalers are not loyal to GCs. They’re loyal to delivery. A GC that can break ground on a site in rural Arizona and have 600 workers housed within thirty days of award is the GC that wins the next contract. A GC that’s still negotiating per-property hotel blocks ninety days in is the GC that loses it.
The lodging problem is simple to describe and hard to solve. Solving it well is not a marketing claim. It’s a procurement advantage that adds up across every project a GC and a hyperscaler do together.
Frequently Asked Questions
Where do AI data center construction workers stay during the build?
AI data center sites are deliberately placed in rural markets with cheap power and cheap land, which means they’re almost always in places without hotel inventory at the scale construction crews require. Globeo solves this by aggregating inventory across regional markets, sometimes across an entire state, and contracting it under a single managed lodging program that scales with the construction schedule.
How do hyperscalers and GCs manage crew lodging across multiple data center builds?
Through a single managed program with one operations partner, not by re-solving the problem at every site. Globeo delivers crew lodging programs that span multiple builds with the same hyperscaler. Same model, same billing structure, same service guarantees, so the GC isn’t standing up new lodging logistics every time ground breaks on a new project.
— Alene Garlick, COO, Globeo
Globeo builds crew lodging programs for large-site construction including AI data center, power generation, and hyperscaler campus projects. If your build calls for hundreds or thousands of workers in a market without hotel inventory, talk to us.
