AINEWS
Hyperscale AI sites — the largest, fastest built data centers — burn $60M to $70M of construction a month, roughly four times a typical hospital project, and contractors say the tempo is forcing safety to react to incidents rather than prevent them
A typical $200 million hospital project moves roughly $15 million of construction work in a month. A hyperscale AI data center moves $60 million to $70 million in the same window, according to Marc Hutson, senior vice president and director of construction operations at JE Dunn (ENR). Hyperscale refers to the largest, fastest-built data centers, mostly built to host AI compute.
That four-times tempo is colliding with a safety operating model built for slower projects, and contractors are now saying so on the record.
"If you have an old-school mentality, you're going to continue to come up short on preventing injuries," Hutson told ENR, calling the prevailing hyperscale approach "very reactionary" rather than proactive.
The pressure is mechanical. Burning through $60 million to $70 million of work in a single month means rotating subcontractor crews faster than any single team can train them, layering multiple general contractors, or GCs, on shared sites, and absorbing high turnover in the skilled trades. "Multiple general contractors on different projects with different scopes create a tangled web of accountability for safety training and management," the ENR feature reported, citing Hutson's framing.
That tangle matters because safety culture on a construction site usually travels with institutional memory: the foreman who knows which welder skipped the briefing, the superintendent who recognizes the corner-cutting shortcut before it becomes an injury. At hyperscale tempo, that memory keeps getting cleared out.
Caterpillar Safety Services, a consultancy run by the heavy-equipment maker, has been brought in by GCs for safety evaluations and training on data center megaprojects. The engagement is explicitly not tied to any Caterpillar equipment purchase or rental, a distinction Hutson drew so the reader would not read the case as a vendor plug. JE Dunn had previously used the consultancy during subcontractor onboarding, then re-engaged as project signs of stress and possible enforcement conflicts emerged, per ENR.
That pattern, treat the safety program as something you can hire in reactively when the project starts to look shaky, is the posture Hutson described as insufficient. A "very reactionary" approach means investigating after a near miss rather than designing the work so the near miss cannot happen.
The structural problem extends past any one firm. Labor shortages, in-demand skilled trades, and high turnover are described as creating cracks in existing safety cultures on data center megaprojects. Each new crew onboarding is a chance for a safety briefing to be skipped, abbreviated, or delivered in a language the worker does not fully follow. Each additional prime contractor on a shared site is another handoff where accountability can blur.
None of this is an indictment of any one firm. JE Dunn, the only general contractor quoted in the available reporting, is the source identifying the problem and reaching for outside help. The limit of the available evidence is that the diagnosis rests on a single trade-press article and one executive's framing. Independent verification from a second general contractor, an insurance carrier, a safety academic, or OSHA recordable-rate data on data center construction has not yet been published alongside Hutson's account. The four-times tempo contrast and the "reactionary" framing should be read as one contractor's diagnosis, not an industry consensus.
What is structural, not editorial, is the math. A site that bills $60 million to $70 million a month cannot run the same onboarding cadence, the same crew continuity, or the same single-prime-contractor accountability as a site billing $15 million. Either the industry redesigns safety around that tempo, or the gap between design-out and incident-response keeps widening.
The next test is whether the outside consultancies, internal safety teams, and multi-GC coordination models now being tried on these megaprojects can keep up with the build cycle, or whether the safety program keeps lagging the schedule by a quarter or two. Right now, the on-the-record answer from the people doing the work is that the old playbook is no longer matching the pace.