AI infrastructure is exposing a major blind spot on data center jobsites

The AI infrastructure boom has produced a surge of headlines about compute power and a relative silence about the construction workforce making all of it possible. U.S. Census Bureau data shows data center construction spending hit $41 billion in 2025, up 32% year-over-year, and the safety programs on those jobsites are being asked to manage a density and pace most of them were never designed for.

I’ve spent the last several years working at the intersection of AI and construction safety, and that disconnect keeps growing.

These projects are a different beast

The defining characteristic of data center builds is extreme schedule pressure reshaping how every decision gets made on site. A construction delay on a hyperscale data center can cost millions of dollars per day in lost compute capacity, and that math changes behavior in ways that have very real safety consequences.

A typical large commercial project has dozens of contractors. A hyperscale data center build can have hundreds of specialized subcontractors working simultaneously across the same floors on compressed timelines. When one trade falls behind, the crews depending on them don’t get more time — they get pushed into a smaller space with less margin. That kind of compression tends to produce cut corners, lapsed procedures, and gaps in hazard communication that accumulate faster than anyone can track manually.

Where incidents start

Most people picture construction accidents as dramatic failures, like a beam collapse or a fall from a scaffold. The more common origin is far less visible. Imagine two crews end up working on the same water line, and one cuts in without telling the other. The result is flooded electrical equipment, a six-month wait for replacement parts, and millions in project losses, not to mention the human cost of anyone caught in that situation.

The incident takes seconds, but the coordination breakdown that set it up may have happened across several hours and shifts. The gap between two subcontractor safety programs that were never designed to communicate with each other leads to a serious accident. On a data center site with dozens of active trades, those gaps happen more often than most people outside of construction realize.

The Bureau of Labor Statistics’ 2024 Census of Fatal Occupational Injuries reported that falls, slips, and trips accounted for approximately 38% of construction fatalities that year, with the construction industry responsible for roughly one in five U.S. workplace deaths annually. Data center sites compound those numbers with hazards that traditional safety walk programs can’t track at the frequency required. Commissioning phases bring electrical crews energizing switchgear near active trades. Deep duct bank excavations need daily monitoring, and heavy equipment moving through crowded zones creates struck-by exposure that shifts from hour to hour.

Traditional safety apps weren’t built for this

Most construction safety management systems were designed for projects where a safety manager can personally oversee most of what matters. That model breaks down in three predictable ways on a hyperscale data center build:

Documentation volume outpaces review capacity. On a large hyperscale site, hundreds of pre-task plans are submitted daily across every active trade. No safety team can meaningfully review that volume through manual processes, which means the hazards flagged in those plans often never get cross-referenced against what the crew in the next zone is doing.

Fragmented records create accountability gaps. Most large data center projects involve hundreds of specialty trade contractors, each running their own safety program and their own form system. Safety directors have limited visibility into what’s being documented at the crew level by a sub working three floors up. The data might be there, but without a shared system, it stays siloed.

Nothing connects hazards across trade boundaries in real time. OSHA’s multi-employer citation policy holds general contractors (GCs) responsible for site-wide hazards, including those created by subcontractors. That liability exists whether the GC had visibility into the conditions that created it, and trade stacking combined with sequencing pressure compounds hazard exposure faster than manual walkthroughs and spreadsheets can track.

The stakes go well beyond OSHA compliance

The risks GCs and project owners carry on data center projects extend past recordable incident rates and OSHA exposure, and they’re more interconnected than most safety conversations acknowledge.

Schedule risk is the most immediate. A single serious incident can halt a project for weeks or months during investigation and remediation, at a cost that overshadows any investment in prevention. On an already compressed timeline, that kind of setback cascades through every trade still waiting to complete their scope.

Litigation exposure is increasing alongside these issues. Marathon Strategies’ 2025 analysis of nuclear verdicts found that 135 lawsuits resulted in jury awards exceeding $10 million in 2024 alone — a 52% increase over the prior year. GCs often end up carrying more liability than their contracts explicitly acknowledge, and after an incident, the question of who documented what — and when — becomes central to the legal conversation. A file full of checkbox observations that don’t describe conditions, trades present, or corrective actions taken won’t survive discovery.

Reputational risk on data center projects is asymmetric, too. Something goes wrong on a hyperscaler’s build, and the owner’s name ends up in the headline alongside the general contractor’s. That creates downstream pressure on future contract relationships that’s difficult to quantify but very real in practice.

Insurance economics add another layer. Large owners managing billion-dollar builds are increasingly adopting Owner Controlled Insurance Programs (OCIPs), which function similarly to a high-deductible health plan where the owner absorbs the first tranche of any claim — often $500,000 or more — directly out of pocket. Every preventable incident becomes a direct financial loss to the owner, and a well-documented safety program stops being a compliance requirement and starts being a profit mechanism. ConstructConnect reported that megaprojects accounted for $197 billion in U.S. construction starts in 2025, and OCIP structures have become more common as owners look for coordinated ways to manage risk at that scale.

What meaningful visibility actually looks like

The safety programs that hold up on these projects share a few specific qualities that traditional programs don’t. Most come down to getting better data from the field without making it harder for crews to do their jobs.

The first breakdown described above — documentation volume that outpaces review capacity — is primarily a data quality problem. Checkbox forms capture that something was observed, not what conditions were present, which trades were nearby, or what corrective action happened. Voice-to-text field reporting changes that equation. A superintendent standing in a congested mechanical bay can describe what they see out loud, and the observation gets structured, categorized, and routed into a dashboard without anyone stopping to type. The difference in data quality shows up in audit and insurance conversations.

The second failure mode — fragmented records across hundreds of subs — requires standardization that doesn’t force every subcontractor onto a new system. A platform that can capture whatever paper form a subcontractor is already using and extract structured data from it automatically closes the gap between subs with mature digital processes and the ones still running on clipboards. Every crew’s activity becomes visible through the same lens, regardless of their internal systems.

That standardization is what makes it possible to address the third failure: connecting hazards across trade boundaries. The coordination breakdowns that cause the most serious incidents on dense, schedule-driven sites are almost never visible in any one subcontractor’s documentation. They surface when you can cross-reference what multiple crews reported about the same floor, the same time window, or the same system.

Well-governed safety programs generate documentation that holds up — an auditor, underwriter, or opposing counsel reviewing a project’s safety record can see evidence that problems were identified, acted on, and closed.

The coordination problem is the safety problem

I’ve spent a lot of time connecting with safety directors and VPs at large GCs who describe the same situation. They know their teams can’t be everywhere. They know the data they’d need to catch coordination failures probably exists somewhere in their documentation. They just don’t have a practical way to surface it. The safety infrastructure the construction industry built over the past several decades was designed for a different scale and a different pace.

The hyperscale data center build cycle isn’t slowing down. The projects that can’t wait are already underway, and the safety programs managing them are being asked to operate at a scale they weren’t designed for.

If I could give one piece of advice to a VP of Safety or a project executive looking at their next hyperscale build, it would be this: stop treating safety documentation as a compliance deliverable and start treating it as a coordination system. The builders who invest in the tools that let them see across trades, zones, and subcontractor boundaries in real time are the ones with the best chance of finishing those projects on schedule and with a defensible safety record. The alternative is managing to the incident after it happens, and on a project this complex, that’s a position no one can afford.