$725 billion in AI infrastructure capex is chasing a short list of qualified electrical and mechanical subs

Q1 2026 earnings confirmed what the construction pipeline already implied: Amazon, Microsoft, Google, and Meta are collectively planning roughly $725 billion in capital expenditure this year — up 77% from 2025's approximately $410 billion. Moody's, reviewing those same results, raised its own forecast to $785 billion. JPMorgan projects cumulative hyperscaler infrastructure spending will reach $5.5 trillion through 2030.

Nearly all of that capital eventually moves through a construction site. US data center construction spending eclipsed $50 billion earlier this year, according to Bloomberg — a pace the market has never sustained before. The firms collecting that work are not a random sample of MEP subs. They're a short list, and most of them locked in their position two years ago.

What data center electrical and mechanical work actually requires

Standard commercial work asks an electrical sub to design and install distribution, lighting, and low-voltage systems according to NEC code and local authority having jurisdiction requirements. Data center work starts there and adds an entirely different layer.

Critical power architecture. Hyperscaler facilities operate at N+1 or 2N redundancy — every major power system component has a hot standby or a full duplicate ready to assume load instantly. UPS (uninterruptible power supply) systems, automatic transfer switches, and backup generation aren't a backup provision on these jobs; they're primary design requirements, commissioned to guaranteed switchover tolerances measured in milliseconds. If your firm has never scoped or installed a 2N switchgear arrangement with full UPS bypass, this will be the first thing a data center client checks.

High-density cooling infrastructure. GPU clusters deployed for AI inference and training run at power densities far higher than traditional server rooms. The mechanical work that follows — precision in-row cooling, rear-door heat exchangers, direct liquid cooling manifolds, and CRAH (computer room air handler) installations — operates to tighter tolerances and tighter commissioning schedules than conventional HVAC. Mechanical subs entering this market for the first time typically discover their standard QA/QC process doesn't match what the client expects.

Integrated systems testing. Acceptance on a data center project isn't a punch list walkthrough. IST (integrated systems testing) protocols simulate power failures, cooling failures, and cascading events on an energized facility. Construction teams are present through this process, not signed off before it starts. Firms that have only done traditional commissioning underestimate the schedule implications.

What getting on a hyperscaler preferred contractor list requires

Hyperscaler procurement teams maintain qualification lists for each geography. Getting on those lists is not a bid exercise — it's a documentation exercise, and the requirements are specific.

Tier III or Tier IV data center experience. The Uptime Institute's facility classification system defines the redundancy and concurrently maintainable criteria that hyperscaler clients use as their baseline. They want prior projects that achieved Tier III or higher certification — not just any electrical work in a building that has servers in it.

NFPA 70E compliance and arc flash programs. Data center work involves regular contact with energized systems at voltage levels and current magnitudes that make arc flash a first-order safety concern. Expect to submit written hazard analysis programs and PPE documentation for every technician on site, not just a generic safety plan.

Safety record documentation. DART (days away, restricted, or transferred) rates are reviewed, and for many hyperscaler programs there are floors below which a firm is ineligible regardless of price or relationship.

Background-checked workforce. Federal and defense workloads hosted in hyperscaler infrastructure mean some projects require ITAR-cleared or otherwise vetted personnel. This is a growing requirement, not an edge case.

None of this is unreasonable for a firm that has been doing critical facility work. For a firm that hasn't, building these programs from scratch while pursuing the work simultaneously rarely goes well.

The window, and the constraint that shapes it

The pipeline is real. Grid connection delays in primary data center markets now stretch past four years in some jurisdictions, which creates schedule uncertainty across new starts — but that constraint doesn't shrink the total volume of work. It defers some of it, concentrates work in markets with better grid access, and creates volatility in start dates that subs have to absorb.

For a trade firm deciding whether to build toward this market: the ramp typically takes 18 to 24 months from the decision to pursue it to a first significant award from a hyperscaler-tier client. That time goes into assembling the right crew, accumulating the qualifying project experience, building the documentation infrastructure that procurement teams ask for, and getting through the evaluation process.

The firms already booked solid on this work made that investment in 2024. The question for a firm looking at the pipeline today is whether the investment is worth it given current commercial backlog. For a firm with clean safety records, capacity in electrical or mechanical disciplines, and tolerance for the learning curve — the margin premium on critical facility work, combined with the volume and duration of hyperscaler programs, makes the arithmetic favorable. It just doesn't happen fast.

Forward this to the owner or VP at an electrical or mechanical sub who's been watching the data center pipeline but hasn't committed to pursuing it.

Construction AI Brief covers the AI moves that change contractor decisions three times a week. Subscribe at constructionaibrief.com.