SECTOR FOCUS — DATA CENTER, HYPERSCALE & CRITICAL INFRASTRUCTURE
Power is the schedule. Everything else runs downstream of it.
This is where the majority of construction capital is moving right now, and it's where we're spending the most attention.
This page reflects general industry practice and BuildIQ Advisors' professional perspective as of publication. It is not tailored to any specific project, does not constitute engagement-specific advice, and does not create an advisory relationship. Verify current codes, standards, and site-specific conditions independently before acting.
CURRENT CHALLENGES
We track this market closely — it's currently absorbing more construction capital and more construction complexity than any other sector in the country. Nearly 100 gigawatts of new data center capacity is coming online between 2026 and 2030, and the five largest hyperscalers alone are putting $600–775 billion into capital expenditure this year — a surge on the U.S. grid unlike anything since the 1980s.
Power and interconnection — the real critical path.
Grid interconnection is running 36 to 84 months in many markets, against a 12-to-24-month construction cycle — in Northern Virginia, the largest data center market in the world, a new grid connection can take more than seven years. That mismatch is why 30–50% of the capacity planned for 2026 is at real risk of slipping to 2028 or later — not because anyone can't build fast enough, but because the power isn't there yet. Hyperscalers have responded by becoming energy developers themselves — industry reporting puts more than 130 gigawatts of energy resources as proposed nationally to serve planned data center projects. Speed-to-power — the time from project commitment to live electrical capacity — has become the dominant site-selection metric for 2026, ahead of land cost or even fiber proximity. That reshapes the whole program: procurement and permitting run in parallel with design far more aggressively than on a conventional building, because the window to correct a scheduling mistake keeps getting thinner.[3][4][5]
Long-lead electrical equipment — the schedule's real long pole.
Switchgear lead times for standard medium- and low-voltage equipment have stretched to 44–54 weeks, up from 20–30 just a few years ago — custom and high-voltage configurations can run as long as 80 weeks. Large power transformers are running roughly 128 weeks. Generator step-up units are near 144 weeks. Cummins said in February 2026 that its backlog for high-horsepower generator sets extends into 2028, with active capacity conversations already running into 2029 and beyond. That means switchgear, transformers, and generators get committed against a projected in-service date before design is even fully resolved — a fundamentally different procurement posture than a conventional building, where equipment typically follows a completed design. Get that sequencing wrong and the long-lead item, not the building shell, sets the critical path.[3][6]
Cooling architecture and density — two mechanical plants, not one upgraded.
AI workloads have pushed rack densities from a traditional 10–15 kW to 100–200-plus kW per rack. Liquid cooling — direct-to-chip, immersion, CDU-based — is now standard for high-density halls, cutting total facility power draw 20–40% against air-only systems. A modern high-density hall runs two independent mechanical plants, air-side and liquid-side, each with its own controls, water treatment, and failure-mode logic, that have to work together and fail safely independently.[7][8]
Commissioning rigor — Level 5 Integrated Systems Testing.
Data center commissioning runs a five-level methodology, culminating in Level 5 Integrated Systems Testing — every critical system tested together, under simulated failure scenarios, against the Owner's Project Requirements and Basis of Design. On a complex facility, Level 5 alone can run 20-plus working days of dedicated engineering time, sequenced against a hard energization date the power utility and the tenant have already committed to.[9][10]
Multi-site program cadence.
Hyperscale doesn't build one building — it runs a program: a repeating design, a consistent set of vendors and workflows, executed across many sites under central governance, so the speed gained at site six doesn't cost the quality standard set at site one. Land partnerships with pad-ready sites — power pre-secured, permitting pre-completed — are reported to compress individual-site timelines by 12 to 18 months, and that only works if the delivery process behind it is genuinely repeatable.[11][12]
Speed-to-market and what it does to procurement.
The same compression that shortens the calendar pushes major procurement and buyout decisions earlier against less-complete design — raising the cost of a wrong call made under time pressure. Initial phases get designed for the fastest possible delivery while longer-lead infrastructure — substations, water systems — gets built to support the program's later expansion.[11]
Labor, concentrated and competed-for.
A single hyperscale campus can pull 4,000 to 5,000 workers at peak — electricians, mechanical contractors, controls specialists, commissioning teams, high-voltage field talent — landing on a national labor market that needs roughly 349,000 net new construction workers this year just to keep pace. When multiple campuses break ground in the same region, local skilled-trade pools are gone within months.[13][14][15]
WHAT SHOULD BE CONSIDERED TO OVERCOME THEM
HOW IT'S ACTUALLY GETTING BUILT
Behind-the-meter (BTM) / on-site generation
Proven at real scale — not a pilot
Roughly 56% of developers are now pursuing co-located or on-site generation rather than waiting on the interconnection queue — natural gas turbines and reciprocating engines dominate because they can be procured and installed faster than a utility substation upgrade. GE Vernova alone expects 20 gigawatts of annualized gas turbine output in 2026, and Texas has an estimated 38 gigawatts of behind-the-meter investment in development against roughly 356 gigawatts of ERCOT interconnection requests. The tradeoff: on-site generation buys schedule, but it relocates the power problem rather than solving it — fuel-supply contracting, air-permit and emissions review on a compressed timeline, on-site storage and safety-code compliance, and, if the plant is meant to be temporary bridge power, a second capital commitment and a second commissioning cycle when permanent grid power eventually arrives. A program that treats on-site generation as a schedule fix without pricing the emissions-permitting and fuel-logistics workstream as its own critical-path item finds that critical path during air-permit review, not during construction.[24][25]
Nuclear PPAs and SMR agreements
Real for reactor restarts. Unproven for SMRs — none operational yet
Nuclear PPAs and SMR agreements get announced constantly — every major hyperscaler has signed at least one, roughly 9.8 gigawatts committed across 13 disclosed projects by mid-2026. For restarting or re-contracting an existing large reactor, that's real, contracted power on a real timeline. For SMRs specifically, no U.S. SMR is commercially operational anywhere yet, and new-build timelines run six to ten years. A nuclear PPA announcement and an operating nuclear PPA are two different deliverables — a program schedule that depends on SMR power inside a three-to-five-year construction window is pricing a technology that hasn't shipped as if it had.[26]
Brownfield siting on retired generation
Real and executed — verification-dependent
Reusing the interconnection point at a decommissioned coal plant, a retired industrial site, an idle enrichment facility is real and executed, not theoretical; it solves the interconnection-queue problem specifically, which is the whole point. But “existing interconnection” at a retired site is a legal and technical starting point, not a finished asset — legacy equipment condition, actual available capacity, and historic load rights all need independent verification before they convert into usable capacity, and a Phase I/II environmental workstream that surfaces contamination erases the schedule advantage the site was purchased for, after the acquisition money is already spent.[27]
Demand-response and curtailment agreements
Emerging, not yet proven at hyperscale volume
Trading controllable demand for faster interconnection is contracted and real in specific cases (Google alone carries roughly a gigawatt of demand-response capacity across multiple utilities), but a single 96-megawatt pilot facility isn't yet the template for a gigawatt campus. Where it's contracted, it works. The catch is architectural: a facility engineered to tolerate load-shedding without disrupting live workloads has to be designed that way from the switchgear and controls level up — that decision has to be made before the electrical design locks, not bolted on after commissioning as a software feature.[28]
HOW IT'S ACTUALLY GETTING BUILT
Standardizing on widely available equipment sizes
Works — the most-repeated advice in the field
The single most-repeated piece of practical advice in the field right now: standardize on widely available equipment sizes — 3-megawatt generators, 3-MVA transformers are the cited sweet spot — instead of custom-engineered gear, because the manufacturer is already running that SKU at volume. The tradeoff: the electrical design now has to fit the equipment, not the reverse, which is a legitimate engineering compromise but has to be an explicit, early, owner-signed-off decision — not something a design team backs into after the transformer's already ordered and the room's already poured around it.[29]
Owner-furnished early procurement
Works — increasingly the only way the math works
The owner buys the long-lead gear directly and furnishes it to the GC, decoupling the order from design completion — isn't optional anymore given the math: a conventional design-then-procure sequence doesn't fit inside a 12-to-24-month construction window against a 128-week transformer lead time. The tradeoff: the GC loses single-point accountability for that equipment. A defective or misspecified owner-furnished transformer means the GC didn't select it, can't warranty it through their own sub-tier, and has to coordinate a fix with a vendor they hold no contract privity with, while the schedule keeps running — a contract term (inspection and acceptance protocol, damage liability, storage responsibility) that has to be settled before the equipment ships, not when it arrives damaged.
Modular and skid-mounted electrical rooms
Proven — moving from option to default
Power skids and cooling plants assembled and factory-tested off-site are moving from an option to the default on large projects; one national mechanical contractor's data-center-driven prefab revenue was up more than 56% year-over-year in Q1 2026, a market signal, not a vendor claim. The tradeoff mirrors every prefab tradeoff: speed bought at the cost of early, hard-to-reverse design commitment. A change order on a modular electrical room means re-opening a factory build, not modifying a wall — and a fabrication delay or a failed factory acceptance test sits off the critical path a site team can actually see, in someone else's shop.[30][8]
HOW IT'S ACTUALLY GETTING BUILT
Closed-loop and zero-water liquid cooling
Works, becoming standard — the power-vs-water tradeoff is genuinely unsettled
Closed-loop and zero-water liquid cooling is becoming standard for new high-density builds — the water benefit rides along with the density-driven adoption of liquid cooling anyway. The tradeoff is real and genuinely unsettled: eliminating water largely trades it for electricity, since air-cooled chillers and dry coolers use materially more power than evaporative systems to reject the same heat load — in a sector where power, not water, is already the harder constraint. A site that solves its water exposure by going all-air-cooled can make its power problem worse at exactly the moment power is hardest to solve. This is genuinely site-specific: a dry, power-constrained site and a water-constrained, power-available site want opposite answers, and a single blanket recommendation across every site means nobody actually looked at the site.[31]
Reclaimed and non-potable water sourcing
Works only where the infrastructure already exists nearby
It doesn't create that infrastructure. A site without reclaimed-water utility access doesn't get this option without funding new infrastructure itself, which reintroduces the exact permitting and construction timeline the site was probably selected to avoid in the first place.
HOW IT'S ACTUALLY GETTING BUILT
Prefabrication and repeatable program design
Schedule gains are proven. Quality-at-scale is contested, not settled
The 30-to-50% schedule compression figures for modular and prefabricated delivery apply across prefabricated mechanical and structural components generally, and the tradeoff pattern holds everywhere it's used: speed bought with reduced field flexibility and earlier, harder-to-reverse commitment. One genuinely open question in the field right now: whether a repeatable-design program actually holds its quality standard at site six the way it did at site one without the central-governance overhead eating the schedule gain it was supposed to produce. That claim gets made confidently in vendor and trade-press material; independent, adversarial reporting testing it program-wide — not just on the flagship examples that get written about — is thinner. Treat “repeatable design scales cleanly” as the industry's stated intent, not yet a fully proven universal outcome.
HOW IT'S ACTUALLY GETTING BUILT
Prefabrication and regional labor strategy
Solves individual projects — no structural fix for the shortage itself
Prefabrication is the dominant lever — it moves trade hours out of a labor-constrained jobsite into a factory, which changes the labor problem's geography, not its total hours. Regional labor strategy and travel crews are the other lever in active use — Microsoft has publicly confirmed flying in electricians from 75-plus miles away or temporarily relocating them to keep projects moving, the plainest evidence this constraint is being solved by labor logistics right now, not by a structural fix. Worth saying plainly rather than smoothing over: travel-crew and relocation strategies solve one project's staffing gap by pulling skilled labor out of another region's available pool — the same mechanism that makes healthcare's competition with data centers for the same electricians worse, not better, region by region. Training pipelines are the only genuine structural fix, and they run on a multi-year timeline that doesn't help a 2026–2027 program at all. On labor specifically, the industry doesn't currently have an answer that beats the shortage — only ways to win the competition for the labor that exists.[32]
WHAT WE WOULD HELP THEM NAVIGATE
- Confirming whether the power plan is grid-dependent, behind-the-meter, or hybrid, and pressure-testing who owns the interconnection application, the utility relationship, and the risk if it slips — before that ambiguity becomes a month-eight discovery instead of a week-one answer.
- Verifying long-lead electrical procurement against confirmed factory slots, not quoted lead times, and settling the contract terms an owner-furnished equipment strategy requires — inspection protocol, damage liability, storage responsibility — before the gear ships, not after it arrives.
- Holding commissioning rigor accountable to the actual energization date the utility and tenant have committed to, not a soft internal target, and confirming Level 5 Integrated Systems Testing is sequenced against that real date.
- Testing whether a program's repeatable-design claim is actually holding site-to-site, rather than assuming site six will perform like site one without the same governance attention that made site one work.
- Confirming a site's water-and-cooling strategy is resolved for its specific climate and utility access, not applied as a generic corporate spec that quietly trades one binding constraint for the other.
The same program discipline behind directing $200–300 million a year across 185 projects, 125-plus active concurrently, for a major public capital program — one design standard, one delivery process, centrally governed across volume, and long-lead procurement sequenced years ahead of need — applies directly to data center and hyperscale programs, which carry that same multi-site program governance and that same long-lead-equipment procurement discipline, at a compressed timeline. Direct experience holding cost and schedule together on a large, security-driven institutional program — complex MEP, advanced security systems, secure detention, delivered under genuine external disruption — is the same kind of complexity a compressed, speed-to-power-driven data center program throws at the people running it.
WEEK-ONE QUESTIONS
WEEK ONE — WHAT WE'D ASK ON A PROGRAM LIKE THIS
- 01
Is the power plan grid-dependent, behind-the-meter, or hybrid — and who owns the interconnection application, the utility relationship, and the risk if it slips? That answer alone determines whether the GC is managing a conventional construction schedule or co-managing an owner-run energy-development program running in parallel.
- 02
What is the actual committed delivery date for every long-lead electrical item — not the quoted lead time, the confirmed factory slot — and is the design frozen enough to hold that order? A quoted lead time and a confirmed slot are different facts. Only one of them is a schedule commitment.
- 03
Is this site's water and cooling strategy resolved for this specific site's climate and utility access, or is it a standard corporate spec applied without a site-specific tradeoff analysis? Given how directly cooling choice trades power for water, a generic answer here is itself a red flag.
- 04
What labor is actually committed under contract for the peak months, versus assumed available in the regional market? “We'll staff it” is not a plan. A named commitment with a travel or relocation contingency is.
- 05
If any part of the power strategy depends on nuclear, SMR, or a demand-response program that isn't yet operating at scale, what's the fallback if that specific technology or contract slips — and is that fallback priced into the program, or is it a hope?
Professional Services Disclosure
BuildIQ Advisors provides construction advisory and consulting services under signed engagement agreements, performed to the standard of care customary for the industry. We are not a licensed architecture, engineering, accounting, or law firm — advice requiring those licenses should come from one. Engagement terms govern each project.