Per-MW pricing, regional variance, and cost drivers for owners scoping hyperscale & AI builds.
Salary benchmarks across the 14 mission-critical disciplines.
If I had to sum it up in one line: most U.S. data centers in 2026 cost far more per MW than early budgets assume.
I’d put the starting ranges like this:
The main reason is simple: the building shell is only a small part of total cost. In many projects, electrical and mechanical systems make up 60%–75% of the budget, while the shell often sits at just 12%–18%. So if you price by square footage first, you can miss the biggest cost items.
Here’s what I’d watch first if I were budgeting a project:
A few numbers stand out:
Bottom line: if you want a budget that holds up, I’d focus less on the shell and more on power, cooling, redundancy, procurement timing, and utility scope. That’s where the money usually goes.
Data Center Build Costs by Facility Type & Tier Level (2026)
Budget accuracy starts with the cost mix. In most data center projects, power, cooling, and utility infrastructure drive the bulk of spend, not the building shell. A simple way to think about the budget is through four buckets: site, shell, MEP, and schedule risk.
Site conditions can change the math fast. Poor soils, floodplain exposure, and seismic requirements push up foundation and engineering costs.
The shell and core - structural steel, roof, walls, and data hall build-out - usually makes up only 12%–18% of total project spend [1]. That includes the main building frame and enclosure, but it’s still a smaller piece of the budget than many teams expect. The shell matters, but it’s not where most of the money goes.
This is where the big dollars sit. Electrical and mechanical systems account for 60%–75% of total spend [1].
Here’s how that budget often breaks down:
Electrical is usually the biggest line item, especially when utility work, generators, UPS, and distribution all stack together. Mechanical costs follow close behind, and AI-focused builds can push cooling costs higher. If a project needs to be liquid-cooling-ready, that alone can add 10%–20% to the design [1].
Utility interconnection can also become the schedule bottleneck. In high-demand markets, grid wait times can stretch past four years [1]. Add in substation upgrades or transformer procurement, and the cost can jump again. At that point, procurement risk stops being just a purchasing problem and turns into a schedule problem too.
Long-lead equipment now drives the critical path on many projects. Generators and transformers usually come with lead times of 12–18 months, while switchgear, UPS systems, and chillers often land in the 25–55 week range [1]. That’s why early procurement matters so much. If key equipment shows up late, a project can lose a year - or more - before core systems even hit the site [1].
And that delay gets expensive in a hurry. A delayed 60 MW facility can mean more than $14.2 million per month in lost revenue [1].
Commissioning talent sits inside soft costs, but it’s far from a minor item. It confirms that the facility is ready for live load before a customer’s workload goes live [1].
Facility type gives you the baseline. After that, density and redundancy are what move the budget. Two data centers can have the same square footage and still land in very different cost ranges. The big reasons are power density, cooling design, and how much backup capacity the site needs.
Facility type changes the cost structure in clear ways. Enterprise builds often aim for Tier III or Tier IV performance, and greenfield projects usually take 24–36 months to deliver. Colocation conversion projects tend to move faster, often in the 12–20 month range, because the focus is speed to market and tenant-ready fit-out. Hyperscale and AI-ready campuses sit at the top end of the cost curve, with modular delivery often taking 16–20 months.
A powered shell pushes more CapEx into the later tenant fit-out stage. That gives developers room to wait on part of the IT infrastructure spend until demand is confirmed.
Once rack density goes up, the project often moves into a different cost bracket.
Rack density is where the gap between a standard facility and an AI-ready build gets hard to miss. Conventional air-cooled designs support 5–15 kW per rack and serve as the baseline for both cost and build complexity. Liquid-ready infrastructure that can handle 15–50 kW per rack may add 10–20% to mechanical cost [1]. AI-optimized builds at 50–150 kW per rack need direct liquid cooling, specialized commissioning, and budgets of $20M+ per MW [1].
Verify utility capacity early. If the utility can’t support the load, the rest of the design discussion starts to fall apart.
After density, redundancy becomes the next big cost multiplier.
Redundancy gets expensive fast. More redundancy means more generators, more UPS modules, more cooling paths, and more switchgear. It also means more engineering work to make sure those systems keep running under failure conditions. Moving from Tier 2 to Tier 3 adds 15–25% to total project cost. Moving from Tier 2 to Tier 4 adds 35–50% [1]. Most of that premium sits in the electrical topology and cooling distribution, which are already the biggest budget items.
Tier 4 costs more because fault tolerance requires fully independent systems, duplicate distribution, and separate cooling loops [1]. For many projects, Tier 3 is the practical middle ground.
The scenarios below are planning examples, not quotes. Actual budgets change by site, market, utility access, labor, and supply chain conditions.
Here, the earlier per-MW ranges are translated into project-level budgets based on size and market.
A small enterprise facility in a secondary Midwest market might support 0.5–1.5 MW of IT load across about 8,000–20,000 sq ft, with a Tier II or Tier III target. At first glance, that may sound modest. But a greenfield site that needs a new medium-voltage utility feed or an on-site transformer yard can push costs much higher than the building footprint suggests. For a 1 MW build in this kind of market, the total budget often lands around $8M–$12M all-in. In coastal markets, costs can climb even before the design gets more complex.
As project phases get larger, utility scope and redundancy tend to push the budget up faster than square footage does.
A mid-sized colocation build with a 5 MW initial phase - roughly 40,000–80,000 sq ft of rentable white space - usually sees its biggest cost pressure from power and cooling infrastructure. That includes large generator sets and paralleling gear, high-capacity UPS systems and medium-voltage switchgear, modular or chiller-based cooling plants, and utility-side upgrades like new feeders or on-site substations. At this scale, teams often budget around $10M–$14M per MW. Total spend climbs if the site needs heavier utility work or sits in a premium labor market.
Once you move to campus scale, interconnection and substation work often become the main budget gate.
A phased hyperscale campus targeting 50–150+ MW across three to four buildings shows how fast total spend can snowball. Assume Tier III-equivalent redundancy, a hybrid cooling setup that pairs air-cooled whitespace with direct-to-chip liquid cooling for GPU-heavy AI blocks at 30–80 kW per rack, plus a new on-site substation.
For conventional air-cooled phases, hard costs run about $7M–$9.5M/MW. AI-optimized phases with liquid-cooling readiness push hard costs to $9M–$12M/MW, bringing total budgets to about $8.5M–$12M/MW for traditional phases and $10.5M–$15M/MW for AI-ready phases [2].
At this scale, utility and substation work often become the largest non-building costs. New on-site substations, offsite transmission or distribution upgrades, and interconnection timelines of 18–36 months can add tens to hundreds of millions of dollars to the full project budget. Just as often, they control the critical path. That’s why early utility engagement and a realistic utility will-serve letter should be treated as core preconstruction milestones. This alignment is critical as construction teams prepare for DOE data center projects and other high-capacity energy requirements.
Regional conditions still shift budgets. In Northern Virginia, permitting can run past 12 months. In Texas, permits often move in 3–6 months, which can reduce general conditions and schedule risk.
In 2026, data center cost comes down to power, redundancy, cooling, and utility access - not just building size. Cost ranges follow scope, not shell square footage. And that makes sense when you look at where the money goes: electrical and mechanical systems account for 60% to 75% of a typical data center budget [1]. So if you're using square footage as the main pricing shortcut, you're probably missing what drives total spend.
Electrical work is usually the biggest cost driver. That means delays tied to transformers, switchgear, and interconnection can hit both the budget and the schedule. In plain terms, good estimating alone isn't enough. Delivery discipline matters just as much.
This is where experienced data center teams earn their keep. MEP coordinators, commissioning managers, schedulers, and mission-critical estimators help cut cost and schedule risk by spotting clashes early, lining up long-lead equipment in the right sequence, and protecting the Ready for Service date.
The cheapest fix is almost always the one you avoid in the first place. AI density needs to be addressed at the concept stage, not after design freeze. If you wait and later need to retrofit air-cooled space for liquid cooling, you're looking at $200–$400 per kW [1]. That's a painful way to learn the value of early planning. Getting ahead of that kind of change helps avoid rework and keeps ROI in better shape.
Accurate cost planning isn't about pretending you can predict every line item on day one. It's about pricing the real risk drivers early and putting the right people on the job to keep them under control.
Early data center budgets often leave out land acquisition, interconnection requests, and early design fees. Why? Those costs may show up before the formal construction budget even starts.
They also tend to underprice MEP scope. And that’s where things can go sideways fast.
Small-looking items get missed all the time, including:
Redundancy targets like N+1 or 2N can also push costs higher. In many cases, they add structural support needs that don’t appear as clearly priced line items in the early budget.
Start with your facility type, target IT load, and tier level. For standard enterprise or colocation projects, costs often land around $8 million to $12 million per MW. If you're planning an AI-focused build with high rack density and liquid cooling, that number can climb to $15 million to $25 million+ per MW.
From there, adjust the estimate based on rack density, cooling setup, electrical and mechanical costs, 2026 market price increases, and the long lead times tied to major equipment.
Plan for liquid cooling at the early design stage, especially in the backbone infrastructure. Waiting and trying to add it later usually costs a lot more - often $200–$400 per kW - and it can force tricky structural or electrical upgrades to support high-density AI workloads.
Building for liquid cooling from the start adds only 10%–20% to mechanical costs. A retrofit after construction is a different story. It can mean disruptive work inside live environments, more risk, and a bigger total capital bill.
