Data Center Site Selection: How Operators Choose Where to Build

Data center site selection has changed more in the past three years than in the previous twenty. Where operators once chased fiber hubs and cheap land near big cities, they now chase megawatts — and increasingly build where power exists rather than where customers live. This article, part of the Data Center Economy hub, explains the criteria operators use, how they weigh them for AI versus traditional workloads, and provides a practical checklist. For cost implications, see data center construction costs per MW; for the power picture, read why power is the new bottleneck.

What factors determine where a data center is built?

Operators assess power availability and cost, grid reliability, fiber connectivity and latency, land and zoning, tax incentives, water and climate, natural-hazard risk, construction costs, labour, regulation and community support. Power now outweighs everything else for large AI campuses.

Cushman & Wakefield’s 2026 Global Data Center Market Comparison scores markets on 24 variables grouped around operational IT load, power availability and delivery, grid stability and the pipeline of preleased capacity. Its top-ranked primary markets are Northern Virginia, Atlanta, Dallas, Columbus and Phoenix, followed by Oregon, Chicago, London, Frankfurt and Shanghai. Among secondary markets it ranks Austin/San Antonio, the Carolinas, Salt Lake City, Reno and Omaha highest.

Weightings differ by workload. A retail colocation operator serving enterprises prioritises carrier density and proximity to customers. A hyperscaler building a cloud region needs low latency to population centres plus multiple sites for redundancy. An AI training campus needs hundreds of megawatts to gigawatts of power and can sit far from cities. Our guide to data center business models explains how these customer types differ.

Source: JLL North America Data Center Report Midyear 2026; Cushman & Wakefield; Datacentres.com transaction tracker; Baxtel; Virginia JLARC; Data Center Watch.

Why is power availability the number-one site selection criterion?

Because AI campuses need hundreds of megawatts, and grid connections in established hubs can take years. Operators now select sites where utilities can deliver large blocks of power quickly, or where they can build on-site generation, and accept trade-offs on latency and land.

The shift is visible in the construction pipeline. JLL’s midyear 2026 report counted more than 66 GW under construction in North America, with roughly 77% in frontier markets led by West Texas, plus Ohio, Louisiana, Indiana and the Carolinas. Cushman & Wakefield lists West Texas, Pennsylvania, Alberta and Cheyenne among emerging Americas markets. A DC Hub market note in July 2026 highlighted Midland–Odessa, Michigan’s Upper Peninsula and Cheyenne as places with high power-availability scores attracting institutional bids.

Established hubs remain enormous but constrained. Cushman & Wakefield put Northern Virginia at 12,338 MW of operational capacity, including self-build, in June 2026, and Virginia’s legislative watchdog JLARC projected that unconstrained power demand in the state would double within ten years, driven mainly by data centers. Operators also evaluate utility tariffs, large-load rules and whether they will bear grid-upgrade costs — issues covered in who pays for grid upgrades. Some sites pair with dedicated generation, including the nuclear deals Big Tech has signed.

How important are fiber connectivity and latency?

Fiber and latency remain critical for interconnection, cloud regions, financial services, gaming and AI inference, which need short round-trip times to users. Large training clusters are more tolerant because they exchange data mostly inside the campus, not with end users.

Operators look for diverse long-haul fiber routes entering the site from different directions, proximity to carrier hotels and internet exchanges, and the cost of building new laterals. Retail colocation leaders such as Equinix build their business model around dense interconnection, which is why their sites cluster in major metros. Cloud regions typically comprise several facilities within a metro area so customers can replicate data across them with low latency.

AI is splitting the map. Training can move to remote, power-rich regions, while inference — serving answers to users in real time — is expected to grow fastest. JLL expects inference to overtake training around 2027, which could pull more capacity back towards metro edges. See US data center investment for how this is playing out from Virginia to Texas.

How do land, zoning and construction costs affect site choice?

Large AI campuses need hundreds or thousands of acres with appropriate zoning, room to expand and access to substations. Construction costs, labour availability and permitting timelines vary widely by market and can change project economics significantly.

Recent land purchases show the scale. Digital Realty acquired about 1,440 acres at Astra Enterprise Park near Kansas City for about $475m in June 2026, according to the Datacentres.com tracker. Microsoft bought roughly 600 acres in Woodward, Iowa, about 30 miles from Des Moines, for around $37.4m in October 2026 for a potential gigawatt-scale campus, according to Baxtel. Operators prefer land zoned for data center use by right, which avoids discretionary hearings that can delay or derail projects.

Building costs are rising. Cushman & Wakefield’s 2026 development cost guide reported average greenfield costs of $17.6m per MW for the most modern facilities, up 21% per MW since its previous report. Turner & Townsend’s 2025–2026 cost index surveyed 250 experts across 52 markets. Labour is another constraint: JLARC found a single large site may have about 1,500 workers at peak construction, though a typical 250,000-square-foot facility employs only about 50 full-time staff once operating. Our water, land and jobs statistics provide more detail.

What role do tax incentives play in data center site selection?

Tax incentives — especially sales and use tax exemptions on servers and equipment, plus property tax abatements — can be decisive because IT equipment is replaced every few years. Many US states and countries offer them, though political scrutiny of their cost is rising.

Virginia illustrates both sides. JLARC found that the state’s data center sales and use tax exemption, in place since 2010, saved the industry about $928m in fiscal 2023, with roughly 90% of the industry using it; the exemption is scheduled to expire in 2035. JLARC also estimated the industry adds about $9.1bn to Virginia’s GDP annually and supports around 74,000 jobs and $5.5bn of labour income, mostly through construction.

Internationally, incentives sit alongside data localisation rules and energy policy. Countries from Malaysia to Brazil and the Gulf states have used tax breaks, special zones and power deals to attract capacity, as covered in our guides to Asia-Pacific, Latin America and Africa and Türkiye. Operators should model the after-tax cost of each site over the full equipment refresh cycle, not just the first build.

How do water, climate and natural hazards factor in?

Operators evaluate water availability and rights, the number of hours suitable for free cooling, extreme heat, and exposure to floods, earthquakes, hurricanes and wildfires. Cooler, drier climates reduce cooling energy, while water-stressed regions push designs towards closed-loop or air-based systems.

Cooling design and location are linked. Evaporative systems use less electricity but more water; closed-loop liquid cooling, increasingly needed for dense AI racks, can cut water draw substantially. JLARC found that most Virginia data centers use about as much water as a large office building or less, though a few use substantially more, and that oversight of how water is shared among local uses is limited. Our guides to cooling technologies and sustainability explain the trade-offs.

Natural-hazard screening includes floodplain mapping, seismic zones, wind loads and wildfire risk. Insurers and lenders increasingly require climate-risk assessments, and some customers specify minimum distances from hazards or require sites in separate risk zones for disaster recovery.

How do community opposition and regulation affect site selection?

Local opposition and regulatory limits can delay or kill projects even where power and land are available. Concerns typically centre on noise, water, electricity prices, land use and the small number of permanent jobs relative to the size of the buildings.

Data Center Watch found that $18bn of US data center projects were blocked and $46bn delayed over roughly two years to March 2025, with at least 142 activist groups across 24 states. Its later briefings cited examples such as a 5–0 vote denying a project in Archbald, Pennsylvania, a withdrawn campus in Apex, North Carolina, and a 150-day moratorium proposed in Oldham County, Kentucky. Its Q2 2026 report described sustained disruption amid expanding local opposition.

Regulation also operates at national level. Ireland’s data centers used 22% of metered electricity in 2024, according to the Central Statistics Office, prompting tight grid connection policies in the Dublin area. Singapore paused and then selectively reopened new capacity. Operators therefore assess political risk, permitting timelines and the likelihood of new rules on energy, water or data sovereignty. See our Europe investment map for country-level policies.

What does a data center site selection checklist look like?

A practical checklist screens sites in stages: first power and land, then connectivity and hazards, then incentives, costs and community factors, before detailed due diligence. Eliminating sites early on power saves the most time and money.

Stage 1: Fatal-flaw screen

Confirm the utility can deliver the required megawatts and by when; confirm parcel size, ownership and zoning; check flood, seismic and other hazard maps; and identify any moratoria or pending regulation.

Stage 2: Technical fit

Map fiber routes and carriers, measure latency to target users and cloud regions, assess water supply and cooling options, and evaluate substation and transmission upgrades needed. Consider on-site generation or storage where grid timelines are long.

Stage 3: Economics and risk

Model construction costs per MW, power tariffs over 15 years, incentives over the equipment refresh cycle, labour availability and insurance. Engage local officials and residents early. Finally, test financing: JLL reports that hyperscaler-leased projects can borrow at low-200 basis point spreads with up to 85% loan-to-cost, while non-investment-grade projects pay about 200–300 basis points more — so tenant and site quality feed directly into cost of capital. Investors can read how to invest in data centers and our forecast to 2030 for the demand side.

Frequently Asked Questions