Data Center Site Selection: The Complete Developer's Guide
Power, land, connectivity, permitting, and cooling — the full site criteria checklist for institutional data center developers.
Data center development has become one of the most technically demanding site selection exercises in real estate. A site that works for industrial or multifamily development will fail for a hyperscale or colocation build if even one critical infrastructure variable is missing.
This guide covers each selection criterion, what institutional developers are actually evaluating, and where AI is compressing the analysis.
The Five Core Criteria
1. Power: The First Filter
Power availability is the dominant filter in data center site selection — and the one most likely to kill a deal.
What to evaluate:
Available load from the local utility, measured in megawatts (MW). Hyperscale builds typically require 100 MW or more at site level; edge and colocation facilities run between 1 MW and 30 MW.
Utility reserve margin — the headroom the local grid has to serve new large loads without major infrastructure upgrades.
Proximity to transmission infrastructure (345 kV, 230 kV, or 138 kV lines depending on load size).
Interconnection queue position. In constrained markets, studies can take 24-36 months and result in significant upgrade obligations.
Time to energize — the period from utility application to power delivery. This ranges from 18 months in favorable markets to 5+ years in congested ones.
PJM, ERCOT, MISO, and the Western Interconnection each behave differently. A developer with sites across multiple regions needs market-specific queue data, not a single national summary.
What AI handles: Querying FERC interconnection data, utility reserve margin reports, and load availability databases across multiple ISOs simultaneously. What takes a consultant days to compile can be assembled in hours.
2. Land: Acreage, Zoning, and Control
The land requirement for a hyperscale campus ranges from 100 to 500+ acres, depending on phasing, power infrastructure, and setbacks.
What to evaluate:
Parcel size and contiguity. Fragmented ownership requires multiple acquisitions and creates assembly risk.
Zoning classification and conditional use requirements. Data centers sit in a grey zone in many jurisdictions — technically industrial use but with energy and infrastructure demands that often require special permits.
Environmental constraints: wetlands, flood zone designation (FEMA FIRM maps), proximity to regulated waterways, and Phase I/II findings.
Control: can the developer secure an option, LOI, or purchase contract ahead of utility commitments?
Developers are increasingly looking beyond primary markets — Northern Virginia, Phoenix, Dallas, Chicago — into secondary and tertiary markets where land is available, zoning is flexible, and utilities are less congested.
3. Connectivity: Fiber and Network Infrastructure
A data center without fiber access has no customers. Connectivity is equally critical to power for most tenants.
What to evaluate:
Proximity to diverse fiber routes and carrier-neutral access points.
Availability of dark fiber for high-bandwidth requirements.
Distance to major internet exchange points (IXPs).
Latency to major population centers — relevant for edge deployments and financial services tenants.
Number and diversity of network service providers available at the site.
What AI handles: Mapping fiber routes against a site, identifying carrier points of presence, and flagging connectivity gaps before a developer commits to land.
4. Permitting and Entitlement
Permitting timelines for large data centers range from 6 months to 3+ years depending on jurisdiction, scale, and local political dynamics.
Key variables:
Building permit complexity — data centers at scale require special inspections and electrical reviews that add time regardless of jurisdiction.
NEPA or SEPA review requirements if federal or state land is involved.
Transmission-level permitting if the developer is building dedicated substation or transmission infrastructure.
Community sentiment. Municipalities with prior data center approvals move faster. New-to-market jurisdictions can be unpredictable.
Several states — including Virginia, Texas, and Georgia — have streamlined permitting pathways for digital infrastructure. Others have imposed moratoriums in response to community opposition.
5. Cooling and Water
Cooling accounts for 30-40% of data center operating costs. The available cooling approach is site-dependent.
What to evaluate:
Climate. Air economization is viable in most of the northern US, Canada, and Northern Europe. Hot, humid climates require more mechanical cooling.
Water availability for evaporative cooling systems. Data centers in water-stressed regions face increasing regulatory and ESG pressure on consumption.
Proximity to natural cooling resources — rivers, ocean water, underground aquifers — for hyperscale builds.
Local regulatory requirements on water discharge and treatment.
Hyperscalers are increasingly moving to direct liquid cooling (DLC) and immersion cooling for high-density GPU clusters. These approaches reduce water dependency but require different site infrastructure assumptions.
Secondary Criteria
Labor market: Data centers require specialized trades during construction — electrical, controls, low-voltage — and ongoing facility technicians post-completion.
Natural hazard exposure: Seismic risk, flood zones, tornado corridors, and wildfire adjacency all affect insurance costs and redundancy design requirements.
Tax incentives: Virginia, Georgia, Nevada, and several other states offer significant sales tax exemptions on equipment. Some markets offer real property tax abatements for large industrial users.
Transportation: Access to major highways and airports matters for equipment delivery and staffing.
How AI Is Changing the Evaluation Process
Site selection for data centers has historically been a months-long manual process. A development team would commission power studies, hire brokers to identify parcels, retain environmental consultants for Phase I work, and engage utility representatives — often in parallel, with poor information sharing between workstreams.
AI is compressing each of these steps.
Power analysis: Automated querying of ISO data, utility GIS layers, and interconnection queue filings can surface load availability at a specific point of interconnection within hours. What previously required a utility pre-application meeting and a 3-6 week response can be modeled as a first pass before any utility contact.
Site screening: AI can run multi-variable site criteria against parcels across a target geography in a single pass — evaluating acreage, zoning, fiber proximity, flood zone designation, and ownership structure simultaneously.
Environmental risk scoring: AI tools can flag obvious risk factors before commissioning a formal Phase I study — ASTM-defined RECs, historical industrial use, regulatory database hits.
Market comparison: Developers can model power cost, incentive value, permitting timeline, and construction cost across markets to rank site candidates on total development economics, not just land availability.
The decisions at each gate still require developer judgment. Power negotiations, community relations, and capital structure remain human work. AI handles the information assembly that feeds those decisions.
The Site Selection Checklist
For a development team evaluating a potential data center site, the minimum diligence set before any capital commitment:
Confirm available load from the utility (MW, timeline, upgrade cost estimate)
Validate transmission infrastructure proximity and interconnection queue position
Confirm parcel control path (option, LOI, off-market purchase)
Review zoning and confirm data center as permitted or conditionally permitted use
Complete Phase I ESA
Confirm fiber route availability (at least two diverse paths)
Assess cooling approach and water availability
Review FEMA flood map and natural hazard exposure
Confirm construction labor market and trades availability
Model incentive stack against baseline development economics
Secondary market sites that clear these ten gates are increasingly competitive with primary market alternatives — particularly as Northern Virginia, Phoenix, and Dallas approach power saturation.