Data Center Site Screening: Criteria, Process, and How AI Does It at Scale
The go/no-go variables every developer must clear before committing to full due diligence, and why screening speed has become a competitive edge.
Site screening is the step before due diligence. It is the filter that determines which sites are worth the $500,000 to $1.5 million it costs to complete a full diligence process. Done well, it eliminates 80% of candidates in the first few weeks. Done poorly, it lets disqualified sites consume months of team time and capital.
For data center development specifically, the screening criteria are more demanding than almost any other asset class. A site that looks perfect on a map can fail on power, water, fiber, zoning, or half a dozen other variables that are not visible without deliberate analysis. AI has changed what is possible in this phase, not by lowering the bar but by evaluating more sites against a higher bar in less time.
The Core Screening Variables
A complete data center site screen covers five primary domains and 30 or more sub-criteria. Development teams that compress this list take on risk they often do not see until late-stage diligence.
Power
Power is screened first, because it is the highest-frequency disqualifier.
Go/no-go thresholds:
Minimum substation capacity: 50 MW committed or immediately available for a viable development site. For hyperscale campus development, the threshold rises to 200 MW or more.
Substation proximity: ideally within two miles, with no major transmission constraints between the substation and the site
Utility service territory clarity: sites on the boundary of two utility territories create interconnection complexity that can add years to the timeline
Known transmission congestion: check FERC-filed Transmission Planning reports for congestion designations in the relevant load zone
Sites that cannot show a credible path to 50 MW of committed capacity within 24 months are generally screened out at this stage.
Land
Go/no-go thresholds:
Minimum acreage: 20 acres for a single-building campus, 50 or more for a multi-phase development
Topography: slope under 3% preferred; greater than 8% typically requires cost-prohibitive earthwork
FEMA flood zone: sites in 100-year (AE/AO) flood zones are screened out unless mitigation measures are already in place. Flood insurance costs and lender requirements make these sites non-competitive.
Geotechnical risk flag: karst regions (Florida, Tennessee, parts of Indiana and Missouri) and high-liquefaction-risk zones (coastal California, Pacific Northwest) require additional scrutiny
Shape and usability: irregular parcels, parcels with significant easement encumbrances, and sites with protected wetland inclusions lose effective area quickly
Zoning
Zoning determines whether a data center is a permitted use, a conditional use, or a non-starter.
What to check:
Industrial or heavy commercial zoning designation (IM, M-1, M-2 designations in most municipalities)
Special use permit requirements: since 2022, dozens of jurisdictions have added data-center-specific SUP requirements in response to community concerns about power draw, noise, and traffic
Noise ordinances: cooling towers and diesel generators are the primary sources. Many jurisdictions limit continuous noise to 55 to 65 dB at property lines, which constrains generator placement significantly.
Height limits that affect cooling tower design
Recent zoning amendments or active rezoning proceedings on adjacent parcels that could affect future expansion
Fiber and Connectivity
Go/no-go thresholds:
Lit fiber within one mile of the site boundary
Minimum two distinct fiber providers (carrier diversity is a tenant requirement, not a preference)
Access to at least one major carrier hotel within the market (30 to 50 milliseconds latency to key IXPs)
Markets with limited fiber diversity are viable only for specific tenant categories (cloud storage, content delivery) where latency sensitivity is lower.
Water Availability
Water has moved from a secondary screen to a primary one in the last two years.
What to evaluate:
Municipal water system capacity and the utility's willingness to issue a capacity commitment
Prior appropriation states: in Colorado, Arizona, Nevada, and parts of Texas, water rights are senior claims. A site without appurtenant water rights or a credible path to them is a high-risk screen.
Cooling tower blowdown regulations: some municipalities restrict discharge volume and chemistry, which affects operational feasibility
Alternative cooling viability: if water is constrained, can the site support dry or closed-loop cooling at acceptable PUE levels?
Secondary Variables
These factors do not automatically disqualify a site but influence risk scoring and priority.
Labor market proximity: within 30 miles of a skilled trades labor market reduces construction cost and schedule risk
Seismic risk: Zone 3 and 4 sites (California, Pacific Northwest, New Madrid zone) require structural design upgrades that add 5 to 10% to hard costs
Hazmat proximity: active industrial facilities, pipelines, or fuel storage within a quarter-mile create environmental liability risk and tenant concern
Permitting jurisdiction track record: some counties have a history of slow or adversarial permitting on data center projects. This is qualitative but consequential.
Expansion land control: a site without adjacent land for future phases has limited long-term value for campus-scale development
How AI Changes the Screening Process
The traditional site screening workflow is analyst-intensive. Each variable requires pulling from a different data source: utility maps, FEMA flood data, county GIS, FCC fiber data, state environmental databases. A thorough manual screen of a single site takes two to three days. Screening 30 sites takes months.
AI-assisted screening changes this in two ways.
First, it integrates disparate data sources into a unified scoring model. Power availability, zoning classification, flood risk, fiber proximity, and water access can all be evaluated in parallel against a defined criteria set, rather than sequentially.
Second, it enables volume. A development team using AI-assisted screening can evaluate 100 or more sites in the time it previously took to evaluate 10, with consistent criteria application across the entire set. That matters in competitive markets where the best sites are optioned quickly.
What still requires human judgment:
Utility relationship intelligence: queue dynamics that are not reflected in public data
Community risk assessment: the likelihood of organized opposition is not captured in GIS data
Land control strategy: which sites to option and in what sequence involves competitive intelligence that AI cannot fully model
Final go/no-go: the screening output informs the decision; it does not make it
The Developer Takeaway
Site screening is not a cost center. It is the decision architecture that determines which opportunities get capital and which do not. A compressed, AI-assisted screening process does not just save time. It improves selection quality by ensuring that the full criteria set is applied consistently to every candidate, not just the ones a human analyst had time to fully evaluate.
The developers building the strongest data center pipelines in 2026 are running AI-assisted screens against larger site universes than their competitors. The advantage compounds: more sites screened means more viable candidates identified, which gives development teams negotiating leverage and optionality when the best sites come to market.