Site Scoring and Weighting Methods for Data Center Selection

Every site search ends with a shortlist of parcels that are each good at different things. One has a 230 kV line on its boundary but sits an hour from the metro; another has fiber and labor nearby but needs a new substation. A scoring model makes the comparison explicit. In the planning literature this is GIS-based multi-criteria decision analysis: methods for combining geographic data with the decision maker’s value judgments to support a choice.¹

The model does two jobs. It forces the team to write down what it values and how much, before anyone falls in love with a parcel. And it gives a record of why one site ranked above another, which matters when the decision is revisited months later. It does not prove that a site works. That still depends on utility answers, field studies and entitlement, as our guide to desktop screening vs. field assessment explains.

The criteria themselves are covered in our site selection criteria checklist. This guide is about the method: how to turn those criteria into gates, scores and weights. For data centers, one criterion carries outsized weight today. A 2025 survey of hyperscalers, colocation developers and utilities, published by an on-site power vendor, found that power access had overtaken fiber proximity as the top factor, with 84% of respondents ranking it among their top three.⁴

There are two basic ways to combine criteria. A Boolean or conjunctive screen asks whether a site meets every threshold and rejects it if it misses any one. A weighted linear combination multiplies each criterion score by a weight and adds them up, so a high score on one criterion can offset a low score on another.² Eastman describes these as points on a continuum: the Boolean AND allows no trade-off between criteria, and the weighted average allows full trade-off.² Malczewski’s ordered weighted averaging method makes the same point formally, spanning minimum-type and maximum-type aggregation with the ordinary weighted combination in between.⁹

For data center land, full trade-off is dangerous. No amount of fiber, labor or tax incentive compensates for a site that cannot get power on the target date, sits in a regulatory floodway, or lacks the contiguous acreage the campus needs. Those criteria belong in gates. Recent data center siting research follows that order: a 2026 national suitability study for the United Kingdom first excluded constrained areas such as protected land and flood zones, then weighted climate and infrastructure factors with AHP and combined them by weighted linear combination on a 1 km grid.¹⁰

Write gate thresholds as testable statements, such as “utility indicates a feasible path to 100 MW by 2029” or “at least 150 buildable acres outside the mapped floodplain.” A gate that fails on a desktop basis but could pass after field work should be recorded as unresolved, not as a pass. Our guide to why data center sites fail lists the issues that most often belong in gates.

Each scored criterion needs three things: a definition that does not overlap with the others, a measurement, and a rule that converts the measurement to a common scale. Overlap is the most common error. If “distance to substation,” “distance to transmission” and “utility capacity” are all scored separately, power is counted three times and the weights no longer mean what they say. The UK government’s multi-criteria manual, written for officials appraising options, treats the structure of criteria as a core step before any weighting.¹¹

Because criteria arrive in different units (miles, megawatts, months, dollars), they have to be standardized before they can be combined; a linear rescaling between the worst and best values in the set is the most common approach in GIS work.⁹ Make the direction explicit. Closer fiber is better; a longer power timeline is worse.

Score power from the best evidence available and say what that is. A distance to a line is a weak proxy for capacity; our guide to substation capacity and proximity explains why. Where scores come from mapped data, record the layer and its date, as covered in our guide to GIS data sources.

Weights express how much a move from worst to best on one criterion is worth relative to the same move on another. Several methods are in common use.

• Direct rating or points allocation. The team spreads 100 points across criteria. Fast and transparent, but prone to anchoring on round numbers.
• Rank-based weights. Criteria are ranked and a formula converts ranks to weights. Useful when the team agrees on order but not magnitude.
• Analytic hierarchy process (AHP). Experts compare criteria two at a time on a scale of absolute judgments, and priorities are derived from the comparison matrix.⁵ A consistency ratio of 0.10 or less is the conventional test that the judgments hang together, applied as a guideline rather than a strict rule.⁶
• Swing weighting. Each weight reflects how much the team values the swing from the worst to the best option actually on the shortlist for that criterion.³

Swing weighting fixes a subtle problem with abstract importance weights. If every shortlisted site has excellent fiber, fiber may matter a great deal in general but it cannot separate these sites, so its weight in this comparison should be small. The UK guidance on options appraisal pairs swing weighting with sensitivity analysis to show how preferences shift when the relative importance of criteria, or the gap between the best and worst option on a criterion, changes.³

Published data center siting studies use a range of methods. One ranked 50 economies on eight main criteria and 45 sub-criteria, deriving weights with a fuzzy form of AHP and ranking locations with fuzzy TOPSIS.¹² Another, by Covas, Silva and Dias, used the ELECTRE TRI outranking method across technical, social, economic and environmental dimensions and handled uncertainty about weights rather than fixing them.¹³ The method matters less than writing the judgments down and testing them.

Suppose four parcels survive the gates. The team scores each 0–10 on five criteria and agrees weights of 35% power timing, 20% transmission proximity, 15% fiber, 15% water and 15% entitlement. The weighted total is simply the sum of score times weight.

Site C “wins,” but only by a tenth of a point. Site A is the strongest power site, and Site B is the most even across criteria. A report that names C as the first choice without saying how fragile that is overstates what the model knows. The next step is to find out how the order changes when the weights move.

Sensitivity analysis asks how much the result depends on the judgments. The simplest form changes one weight at a time, re-normalizes the others, and records when the top site changes. Chen and colleagues built this into a GIS tool that varied the weights of all criteria in an AHP suitability model to see their relative effect on the final map.⁷ Ligmann-Zielinska and Jankowski went further, treating weights as uncertain and analyzing uncertainty and sensitivity together in a spatially explicit way, an area they note had received less attention in land suitability work.⁸

Rankings can also change for reasons unrelated to the weights. Belton and Gear showed in 1983 that adding an alternative to an AHP comparison could reverse the order of the original alternatives, and later work found that adding or removing other alternatives can do the same.¹⁴ The practical lesson is to fix the shortlist before final scoring and to rerun the model if a site is added or dropped.

• Report the margin between first and second place, not only the order.
• Name the weight change that would flip the top two sites.
• Show results under at least one alternative weighting, such as a “power first” view and a “balanced” view.
• Treat sites whose ranges overlap as tied and decide between them on the open diligence items.

1. 01Set the gates and their thresholds first, in writing, tied to the project’s load, ramp and schedule.
2. 02Screen every candidate against the gates and record which gates are confirmed and which are unresolved.
3. 03Define non-overlapping criteria, measures and scoring scales for the survivors.
4. 04Agree weights in a session with the people who own the decision, using swing weights or AHP, and record the reasoning.³⁵
5. 05Score, then run sensitivity analysis and report the margin of victory.⁷
6. 06Feed the result into diligence: the open questions on the top two or three sites decide more than the decimal points.

Scoring works best as one part of a broader portfolio screen and should be read with the cautions in our guide to reading a feasibility report. Our methodology explains how BlackForge separates gates from graded findings. If you have parcels to compare, you can get a site reviewed.