Methodology

The Data Center Readiness Index ranks all 50 US states on their suitability for large-scale data center deployment. It aggregates publicly available federal and third-party data across nine dimensions to give hyperscaler and large-colocation site-selection teams, infrastructure investors and lenders underwriting multi-billion-dollar campus builds, and policy analysts and journalists a defensible, reproducible basis for comparing states.

The dimensions, indicators, and weights are tuned to multi-hundred-megawatt campus-scale projects, but the Index is directionally useful for any party evaluating where to place a large industrial load.

The Data Center Readiness Index ranks states on dimensional readiness, not on existing market presence or engineered-solution offsets. A state’s rank reflects raw operating-environment quality, not its current campus footprint. Three of the most active US data-center markets (California, New Jersey, and Nevada) rank in the lower half of the composite. They are not low-ranked because the Index is wrong about them. They are low-ranked because the underlying operating environment in each is materially harder than in the high-ranked states, and operators who build there do so by buying their way around those constraints: long-term water contracts, custom interconnection deals, dry-cooled or closed-loop facility designs, premium real estate, premium labor.

The most useful reading of the Data Center Readiness Index is as a complement to existing-market data: a high-ranked state is one where a new build faces less friction; an existing major market that ranks low is a place where capital has accumulated despite friction, typically through engineered solutions and incumbency effects that a new entrant would not start with.

Every state receives two composite scores per release. The published 1–50 ranking is produced by a non-compensatory geometric-mean composite; a supplementary arithmetic-mean composite is computed from the same inputs as a robustness check on the headline. Both follow standard practice in the OECD Handbook on Constructing Composite Indicators.

Each indicator is winsorized at the 5th and 95th percentiles across the 51 jurisdictions before any normalization. This caps extreme values (a single hurricane year, for example, would otherwise dominate the outage-duration scale) without re-ranking the underlying data. Indicators where lower is better (electricity price, outage minutes, hazard score, opposition actions) are inverted so that 100 always means best in class.

The headline ranking uses a geometric-mean composite. Each winsorized indicator is normalized to a 0–100 min-max scale; indicators are combined within each dimension by weighted geometric mean, and dimensions are then combined by weighted geometric mean across the composite. The result is sorted to produce the 1–50 ranking.

The geometric mean is non-compensatory by design: a state cannot trade catastrophic weakness on any single dimension off against strength on another. Weakness in any one dimension is penalised multiplicatively rather than averaged out. Hyperscalers require minimum acceptable thresholds across all criteria simultaneously, not merely a high average; the headline ranking is built to reflect that.

A supplementary composite is computed from the same winsorized inputs using percentile-rank normalization and weighted arithmetic mean. The Spearman rank correlation between the geometric-mean headline and the arithmetic-mean supplementary composite is reported with every release as a methodological witness to the structural agreement between the two aggregations. Disagreement between them is concentrated in states with mixed performance across dimensions; uniformly strong or uniformly weak states rank similarly under both methods. The divergence is informational, reflecting different commitments about whether averaging out weakness is acceptable.

Energy Supply is the most heavily weighted dimension; the other weights reflect operator priorities surfaced through expert interviews and review. The exact dimension weights are part of our proprietary methodology and are not published.

States are assigned to one of four tiers, recalculated on each release: Top Quartile covers ranks 1–13 (strongest overall environment for data center deployment); Above Average covers ranks 14–26 (solid fundamentals with one or two notable gaps); Below Average covers ranks 27–38 (meaningful weaknesses that warrant attention); Bottom Quartile covers ranks 39–50 (significant structural barriers).

A separate Insufficient Data designation applies to jurisdictions that fall over the missing-data threshold and are not assigned a 1–50 rank.

Every set of published scores is linked to a release identifier that records the weights used, the source data vintage, and the timestamp of the run. Rerunning with the same inputs produces identical scores. Score history is preserved; no published row is ever overwritten.

We test weight sensitivity by perturbing each dimension weight by ±25 percent and observing tier movement. Top-tier states are expected to remain stable under these perturbations.

OECD/JRC (2008). Handbook on Constructing Composite Indicators: Methodology and User Guide. OECD Publishing, Paris. DOI: 10.1787/9789264043466-en.

Saisana, M., Saltelli, A., & Tarantola, S. (2005). Uncertainty and sensitivity analysis techniques as tools for the quality assessment of composite indicators. Journal of the Royal Statistical Society: Series A, 168(2), 307–323.