Methodology — US Data Center Readiness

The Data Center Readiness Index (DCRI) v1 ranks all 50 US states plus the District of Columbia on their suitability for large-scale data center deployment. It aggregates publicly available federal and third-party data across eight dimensions to give hyperscaler site-selection teams, infrastructure investors, policy analysts, and journalists a defensible, reproducible basis for comparing states.

DCRI 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 v1 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 DCRI 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.

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) are inverted so that 100 always means best in class.

The rank pipeline (headline) converts each winsorized indicator to a 0–100 percentile rank, combines indicators within each dimension by weighted arithmetic mean, and combines dimensions by weighted arithmetic mean to produce the 1–51 composite ordering. This is the published ranking.

The score pipeline (continuous) normalizes each winsorized indicator to a 0–100 min-max scale and combines via weighted geometric mean within and across dimensions. The geometric mean is non-compensatory by design: a state cannot trade catastrophic weakness on any dimension off against strength on another. Hyperscalers require minimum acceptable thresholds across all criteria simultaneously, not merely a high average — the score pipeline reflects that.

The two pipelines are a robustness device, not a redundancy. If both produce highly correlated state orderings, each serves as a methodological witness to the other. DCRI v1 ships with a published Spearman ρ of 0.7772, recorded in every run’s metadata. That value is below the 0.85 OECD convention; we publish it openly under a documented rationale: both pipelines agree on the top-five and bottom-five clusters across each step of the v1 build, geometric-mean aggregation is non-compensatory, and the divergence is concentrated in the mid-table where weighted composites are expected to be sensitive.

Exact numerical weights are internal to Labrynth methodology and not published. Energy Supply is the most heavily weighted dimension; other weights reflect operator priorities from Phase 1 research.

States are assigned to one of four tiers, recalculated on each run: 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 tier applies to jurisdictions that fall over the two-dimension missing-data threshold. The District of Columbia is in this tier for v1 and is not assigned a 1–51 rank.

DC has null dimension scores on Water Supply because the U.S. Drought Monitor — the single indicator that v1 Water Supply is built on — does not produce classifications for the District (the Monitor’s smallest reporting unit is the state). DC also has null scores on two other dimensions where federal source data is structured around states and DC’s small geographic footprint, lack of in-state generation, and non-state administrative structure produce non-comparable inputs. Together those three null dimensions push DC over the two-dimension threshold for ranking.

Before the v1 redesign, DC was scored in Water Supply on a flood indicator alone. That flood indicator was removed in v1 because it double-counted FEMA’s National Risk Index composite. Publishing DC as “insufficient data” is the methodology-coherent outcome of that decision — better than backfilling from a structurally inferior source.

Every set of published scores is linked to a run_id that records the weights file used, the git commit of the scoring notebooks, the source data vintage, and the timestamp of the run. Rerunning with the same inputs and weights 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 Quartile states are expected to remain stable under these perturbations. Sensitivity tables are internal methodology documentation; the dual-pipeline Spearman ρ in run metadata is the published audit signal.

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.