Hex-grid scoring for datacenter siting across utility service territories. The map covers Alliant Iowa, Indiana & Michigan Power, and SWECI — ~3,870 H3 res-6 hexes (each ~36 km² / ~14 mi²). Each hex is scored 0–100 across power, land, environmental, access, and economic criteria; high-scoring contiguous hexes are clustered into candidate regions for triage.
Each utility territory is tinted in its own hue with a label badge. Notable
clusters — contiguous groups of hexes that pass all the scoring filters
— render as green
polygons; singletons (1-hex clusters) get a dashed border.
Hover any cluster for rank, mean score, hex count, primary utility,
and the top contributing signals. Click to fly to the cluster bounds.
Use the View by dropdown to switch between scoring modes.
Consensus — hex must clear the 90th percentile in both
Weighted and FV Similarity. The most defensible mode — agreement across
approaches reduces false positives.
Weighted Score — the explicit hex-scoring formula
(see Weights tab).
FV Similarity — cosine similarity against a 327-site rural
DC reference embedding across infrastructure-distance features. Data-driven,
no explicit weights.
DC Valuation — kNN-based modeled dollar estimate of
datacenter-ready infrastructure value at each hex (trained on disclosed
comp transactions).
Two pre-filters run before clustering, so we don't surface candidates we
couldn't actually act on:
• Utility filter — hexes outside our tracked utility
territories (primary_utility = null) are dropped.
• Urban cap — hexes with ≥50% Built Area
(Impact Observatory LULC class 7) are dropped — hyperscale DCs aren't
going on top of city centers.
Toggle context layers on/off. Substations and transmission lines on by default.
Read-only breakdown of the criterion weights baked into the
weighted_score formula. Tunable presets are tracked on the
roadmap.
Connected high-scoring hexes after the utility + urban filters. Green polygons; dashed border = singleton (1 hex).
Service-area polygons for each tracked utility. Tinted in distinct hues with a label badge.
Source: HIFLD Electric Retail Service Territories. IMPA assembled from member-muni HIFLD entries; SWECI uses an 11-county fallback when HIFLD entry is stale.
Substations are toggleable by voltage tier in the Layers panel: 765 kV (ultra-backbone, ~22 sites — AEP eastern corridor + ComEd Wilton/Plano/Collins + Cook nuclear), 345 kV+ (bulk transmission, the standard hyperscale interconnect tier), 138 kV+ (regional distribution, adequate for sub-gigawatt and small-site campuses), and Other (sub-138 kV; off by default — usually local distribution and visually noisy at regional zoom).
Source: HIFLD Electric Substations (OSM-derived bulk).
High-voltage lines (100kV+), color-coded by voltage class.
Source: HIFLD Electric Power Transmission Lines.
Interstate natural-gas transmission lines (purple dashed).
Source: HIFLD Natural Gas Pipelines.
Planned generation projects: solar (yellow), wind (cyan), battery (purple), gas (orange). Diamond size = MW capacity. Currently visualization only — not yet a scoring input (tracked as "MW-available-by-year" on the roadmap).
Source: MISO Generator Interconnection Queue.
Operating, planned, and retired units across the region. Color by status.
Source: EIA Form 860 (2024 vintage).
Existing datacenter facilities, rendered as cyan squares (visually distinct from substation circles + GI-queue diamonds). The popup notes "Rural siting" where applicable — those entries are the curated comp set for FV scoring.
Source: datacentermap.com.
NHD lakes & rivers (cooling-water context).
Source: USGS National Hydrography Dataset.
IRS-designated OZ census tracts — meaningful federal capital-gains tax deferral for investments inside the boundary.
Source: CDFI Fund / IRS.
Hex grid — Uber H3 resolution 6 hexes (~36 km² / ~14 mi² each)
covering the union of utility service territories. Land cover comes from
Impact Observatory LULC v02
on Microsoft Planetary Computer; slope from the
USGS 3DEP National Elevation Dataset;
county hazard scores from FEMA National Risk Index;
place / population centroids from TIGER/Line Places.
Clustering — for each scoring mode, hexes above the
90th percentile (after the utility + urban filters) are
grouped into connected components on the H3 adjacency graph. Each
component is one cluster; singletons (1-hex) are kept and flagged.
Consensus mode requires a hex to clear the 90th percentile in
both Weighted and FV simultaneously.
FV Similarity compares each hex's infrastructure profile
to a curated 327-site rural DC reference embedding (cosine similarity).
Useful as a foil to the explicit weights — if Weighted and FV both
rank a hex highly, that's a stronger signal than either alone (hence
Consensus).
Navier Region Triage — May 2026