Planetary-scale answers, unlocked.
A Hands-On Guide for Working with Large-Scale Spatial Data. Learn more.
Authors
Data center site selection is the process of finding land where a facility of 50 to 1,000 megawatts can be powered, connected, cooled, permitted, and kept out of harm’s way. It screens candidate parcels against grid capacity, fiber routes, water availability, natural hazards, zoning, and latency to users, then ranks the survivors on cost and time to build.
Data center site selection is a spatial screening problem with power at the center. A hyperscale campus draws as much electricity as a mid-sized city, and the grid connection sets the schedule.
The defining property is that the criteria are mostly geographic and mostly in conflict. Cheap land is far from fiber; abundant water sits in flood plains; available substation capacity sits where latency to users is poor.
Site selection runs in layers. A first screen removes land that fails a hard constraint: inside a flood zone, on a fault, in a protected area, or more than a set distance from transmission.
A second pass scores the remaining parcels on soft criteria, such as distance to fiber, climate for cooling, tax incentives, and community acceptance.
Developers, hyperscalers, colocation operators, and their advisors run this process for every new campus, and the winners are decided on data before anyone visits the land.
Two site selection layers are ready to query in the Havasu catalog: Copernicus DEM GLO-30 for elevation and slope, and Overture infrastructure for towers, bridges, and power lines.
Data center site selection turns a region into a ranked shortlist of parcels, each with a power, fiber, water, hazard, and permitting profile attached.
Hyperscalers evaluate hundreds of candidate sites a year across multiple markets, and only a few advance to land option and interconnection application.
Utilities use the same analysis in reverse, mapping where they can serve a large new load and marketing those sites to developers.
Economic development agencies pre-qualify sites by assembling the power, fiber, and environmental data a developer would otherwise spend months collecting.
Investors screen data center portfolios for hazard exposure and stranded-asset risk.
Each of these decisions rests on the same joins between parcels and the layers that describe what surrounds them.
A data center site selection model joins parcels to a stack of layers, applies hard filters, and scores what remains.
Each layer is a spatial join. Distance to transmission is a nearest-neighbor query, flood exposure is a point-in-polygon or raster sample, and slope is zonal statistics on an elevation model. Latency is a drive-time or fiber-distance calculation.
Weights turn the joined columns into a score. A power-constrained market weights substation capacity highest, and a latency-sensitive workload weights population proximity. Sensitivity tests show which weights change the ranking.
The output is a ranked table of parcels with every criterion as a column, plus maps for the utility meeting that names a specific substation and interconnection point.
Refresh matters. Interconnection queues, incentive zones, and zoning change monthly, so the model is rerun on a schedule.
WherobotsDB runs the whole site selection stack, parcels against power, hazard, terrain, and context layers, as SQL queries over data that stays in the customer’s cloud storage.
Parcels load from GeoParquet, File Geodatabase, or an ArcGIS Feature Service. Transmission lines and substations from HIFLD or a utility join with a nearest-neighbor query that returns distance to the closest line of each voltage class.
Hazard screens are one join each. FEMA flood zones come in through the ArcGIS Feature Service reader and intersect parcels with ST_Intersects; seismic and wildfire hazard rasters sample per parcel with RS_ZonalStats.
Terrain comes from the Copernicus GLO-30 DEM available in Wherobots Cloud, with mean elevation and relief per parcel from RS_ZonalStats.
Context comes from the Wherobots Data Hub. Overture Maps supplies roads, buildings, and places for proximity scoring, and ESA WorldCover supplies land cover to flag wetlands and forest. Paid editions add Overture places with precomputed driving isochrones for latency-to-population proxies.
Wherobots documents the pattern in its Site Selection use case, and a scheduled Job Run reruns the model each month as queues and zoning change.
The Wherobots MCP Server exposes the same tables to Claude Code and other agents, so a developer can ask for parcels over 100 acres within two miles of a 230 kV line and outside the 500-year flood zone, and get the list.
-- First-pass screen: large parcels near high-voltage transmission, outside flood zones, low relief SELECT p.apn, p.acres, ST_Distance(ST_Transform(p.geometry,'EPSG:4326','EPSG:5070'), ST_Transform(t.geometry,'EPSG:4326','EPSG:5070')) AS m_to_230kv, RS_ZonalStats(d.rast, p.geometry, 1, 'max') - RS_ZonalStats(d.rast, p.geometry, 1, 'min') AS relief_m FROM parcels p JOIN transmission_lines t ON t.voltage_kv >= 230 AND ST_DWithin(p.geometry, t.geometry, 3218.0, true) LEFT ANTI JOIN fema_nfhl_zones z ON ST_Intersects(z.geometry, p.geometry) JOIN wherobots.copernicus_dem.glo_30m d ON RS_Intersects(d.rast, p.geometry) WHERE p.acres >= 100;
Screen parcels against power, hazard, and terrain layers in WherobotsDB on the free tier at cloud.wherobots.com.
Data center sites are chosen in two passes over candidate parcels. A first screen removes land that fails a hard constraint: inside a flood zone, on a fault, in a protected area, or too far from transmission. A second pass scores the remaining parcels on power capacity, fiber, water, climate, latency, and incentives. Due diligence and utility discussions then narrow the ranked shortlist to a site.
The main data center site selection criteria are power availability and cost, fiber connectivity and route diversity, water for cooling, natural hazard exposure (flood, seismic, wildfire, severe wind), land size, zoning and slope, climate for free cooling, latency to users, and tax incentives. Power access is usually the deciding constraint.
Power dominates data center site selection because a hyperscale campus needs hundreds of megawatts of firm capacity, and grid interconnection can take three to seven years in constrained markets. Land, fiber, and buildings can be arranged in months; substation capacity and transmission upgrades cannot, so sites with available power are scarce and expensive.
Data center site selection uses parcel boundaries and zoning, transmission line and substation locations with voltage, interconnection queue records, fiber route maps, FEMA flood zones and other hazard layers, elevation models for slope, water rights and aquifer data, land cover, climate normals, population and enterprise density, and incentive zone boundaries.
A first-pass geospatial screen in data center site selection takes days once the data is assembled, narrowing thousands of parcels to a few dozen. Due diligence on the shortlist, including utility discussions, environmental review, and land negotiation, takes six to eighteen months. Grid interconnection, which starts after site selection, often takes longer than everything else combined.
What Is a Spatial Join? Predicates, Types, and SQL
A spatial join matches rows from two tables by where their geometries sit. How spatial joins work, the predicates and join types, and how to write one in SQL.
What is Sentinel-2? Bands, Resolution, Data Access, Uses
Sentinel-2 is the Copernicus optical satellite mission that images Earth's land and coasts in 13 spectral bands at 10 to 60 m every five days. Its bands, resolution, products, free data access, and uses.
What is Synthetic Aperture Radar (SAR)? Bands, Data, Uses
Synthetic aperture radar (SAR) images the ground with microwaves, day or night, through cloud. How SAR works, its bands, products, and data sources.
share this article
Awesome that you’d like to share our articles. Where would you like to share it to:
