Data center site selection criteria: how to find powered land near a substation
Ask anyone who sites data centers what matters most and the answer is the same: power. Fiber, water, flood risk, zoning and price all matter, but a parcel with everything except a path to tens or hundreds of megawatts is not a data center site. It is a field.
This guide is about that first, hardest filter: how to find powered land, what the public grid data can and cannot tell you, and how to join grid layers to parcels so the screen ends in a list of real properties. For the general site selection method that works for any facility type, start with AI site selection with GIS. This one goes deeper on power.
Why power comes first in data center site selection
Most site selection treats criteria as roughly equal. Data centers do not. The electrical load is so large that it decides whether a site is possible at all, and how long it will take to energise.
Some reference points from neutral sources:
- Data centers used about 4.4% of total US electricity in 2023 (176 TWh), and are expected to use roughly 6.7 to 12% by 2028, according to the Lawrence Berkeley National Laboratory report released by the US Department of Energy.
- EPRI's 2024 Powering Intelligence study noted that it is "not unusual to see new centers being built with capacities from 100 to 1,000 megawatts", roughly the load of 80,000 to 800,000 homes, and that connection lead times of one to two years create local supply challenges (summarised by the American Public Power Association).
- In Texas, ERCOT's large load reporting defines a large load as one or more facilities at a single site with an aggregate peak demand of 75 MW or more (ERCOT NPRR 1267 board report).
Loads of that size are rarely served from an ordinary distribution feeder. A utility presentation to WECC, from NV Energy, put it plainly: serving large loads "depends heavily on high-voltage transmission (100 kV and above)" (WECC, large load considerations). NERC's August 2026 proposal for registering computational load sites uses the same voltage, a site of 50 MW or more connected at 100 kV or above (NERC proposal, hosted by Texas RE). For a site screen, that is a practical cue: start from the transmission system, not from any pole line on the street.
Proximity vs available capacity: what a map can and cannot tell you
The most common mistake in a powered land search is to treat distance to a substation as if it were power. It is a necessary condition, not a sufficient one. A substation two hundred metres away may already be fully committed to other customers, and a line passing over a parcel may be at a voltage that cannot serve the load without a new substation.
| Question | Can public GIS data answer it? | Where the answer comes from |
|---|---|---|
| Where are substations and transmission lines? | Yes, approximately | HIFLD, OpenStreetMap, state or utility open data |
| What voltage is the nearest line or substation? | Often | Voltage attributes on the HIFLD or utility layers |
| How much spare capacity does that substation have for new load? | Rarely | A formal request to the utility; a few utilities publish capacity maps |
| When could a new load be energised? | No | The utility's large load interconnection process and its study results |
Some utilities publish hosting capacity maps. The DOE describes them as an overview of a distribution system's ability to host additional capacity, "either generation or load", at specific locations, while noting that they "do not address site-specific interconnection questions" (DOE U.S. Atlas of Hosting Capacity Maps). Most of these maps cover distribution feeders, so for a load in the tens or hundreds of megawatts they are a hint at best.
Power supply is also tied to new generation. Berkeley Lab's Queued Up report counted about 1,312 GW of generation actively seeking grid interconnection at the end of 2025, and found the median time from interconnection request to commercial operation was over five years for projects built in 2025 (Berkeley Lab, Queued Up 2026 edition). That report covers generation only, not load requests, but it explains why utilities are cautious about promising new capacity quickly.
The layers in a powered land screen
A full data center screen usually combines eight to ten layers. Here is what each one does and where it typically comes from:
| Layer | What it tells you | Typical public source |
|---|---|---|
| Transmission lines | Where high-voltage lines run, and at what voltage | HIFLD, which describes its lines as operating "from 69 kV up to 765 kV" |
| Substations | Points of possible interconnection, often with a maximum voltage | HIFLD, OpenStreetMap power=substation, utility open data |
| Parcels | Ownership, acreage, land use, assessed value | County appraisal district or city open data portal |
| Zoning | Whether a data center is allowed, by right or with approval | City or county zoning layer |
| Flood hazard | Special Flood Hazard Areas (the 1% annual chance flood) | FEMA National Flood Hazard Layer |
| Wetlands | Areas likely to need permitting or avoidance | US Fish and Wildlife Service National Wetlands Inventory |
| Fiber | Proximity to long-haul and metro fiber routes | City or state open data where published; often a private layer |
| Water and sewer | Supply for cooling and discharge | Utility or city open data, often incomplete |
| Roads and rail | Construction and equipment access | OpenStreetMap |
| Incentive areas | Tax exemptions, enterprise zones, Opportunity Zones | State programs and federal designations |
A few notes on the sources:
- HIFLD (Homeland Infrastructure Foundation-Level Data) is the usual national starting point for grid infrastructure. Its transmission line layer describes lines "operated at relatively high voltages varying from 69 kV up to 765 kV" (Data Rescue Project archive of HIFLD Open transmission lines), and its substation layer carries a maximum voltage field defined as the "maximum voltage of all transmission lines that feed into the substation" (HIFLD substation metadata, Florida Geographic Data Library). Treat both as approximate; utilities hold the authoritative versions.
- FEMA's NFHL is "a geospatial database that contains current effective flood hazard data" and covers over 90% of the US population (FEMA).
- The National Wetlands Inventory uses a biological definition of wetlands and makes no attempt to define regulatory jurisdiction; the US Fish and Wildlife Service points regulatory questions to the Army Corps of Engineers (USFWS NWI FAQ). It is a screening layer, not a delineation.
- Incentives vary by state. NCSL reports that the 38 states offering data center tax incentives all waive sales tax on essential equipment, and 14 extend the exemption to electricity used at the facility (NCSL, April 2026). Incentives are usually a statewide rule rather than a map layer, so they shape which state you search more than which parcel you pick.
The core problem: grid data and parcel data live in separate systems
Here is the step that takes the time. Grid infrastructure is published by federal programs, utilities and OpenStreetMap volunteers. Parcels are published by county appraisal districts and city GIS departments. They use different schemas, different projections and different update cycles, and nobody has joined them for you.
Answering "which parcels are within 1.5 km of a substation" means:
- 1Getting the parcel layer for the right jurisdiction, and checking that it actually covers the study area.
- 2Getting substations and transmission lines for the same area, from a source you trust.
- 3Reprojecting both to a projected coordinate system, so distances are in metres rather than degrees.
- 4Buffering the grid features by the distance you care about.
- 5Intersecting the buffers with the parcels, and attaching the nearest substation (with its voltage) to each parcel.
- 6Only then filtering by acreage, flood, wetlands and zoning.
Every one of those steps is routine GIS. The difficulty is that they cross two data worlds, which is why a powered land search so often stalls at step one. Once the join is done, the rest of the screen is fast.
A step-by-step data center site screen
Run the criteria in this order. Each step removes land, and the cheapest, most decisive tests go first.
- 1Choose the region. State incentives, utility territory and market (latency to users, existing campuses) narrow the search to a metro or a few counties.
- 2Power proximity. Keep parcels within a stated distance of a substation or transmission line, ideally filtered by voltage (for example, substations of 138 kV or higher, if the voltage field is populated).
- 3Size. Keep parcels above your minimum acreage, or groups of adjacent parcels under one owner that add up to it.
- 4Hard exclusions. Remove parcels in the FEMA 1% annual chance flood zone, and parcels largely covered by mapped wetlands.
- 5Zoning. Keep industrial or compatible districts, or flag parcels that would need a rezoning.
- 6Soft ranking. Score what is left on distance to fiber, distance to a highway, land value and distance to the substation itself, using weights you state.
- 7Utility conversation. Take the top candidates to the utility for a capacity and timeline answer before anyone spends money on due diligence.
Step 6 is a weighted overlay. If you have not built one before, our guide to site suitability analysis and weighted overlay explains how to choose the scale and the weights, and why hard constraints should never be mixed into the score.
What this looks like on real data
Two real NeuraCities runs show both the method and its limits.
Houston. Asked where a data center could go on land close to a power substation and a fiber line, NeuraCities pulled parcels from the City of Houston open data hub, 402 fiber optic cable features from the Houston GeoHub and 322 substations from OpenStreetMap. It buffered substations by 1,500 m and fiber by 500 m, intersected both with the parcels and attached the nearest substation. 34 parcels met both tests.
Show the data
| Under 125 m | 11 |
| 125 to 250 m | 6 |
| 250 to 375 m | 6 |
| 375 to 500 m | 11 |
Dallas. In a data center shortlist by Thursday, the brief was big parcels, industrial or commercial zoning and close to power. NeuraCities used Dallas's tax parcels and base zoning, and for power it used the EPA ECHO register filtered to Toxics Release Inventory facilities, where generating stations and other large industrial sites are recorded. Keeping parcels over five acres in industrial or commercial zoning within 5 km of a facility left four sites.
The Dallas run is a useful lesson in itself. A power plant is a proxy for power, and a generous one: it says nothing about substation location, voltage or capacity. The post says as much, and the natural next step is to add the utility's substations and fiber routes and rerun the last step. The four either hold up or get replaced. A screen is only as good as its power layer.
How to run a powered land screen in NeuraCities
Describe the screen the way you would brief a colleague, with the distances and thresholds written out:
Find parcels in Harris County, TX of at least 20 acres within 1,500 m of an electrical substation, outside the FEMA 100-year floodplain and not mostly covered by NWI wetlands. Attach the nearest substation and its voltage to each parcel. Rank what is left by distance to fiber, distance to a highway and zoning compatibility, and show the top 20 in a table.
What happens next:
- The orchestrator writes a checklist of the layers and steps it will use, so you can see its reading of the question (for example, which flood zone it treats as the 100-year floodplain) before anything runs.
- The Data Agent looks for parcels, zoning and fiber on the county and city portals in its registry of 600+ open data portals, and for substations, transmission lines and flood hazard in federal sources such as HIFLD and FEMA, national catalogs such as data.gov, and OpenStreetMap. It probes each candidate for coverage before downloading it.
- The GIS agent reprojects the layers, runs the buffers, intersections, nearest-feature join, area filter and overlay, then normalises the soft criteria and combines them in a weighted overlay using the weights you gave.
- You get a map with every layer and its source, a ranked table you can export to Excel, charts and a written summary. Download the candidates as a Shapefile or CSV for your GIS team.
If a layer does not exist publicly (fiber and substation capacity are the usual gaps), the answer says so instead of quietly dropping the criterion. Upload the utility's own substation file or your fiber provider's routes, or connect a PostgreSQL database, and your data takes priority over the public layer.
Then keep asking: "Only substations of 138 kV or more." "Drop parcels under one owner with less than 50 contiguous acres." "Which of these are in an Opportunity Zone?" Each follow-up reuses the layers already loaded. When the screen works, save it as a workflow with {{county}} and {{min_acres}} variables and run the same criteria in the next market.
Tips for a defensible powered land screen
- Write the power test with a voltage, not just a distance. "Near a substation" is ambiguous; "within 1.5 km of a substation of 138 kV or more" is a criterion. Check that the voltage field is populated first, because many records are not.
- Know your power layer's lineage. OpenStreetMap, HIFLD and utility layers disagree in places. Open the layer's source in the map and note which one the screen used.
- Do not confuse proximity with capacity. Label the output as "close to grid infrastructure", and keep the utility's capacity answer as a separate column you fill in later.
- Screen flood and wetlands as exclusions, not scores. A site in the floodway should not survive because it is close to fiber.
- Think in assemblages. Large campuses are often several parcels under one owner. Dissolve by owner before the acreage filter if that is how land is bought in your market.
- Rerun when the inputs change. Substations get built and zoning gets amended. A saved workflow makes the refresh one click.
Frequently asked questions
What are the most important data center site selection criteria?
Power comes first: access to high-voltage transmission and a substation with capacity for a large new load. After that, typical criteria are parcel size, flood and wetland exclusions, zoning, fiber proximity, water supply, road access, land cost and state incentives.
How do I find powered land?
Start with GIS: buffer substations and transmission lines (from HIFLD, OpenStreetMap or utility data) by a stated distance, intersect them with county parcels, and filter by acreage, flood zone and zoning. That gives you land that is close to grid infrastructure. Whether it is actually powered, meaning the utility can serve the load, has to be confirmed with the utility.
How close does a data center need to be to a substation?
There is no universal rule; it depends on the load, the voltage and the cost of building a line. Screens commonly use a fixed distance as a first pass, such as the 1,500 m used in a real NeuraCities Houston run, and then test the shortlist against the utility's actual interconnection options.
Where can I get substation and transmission line data?
Nationally, HIFLD publishes electric substation and transmission line layers, with transmission lines described from 69 kV up to 765 kV. OpenStreetMap maps many substations as power=substation. Some states, cities and utilities publish their own layers, which are usually more accurate.
Can I see substation capacity on a map?
Sometimes, but rarely for large loads. Some utilities publish hosting capacity maps that show the ability to host generation or load on distribution circuits. The DOE notes those maps do not answer site-specific interconnection questions, so large loads still require a request to the utility.
Can AI do data center site selection?
It can run the screen. In NeuraCities you describe the criteria in plain English; the agents find parcels, grid infrastructure, flood, wetland and zoning layers, run the buffers, overlays and joins with standard GIS code, and return a ranked, sourced shortlist. Capacity and interconnection still come from the utility.
Describe your power, size and flood criteria and get a sourced list of parcels near the grid.
Screen for powered land