indahl/atlas-data-pack

ATLAS Data Pack

This is the same data as the download on indahl.ai, published here for the Hugging Face viewer and the datasets library. The page is https://indahl.ai/atlas . The files were tested before release, and what was opened and compared is in TESTED.md. The checking script verify.py and the guide page START_HERE.html are in the download from indahl.ai and are not copied here. Licences differ by file. OpenStreetMap-derived files are under the ODbL 1.0 and Epoch AI-derived files are under CC BY 4.0, as the pack LICENSE.txt says. Files that hold both carry both. EIA, EPA and US Census data are public domain. Full terms: LICENSE.txt.

Heuremen LLC · https://indahl.ai/atlas

The data behind the ATLAS pages at https://indahl.ai/atlas, as files you can open in a spreadsheet, a GIS tool or Python. Each dataset comes as the original JSON file unchanged, as CSV with one row per record, and for points on a map as GeoJSON. Each dataset carries its own build or pull date, or data year, listed in SOURCES.md. Epoch AI values are as of the pull date, 2026-10-28, and Epoch has updated some sites since.

What is in each file

Data centers (json/datacenters.json): 1,649 features from OpenStreetMap and 77 US sites from Epoch AI. They come as csv/datacenters_osm.csv, csv/datacenters_epoch.csv, geojson/datacenters_osm.geojson and geojson/datacenters_epoch.geojson. 10 Epoch sites have no coordinates, so their GeoJSON features have a null geometry. The place_method column says how a point was found and place_match holds the matched address text when there is one. Each OpenStreetMap row carries its source and its OpenStreetMap object id. Each row in both files carries the pull date in a pulled column.

Operating power plants (json/plants.json): 14,280 plants from the EIA monthly generator inventory. They come as csv/plants.csv and geojson/plants.geojson. Capacity by fuel is in the columns that start with mw_. A blank means the plant has no capacity in that fuel.

Planned power plants (json/planned.json): 1,608 plants. They come as csv/planned.csv, csv/planned_generators.csv and geojson/planned.geojson. The planned_generators.csv file has one row for each planned generator with its year, megawatts and fuel, taken from the years list of each plant. The same list is kept as text in the years_json column of planned.csv.

Electricity (json/electricity.json): prices, sales and generation for 51 states, including the District of Columbia. They come as csv/electricity_states.csv, one row per state with the data center counts kept in the data_centers_ columns. The national figures are in csv/electricity_us.csv and csv/electricity_us_ownership.csv.

Utilities (json/utilities.json): 1,823 rows of bundled EIA-861 data, one per utility and state, as csv/utilities.csv.

Carbon (json/carbon.json): EPA eGRID emission rates for 51 states as csv/carbon_states.csv, 27 grid regions as csv/carbon_subregions.csv and the national figures as csv/carbon_us.csv.

Cost to scale (json/scale.json): one year of electricity at the stated power of the Epoch AI sites. It comes as csv/scale_sites.csv with 77 sites, csv/scale_states.csv with 25 states and csv/scale_corrections.csv with 1 correction. The method and the totals are in csv/scale_meta.csv.

Every dataset also has a csv/<name>_meta.csv file with two columns, key and value. It holds the fields that describe the whole dataset, such as the source text, the method, the units, the legend bins and the build date.

Sources and dates

SOURCES.md lists the upstream sources, their links, their licences and the dates found in the data. NOTICE.txt carries the attribution lines. The OpenStreetMap rows are shared under the ODbL 1.0 with the line © OpenStreetMap contributors.

Only the data center rows carry a source in every row. The plants, planned plants, electricity, utilities, carbon and scale files carry one source statement for the whole dataset, in the source or sources entry of the JSON and in the _meta.csv file. They have no source column per row. Their rows carry the build date or the data year as a column instead.

How to open the files

Spreadsheet: open any file in csv/. The files are UTF-8 with a byte order mark so Excel reads names correctly. Empty text and missing values are both shown as blank cells.

GIS tool: add any file in geojson/ as a vector layer. Coordinates are in longitude and latitude order in WGS 84. Properties keep their nested structure, and a tool that cannot read nested values will show them as text.

Python, standard library only:

import json, csv
rows = list(csv.DictReader(open("csv/plants.csv", encoding="utf-8-sig")))
plants = json.load(open("json/plants.json"))["plants"]
points = json.load(open("geojson/plants.geojson"))["features"]

Check the pack with python verify.py. It recounts the rows of every file, compares every CSV cell with the JSON and compares the GeoJSON features with the JSON records.

What is here

• json/: the original files from the ATLAS data folder, unchanged
• csv/: one CSV for each table, plus the _meta.csv files
• geojson/: data centers, operating plants and planned plants as FeatureCollections
• SOURCES.md: upstream sources, links, licences and dates
• TESTED.md: what the files were opened with and what was compared
• NOTICE.txt: upstream licences and attribution
• verify.py: recounts and compares the files
• START_HERE.html: this guide as a page
• LICENSE.txt, CITATION.cff

Links

• ATLAS: https://indahl.ai/atlas
• Electricity: https://indahl.ai/atlas/electricity
• Power plants: https://indahl.ai/atlas/power-plants
• Planned plants: https://indahl.ai/atlas/planned-plants
• Carbon: https://indahl.ai/atlas/carbon
• Code: https://github.com/Heuremen-LLC/indahl.ai
• Data on Hugging Face: https://huggingface.co/indahl
• OpenStreetMap: https://www.openstreetmap.org/copyright
• Epoch AI: https://epoch.ai/data/ai-data-centers
• EIA-861: https://www.eia.gov/electricity/data/eia861/
• EIA-860M: https://www.eia.gov/electricity/data/eia860m/
• EPA eGRID: https://www.epa.gov/egrid

Notes on the data

The text in json/scale.json refers to our build step and names a script and a file path. That file is not in this pack. The JSON files are kept unchanged, so the text stays as it was written.

In csv/utilities.csv the column ba_code holds the text NA in 23 rows (AK 11, HI 8, ME 2, NY 2). NA is a code in EIA's file and not a missing value. The pack's data does not define the code and the EIA files we hold do not either, so we leave its meaning to EIA. Pandas and many spreadsheets read the text NA as missing, so in pandas keep it with pd.read_csv("csv/utilities.csv", keep_default_na=False, na_values=[""]).

The ATLAS page counts 8 sites as not placed in a state, while the Epoch GeoJSON has 10 points without coordinates. The difference is 2 sites that got their state from the address text without a point.

The datacenters_epoch files split the JSON field placed in two. place_method is one of nominatim address, nominatim city/zip only, address match, corrected, or blank. place_match is the matched address text when there is one. The JSON keeps the original placed field.

Known data issues

This list comes from two rough checks and is not complete. It names records whose point may not match the stated state. The records are drawn as given and are not corrected in any file. Colossus 2 in datacenters_epoch sits about 57 m from the Tennessee and Mississippi line, and its state in the scale files comes from the simplified state outlines.

Rule 1, state box: a record is listed when its point lies more than 1 degree outside an approximate latitude and longitude box for its stated state. Rule 2, nearest same-state record: plants and planned plants are pooled by stated state, and a record is listed when the nearest other record of its state is more than 140 km away. Alaska and Hawaii are skipped by rule 2. Neither rule catches a wrong point that sits close to other records of the state or inside the box. A flag is a reason to look and not a finding. Some remote plants are flagged by rule 2 alone. The rules list 9 records.

• plants.json code 6584 "Stock Island": stated state FL, point at 24.563333, -81.7342. Flagged: 161 km from the nearest record of its state.
• plants.json code 64638 "Borderlands Wind, LLC": stated state NM, point at 34.159842, -108.894458. Flagged: 147 km from the nearest record of its state.
• plants.json code 59655 "Campbell County Wind Farm": stated state SD, point at 45.753333, -100.275. Flagged: 143 km from the nearest record of its state.
• plants.json code 70519 "Rangeley": stated state ME, point at 40.981022, -70.651702, which is in the Atlantic south of Nantucket. Flagged: outside the state box and 250 km from the nearest record of its state.
• plants.json code 70557 "AmericoldCA - Plainville, CT": stated state CT, point at 37.429933, -121.925782, which is in California. Flagged: outside the state box and 4125 km from the nearest record of its state.
• planned.json code 68296 "Mullin Creek #1 Generating Station": stated state MO, point at 38.00595, -97.92166. Flagged: outside the state box and 293 km from the nearest record of its state.
• planned.json code 64670 "Big Creek Solar LLC": stated state AR, point at 34.870404, -117.9656, which is in California. Flagged: outside the state box and 2121 km from the nearest record of its state.
• planned.json code 68876 "Okolona Solar": stated state MS, point at 37.693, -87.549. Flagged: outside the state box and 351 km from the nearest record of its state.
• planned.json code 68028 "Delta Solar Plant": stated state LA, point at 33.794007, -90.71392, which is in Mississippi, and its utility is Entergy Mississippi LLC. Flagged: 145 km from the nearest record of its state.

Terms

Each part of this pack keeps the licence of its source, and LICENSE.txt gives them by file. Files derived from OpenStreetMap are under the ODbL 1.0 with © OpenStreetMap contributors, and no further restrictions may be added to them. Files derived from Epoch AI are under CC BY 4.0 with Epoch's citation, and where a file holds both, both apply. EIA, EPA and US Census data are US government works in the public domain.

Heuremen LLC permits anyone to use, copy and change its own parts with attribution to Heuremen LLC and https://indahl.ai/atlas. For its own text files (the README, SOURCES.md, NOTICE.txt, LICENSE.txt, TESTED.md and START_HERE.html) that permission does not include use as AI training data. Its reservation against text and data mining and AI training covers those text files only and does not apply to any OpenStreetMap-derived or Epoch-derived file or to public-domain data.