Methodology
Where the data comes from, what the labels mean, and how fresh it is.
US Energy Map combines two very different kinds of data: a federal registry for battery storage, and hand-curated research for data centers. We keep them transparent and side by side rather than blending them into a single opaque number.
Battery storage (BESS)
Battery entries are ingested automatically from EIA-860M, the U.S. Energy Information Administration's preliminary monthly inventory of electric generators. We filter the inventory to battery technology and aggregate individual generator rows up to the plant level (a single site often reports several generators).
- Coordinates are the exact site coordinates reported to the EIA.
- Power capacity (MW) is the nameplate value; energy capacity (MWh) is shown where reported — EIA reports it for operating sites but not on its planned-generator sheet, so under-construction and planned batteries show MW only. Each build records the exact EIA-860M file it was ingested from.
- Entries from this pipeline carry an
eia-id prefix.
EIA-860M is not exhaustive. It captures utility-scale generators (roughly ≥1 MW) that report to the federal government. Behind-the-meter, residential, and small commercial batteries are not included.
Data centers
There is no federal registry of data centers, so these entries are hand-curated. Every data center on the map has a public source URL — we don't publish a project we can't point to. Sources include operator press releases (AWS, Google, Microsoft, Meta, Apple), trade reporting such as Data Center Knowledge and Black Ridge Research, and the public portfolios of major colocation providers (Equinix, Digital Realty, CyrusOne, CoreSite, NTT, DataBank, and others).
- Where an exact site isn't publicly disclosed, coordinates fall back to the city or county centroid.
- Coverage prioritizes named hyperscale and large colocation campuses; it is not a complete census of every facility.
Status & capacity definitions
| Label | Meaning |
|---|---|
| Operating | The project (or a phase of it) is energized and in service. |
| Under construction | Construction has begun but the site is not yet fully in service. |
| Planned | Permitted or in active development with a target year, but construction has not started. |
| Announced | Publicly announced, but earlier-stage than Planned: site work has not begun and key permits, agreements, or financing may still be pending. |
Capacity is reported in megawatts (MW) of power. Large AI training clusters are evolving quickly and figures may lag announcements.
Multi-phase campuses
Hyperscale campuses are built in phases over years, so a single row has to summarize a moving target. Two conventions keep that honest:
- Capacity is the full announced campus buildout. If an operator has committed to a 5 GW campus, the entry says 5,000 MW even while only the first phase is energized.
- Status reflects the most advanced phase. A campus is Operating as soon as any phase is live, even if later phases are still under construction.
Where the capacity figure is an estimate — or the buildout it describes is incomplete or in flux — the entry carries a capacity_estimated: true flag.
Optional fields
Hand-curated entries may carry three optional fields when the information is publicly reported; they are omitted when unknown:
| Field | Meaning |
|---|---|
investment_usd | Total announced investment in US dollars, from the most recent public figure. |
energy_source | How the project is powered: grid, onsite_gas, nuclear_ppa, solar_storage, or mixed. |
capacity_estimated | The MW figure is an estimate, or describes a buildout that is not yet complete. |
Deduplication
Because batteries are auto-ingested and data centers are hand-entered, an automated battery entry can overlap a curated one. The ingest skips an EIA entry when it falls within ~10 km of a hand-curated entry, or matches one strongly by name in the same state. Hand-curated entries are always preserved.
How fresh is it?
The battery dataset refreshes on a monthly cadence: an automated job re-pulls the latest EIA-860M release around the 26th of each month and proposes an update. Data center entries are updated as announcements warrant. The "updated" date shown across the site reflects the most recent dataset build.
Found something wrong or missing? The dataset and its history are public on GitHub.