Methods & sources

How plumb knows what it knows

Every number has a tag that says where it came from. Tap any tag on plumb to see its source and date.

What the tags mean

  • ✓ Verified

    From an official record: a government body, regulator or public agency, like the SCC, JLARC, EIA, the Census Bureau or a county.

  • Researched

    From a cited source that isn't an official record: a news report, a study, a tracker such as Cleanview, or a company page.

  • ≈ Calculated

    Our own math from public capacity and statewide totals. Always shown with its method and range. Not a measurement.

  • Census

    Census figures for the neighborhood (tract) around a site. They describe what is there, never what a data center caused.

  • Estimated

    Not fully checked: low research confidence, sources that conflict, or no source recorded. Treat it as an estimate.

  • Demo

    Illustrative only, and badged wherever it appears.

  • No data

    Nothing to show, or a live lookup failed. A blank is never shown as zero.

How we do the math

Where nothing is published, we calculate. These are the formulas behind every Calculated tag.

  • Electricity a year

    MW × 8,760 h × 23 %

    Show the math

    Capacity (MW) × 8,760 hours × 23 % average draw, with a range of 18 to 30 %. The 23 % comes from Dominion's 2025 data-center peak (4 GW, in its letter to PJM of 6 Jan 2026) divided by the 17.3 GW of operating capacity Cleanview lists.

  • Homes equivalent

    MWh ÷ 11.9 MWh per home

    Show the math

    Estimated MWh ÷ 11.9 MWh, the yearly electricity use of an average Virginia home (EIA 2023, Table 5A). It is a way to picture the size, not a count of homes served.

  • Water a year

    Area total × share of capacity

    Show the math

    Sites in Loudoun share Loudoun's 952 million gallons (2025) in proportion to their capacity. Everywhere else, sites share Virginia's 2.1 billion gallons (2023, JLARC) in proportion to capacity. Planned sites are counted "if built". These figures are likely low.

  • Census context

    ACS 2019–23, tract around the site

    Show the math

    Median rent, home value and household income come from the ACS 2019–23 5-year estimates for the tract around a site. They show what is nearby, not that a data center caused it.

  • Totals over a radius or county

    Blanks left out, and counted

    Show the math

    Totals leave out facilities without capacity data and canceled sites, and say how many were left out beside the number. A blank is never counted as zero.

Good to know

Trade secret

Virginia keeps each site's power and water use secret

No public figure exists for any single data center. So every power and water number for a site on plumb is tagged Calculated: the right order of magnitude, not a meter reading.

How sure our research is

Every facility carries a rating. We show it, never hide it.

High
101
well documented
Medium
499
partly documented
Low
153
thin or conflicting

Facts from low-confidence rows, or rows with no recorded source, are tagged Estimated.

Limits

  • Distances are straight lines from your address, not driving distance.
  • Locations and capacity come from Cleanview and can be approximate. 67 facilities list no capacity, so they get no estimate.
  • Zero complaints means none found in our research, not that there is no noise.
  • Census tracts can be larger than the immediate neighborhood.
A caution on rent “change”It tracks inflation, not data centers.

The census data we hold includes a rent change figure. It compares two overlapping 5-year periods (2017–21 against 2019–23) in nominal dollars. It is not a two-year change, it is not adjusted for inflation, and the typical value (about +12 %) matches inflation. It is never a data-center effect, so plumb does not headline it. If it appears, it is labeled “ACS 2017–21 vs 2019–23 (overlapping, not inflation-adjusted)”, shown beside the county value and tagged Estimated.

Sources

58 sources behind the numbers on plumb, from data providers to the reports our research cites.

Show all 58 sources

Data and models (8)

Cited in our research (50), most used first