Water-use figures on Gridwatch

How we source and label water data, and the explicit limits of what our figures should be used for.

The two data classes

Every water-use figure on Gridwatch is labeled with one of three confidence levels. The label is structurally part of the data, not a footnote — every CSV export, API response, and UI surface carries it.

Disclosed

The figure comes directly from an operator disclosure: a corporate sustainability report, a filed zoning application, a county records-request response, a state utility-commission docket, or a press release with attributed numerics.

When you see "Disclosed," the source and year are cited inline. Click the citation to read the source document.

Estimated

A figure computed by Gridwatch from three inputs: the facility's IT capacity in MW, its cooling system type, and the climate zone of the host state. Output is a ±30% confidence band, never a point estimate. An estimate is only produced when all three inputs are themselves sourced — which today is rare, because capacity and cooling type are usually undisclosed.

When you see "Estimated," the figure is always shown as a range with the methodology attached. An estimate never overwrites a disclosed figure, and it is never presented as a measurement.

Unknown

No operator disclosure exists, and the inputs needed for an estimate (capacity or cooling type) are missing. The water row shows "Not disclosed." We do not guess.

How estimates are computed

  1. Base water-use effectiveness is set by cooling type. Industry-standard values: evaporative ≈ 1.80 L/kWh, hybrid ≈ 0.90, closed-loop liquid ≈ 0.05, air-cooled ≈ 0.05, immersion ≈ 0.02.
  2. Climate multiplier adjusts for regional evaporation efficiency: hot/dry × 1.35, hot/humid × 1.00, mild × 0.85, cold × 0.65.
  3. Annual utilization is held at 70 % — the industry average for steady-state operating facilities. Sites flagged as "under construction" or "permitted" use the same factor since the number describes their projected operating-state withdrawal.
  4. Output is a ±30 % band around the central value to make uncertainty structural.

State-level climate granularity is coarse. Texas coastal facilities and Texas inland facilities are estimated identically; we know this is wrong and we'll refine to county-level when capacity allows. The ±30 % band is wide enough to cover this in most cases. Where we know state granularity is misleading for a specific site (e.g., a high-elevation Texas facility with desert climate), the site-specific methodology note overrides the default.

What these figures are not certified for

Gridwatch water figures are intended for general market intelligence and editorial research. They should not be used as the basis for:

For those purposes, the right path is to contact the operator directly or file a public-records request with the host county or state utility commission. Gridwatch updates its figures whenever authoritative disclosures supersede our estimates — and we'll replace an estimate with the disclosed figure within seven days of publication, with the prior estimate preserved in the change log.

If you spot a figure that looks wrong

Tell us through the contact form — the facility and the source document showing the correct figure. We update typically within 24 hours and credit the source. If you're a facility operator, the same form works — we'd rather use your disclosed number than our estimate.

Why this exists at all

Water use is increasingly contested at the county level. Mesa, Newton County, Holly Ridge, The Dalles — every one of those fights turned on whether anyone outside the operator had clear water-use numbers. The four-year litigation it took to get Google's The Dalles figures is the canonical example of what happens when this data lives only with operators.

Gridwatch's position: the public deserves figures that are transparent about their provenance and clearly bounded in their confidence. An estimate plus its methodology is a deliverable; either one without the other is not.

The honest state of play, today: almost no operator discloses water use, and for most facilities the inputs required for an estimate are not on the record either. The result is that "Unknown" is by far the most common label in the registry, and we would rather show you that than fill the column with numbers we cannot source. That is a finding about disclosure, not a defect in the data.

It is also changing. Florida's Chapter 2026-65, effective July 1, 2026, establishes a distinct consumptive use permit process for large-scale data centers and allows regulators to require reclaimed water as a condition of approval. Consumptive use permits are public records. As the largest facilities move through that process, water figures will move from "Unknown" to "Disclosed" — sourced to a filing, not to an estimate.