Data-center water intake and consumption measure different things

For 2024, Google’s own report gives The Dalles, Oregon, 461.1 million gallons withdrawn and 361.4 million consumed. The city’s records, as OPB reported them, list Google’s water use at 434.4. That is 26.7 million gallons below Google’s withdrawal and 73.0 above its consumption, and the city’s own master plan defines “consumption” as billed, metered use, which is a different measure from Google’s withdrawal minus discharge. This desk-researched part of The Power Plant Next Door lines the records up for that reporting year, using the one place where we could find both.

The distinction matters at a public hearing. A utility needs to know how much water it must supply and when. Residents may also ask how much returns, where it goes and what remains unavailable locally. One annual total cannot answer every question. Below, we set the labels in the records side by side and compute the gaps between Google’s two numbers and the city’s one.

Google’s 2024 water ledger

The 2025 Google Environmental Report (location table on PDF page 110 and continued on page 111; the same figures in Appendix A on page 114) covers these location figures for calendar year 2024. Its columns distinguish incoming water (withdrawal), water leaving the facility (discharge) and consumption. The company calculates consumption as withdrawal minus discharge. The table below reproduces two locations from the same report, in the same units. These are company-reported location totals. Ernst & Young LLP reviewed them within Alphabet’s Schedules of Select Environmental Metrics under a limited-assurance review (report dated June 16, 2025); a review is a narrower check than an audit, and EY stated that it was not aware of material modifications needed.

Google’s methods matter as much as its column headings. The report describes meter and invoice data where available, with estimates used where actual data are unavailable. Discharge estimates can include engineering calculations at data centers. The published values combine measurements and estimates; consumption is then a calculated result.

The report’s term withdrawal also needs its own context. In USGS water-use terminology, withdrawal describes water removed from a source, while delivery describes water supplied at a point of use. An operator’s incoming-water accounting can include utility supplies, so city and company records may cover different meters or supplies. Google’s endnote 164 adds that, unless otherwise specified, its withdrawals are potable water.

One line on the small row: Pflugerville’s displayed consumption of 0.01 beside equal withdrawal and discharge of 0.1 is a rounding effect, because Google rounds to the nearest tenth of a million gallons and, under 50,000 gallons, to the hundredth (PDF page 114), so a displayed 0.01 stands for 5,000 to under 15,000 gallons. Google labels the site air-cooled, which says nothing about how the other two sites are cooled.

The Dalles: Google’s numbers beside the city’s

OPB’s January 23, 2026 article, headlined “The Dalles’ mayor called OPB’s data center story inaccurate. Here are the facts,” answers the mayor’s criticism of the station’s January 15 investigation. OPB found one error in the earlier story, a projected water-rate increase, corrected it and apologized, and said the error does not change the story’s main points; the article does not flag any water-use figure as wrong. In it, OPB said a city spreadsheet provided on December 24, 2025 in response to a public records request listed Google’s water use in The Dalles as 383.9 million gallons in 2023 and 434.4 million gallons in 2024. The Google report used here covers 2024, so the 2023 figure has no Google counterpart in this comparison. OPB does not define the city’s measure further, and its article carries no Google statement about water use. OPB also reports that Google uses a third of the overall water the city consumes. The city had earlier fought for 13 months to keep Google’s water use confidential, in a lawsuit it filed on Google’s behalf against The Oregonian. It dropped the fight in December 2022 and committed to release the company’s water consumption in future years, according to the Associated Press. The table puts the 2024 figures in one place and subtracts them.

The city’s figure sits closer to Google’s withdrawal than to its consumption. It is 73.0 above Google’s consumption, so it is not comparable with that column unless the city built its figure by the same subtraction. The city’s own published use of the word, below, suggests it did not. None of this shows that either record is wrong. The two series may differ in meters, supplies or dates.

The city’s own meaning of consumption

The city publishes its own definition of the word. The Dalles’ Water System Master Plan Update (November 2024), section 2.3.1, says its water consumption data “is taken from the City’s customer billing records and includes all revenue metered uses,” and keeps that apart from production, the water supplied to the distribution system. Table 2-1 puts citywide consumption at 1,190 million gallons in 2022. In the city’s usage, consumption is billed, metered water. Google’s report uses the same word for withdrawal minus discharge. Two documents say consumption and mean different measures, so a hearing that sets one beside the other without saying which is which is comparing unlike things.

The plan does not mention Google and does not say how the Google line in OPB’s spreadsheet was built. OPB’s caption describes the spreadsheet as showing how much water Google used annually in The Dalles and the city’s total water consumption every year, which suggests the Google line is kept on the same footing as the city’s total. If the line comes from billing records, as the plan’s definition of consumption would suggest, it measures water delivered. That is the same kind of quantity as Google’s withdrawal (461.1) and a different one from Google’s net consumption (361.4). This is our inference from the two definitions, and the city can confirm or correct it.

Four questions still open

Four questions remain open, and they are for the city and Google, not explanations we have confirmed. Does the city series count only municipal deliveries billed to Google, as the plan’s billing-record definition suggests, or also other supplies? Does it run on a calendar year, as Google’s report does, or on a fiscal year? Do both cover the same campus facilities? Does the 434.4 figure count water before any return to the city system, the way withdrawal does? A resident at a hearing can ask for the city’s billing records for the same dates and meters.

Cooling equipment outside the internal loop

A cooling-system description can leave out a boundary. The Department of Energy’s Federal Energy Management Program describes internal chilled-water loops that pass heat to a separate condenser loop and cooling tower, and its cooling-tower guidance explains that evaporation and blowdown, the removal of circulating water to control dissolved solids, are different paths. A closed internal loop is therefore not evidence of zero facility water consumption; a system diagram has to extend through the final heat-rejection equipment.

The limits of a campus total

The spread between The Dalles (361.4 million gallons consumed) and Pflugerville (0.01) in Google’s table is a documented difference in reported annual totals. It does not isolate the cause. The table does not provide a matched comparison of computing work, utilization, weather and cooling design. A larger total alone cannot identify which facility uses less water for an equivalent amount of computing.

The same boundary problem applies to electricity. Water used at a power plant to produce a campus’s electricity is separate from water entering that campus. A claim about direct on-site consumption should say so. A broader claim about the water associated with computing needs an electricity-related estimate, its geographic assumptions and its method; that estimate should not silently replace a facility’s reported total.

For a proposed project, annual demand also leaves the peak supply question open. Ask for expected summer or peak-day deliveries alongside annual volumes, and separate projected full build-out from measured operating-year results. A design forecast is useful planning evidence when labeled as such. It becomes an observed result only when operating records support it. The records compared here do not determine the local effects of an unnamed future campus.

Correction, October 29, 2026: The earlier version mixed city water-use figures with company consumption figures and attributed differences between campuses to cooling design without a matched comparison. It also carried a pollution-and-lawsuit roundup whose primary records were not verified; removing it does not show the reported allegations were false. The page now separates the measures, sets Google’s 2024 figures for The Dalles beside the city figure OPB reported, gives the city’s own definition of consumption and describes the OPB article as what it is, a reply to the mayor’s criticism. The Council Bluffs row was cut.

Editor written. AI used for spelling and grammar only.