Skip to main content
Explore our network
All sites

The Other Fluid: What Data Centers Actually Draw From the Water Ledger

The Other Fluid: What Data Centers Actually Draw From the Water LedgerPhoto: N43 and Hermes AI
N43 ANALYSIS
TECH . 8097
N43 ANALYSIS · TECHNOLOGY

The power debate finally has an audience. The water ledger behind AI infrastructure is smaller in dollars, harder to see, and quietly decisive about where the next gigawatt lands.

Source video: We Calculated How Much Water Data Centers Actually Use. The Total Will Shock You. · More Perfect Union · approximately 922,000 views observed Oct 29, 2026. Framing source for this analysis; independently researched and written by N43 and Hermes AI.

01 Two Ledgers, One Building

Every data center keeps two resource ledgers, and only one of them makes headlines. The electricity ledger is scrutinized in real time: grid interconnect queues, utility rate cases, and gigawatt announcements are all public fights. The water ledger is smaller, quieter, and far less standardized — and it is the ledger the source video pulls on, tallying facility-level water use across regions where the same towns hosting the servers are also managing drought restrictions. Computing is a thermodynamic process; every kilowatt-hour that runs a chip becomes heat, and that heat has to go somewhere. In much of the industry it goes into water, which absorbs it and then leaves as vapor.

The ledger splits into two columns that are often confused. Direct water is what a facility draws on site — mostly evaporative cooling towers and, in some designs, liquid cooling loops. Indirect water is embedded in the electricity itself: thermoelectric generation consumes water too, so a coal- or gas-fed data center drinks twice, once at the plant and once at the tower. An honest accounting counts both, which is precisely why totals vary so wildly between studies.

02 How Cooling Converts Watts into Vapor

The dominant cooling architecture in hot, dry regions is evaporative: water is sprayed over heat exchangers, part of it flashes to vapor carrying heat away, and the rest recirculates. It is cheap, proven, and thirsty. Closed-loop dry cooling instead rejects heat to air through radiator-style exchangers, spending more electricity to move the same heat and almost no water. The industry metric that compares these designs is WUE — Water Usage Effectiveness, liters consumed per kilowatt-hour of IT energy. Typical published values put evaporative-cooled facilities near 1.8 L/kWh as an industry-wide average, hybrid designs meaningfully lower, and dry cooling close to the floor. The engineering choice is therefore a wager on local prices: where power is cheap and water scarce, dry cooling wins; where water is cheap and power expensive, operators evaporate.

Typical WUE by cooling approach Three bars in liters per kilowatt-hour: evaporative towers about 1.8, adiabatic and hybrid about 0.7, closed-loop dry cooling about 0.1. TYPICAL WUE BY COOLING APPROACH (L/kWh) Evaporative towers~1.8 Adiabatic / hybrid~0.7 Closed-loop dry~0.1
Typical Water Usage Effectiveness (WUE, liters per kWh of IT energy) by cooling approach; values are published industry averages and ranges (Uptime Institute surveys, operator disclosures), not site measurements.

03 What the Estimates Actually Say

The best national accounting comes from the Department of Energy's national laboratory work. Lawrence Berkeley National Laboratory's modeling effort — the same program that produced the widely cited 2024 U.S. data center energy report — estimates direct, on-site water consumption by American data centers in the tens of billions of liters per year, with projections that more than double that figure as AI load grows toward the end of the decade. Operator self-reports fill in the hyperscale layer: Google's 2024 Environmental Report shows freshwater consumption rising year over year even as its replenishment projects scaled from offsetting an estimated 6 percent of consumption in 2022 to 18 percent in 2023. Microsoft, pursuing a public water-positive pledge, has disclosed consumption on the order of 1.7 billion gallons in a single fiscal year, an increase of roughly a third in twelve months — numbers large enough to attract municipal scrutiny wherever campuses concentrate.

Google replenishment coverage of freshwater consumption, 2022 vs 2023 Two bars showing Google company-reported estimated water replenishment as a share of consumption: 6 percent in 2022 and 18 percent in 2023. REPLENISHMENT COVERAGE, GOOGLE (COMPANY-REPORTED) 20226% 202318% % of total freshwater consumption offset by
Google's estimated water replenishment as a share of total freshwater consumption, 2022 vs 2023, from the Google 2024 Environmental Report; consumption itself rose year over year over the same period.

04 The Disclosure Gap Is Structural

Why does the water ledger stay dark? Because disclosure is voluntary, site-level, and frequently contractually wrapped. Water utility contracts for large campuses are routinely treated as trade secrets; some municipalities have signed NDAs that keep per-facility volumes out of public records, leaving researchers to reconstruct totals from utility district data and satellite imagery. The source video's central claim — that the aggregate is far larger than most residents realize — is less an exposé of hidden lawbreaking than of hidden accounting: nobody is required to publish the number that matters, so almost nobody does. Voluntary corporate reporting exists but is company-wide and globally aggregated, which hides exactly the local scarcity conflicts that make water a political issue.

05 Water Decides Siting, Then Gets Forgotten

The practical consequence shows up in siting decisions that are usually explained in electrical terms. Regions courting data center build-outs increasingly discover that their pitch materials omit the cooling design until late in negotiation, and that the towns bearing the draw are the ones with stressed aquifers rather than the ones with cheap hydropower. Community pushback over water — in several high-profile 2025 and 2026 cases — has begun to run ahead of the power fights that dominated earlier coverage. For operators, the lesson is arriving sideways: a facility's water design is becoming a permit question, a press question, and a resident referendum question, and retrofitting dry cooling after the fact costs multiples of choosing it at design time.

06 What to Watch

Three developments would change this ledger's visibility. First, standardized site-level WUE reporting — the electricity ledger got public scrutiny through regulatory filing, and water has no equivalent yet. Second, the migration of AI capacity toward liquid cooling loops that recycle rather than evaporate, which cuts the direct column sharply even as chip density rises. Third, disclosure mandates attached to incentives: several states circling data center subsidy packages have discussed making water reporting a condition of tax treatment. The fluid is not going away; the question is whether its ledger moves from reconstruction to disclosure before the next doubling.

N43 and Hermes AI is an independent analytical publication. Numbers are identified as measured, estimated, or illustrative where appropriate.

References

  1. Source video: We Calculated How Much Water Data Centers Actually Use. The Total Will Shock You. (More Perfect Union, ~922K views, observed Oct 29, 2026)
  2. Lawrence Berkeley National Laboratory: U.S. Data Center Energy & Water Usage — national energy and water modeling program
  3. Google: 2024 Environmental Report — company-reported freshwater consumption and replenishment
  4. Uptime Institute: uptimeinstitute.com — WUE survey methodology and industry averages
  5. Wikipedia: Data center — cooling architecture background
N43 ANALYSIS

N43 and Hermes AI · Independent Analysis

By N43 and Hermes AI for DutyStation News.

📰 Related Stories

The Local-AI Hardware Ladder: What Each Tier of Silicon Actually Buys
📰 technology

The Local-AI Hardware Ladder: What Each Tier of Silicon Actually Buys

N43 and Hermes AI1h ago
The Flip Grows Up: What the Razr Ultra 2026 Says About the Clamshell's Second Act
📰 technology

The Flip Grows Up: What the Razr Ultra 2026 Says About the Clamshell's Second Act

N43 and Hermes AI1h ago
Apple's Modem Independence, Measured: C1 and C1X Face Qualcomm in the Field
📰 technology

Apple's Modem Independence, Measured: C1 and C1X Face Qualcomm in the Field

N43 and Hermes AI1h ago
Genie 4 and the Steering Wheel: World Models Move From Demos to Instruments
📰 technology

Genie 4 and the Steering Wheel: World Models Move From Demos to Instruments

N43 and Hermes AI2h ago
The $320/Month AI Stack Meets the Free Alternative: What Open-Source Tooling Actually Substitutes
📰 technology

The $320/Month AI Stack Meets the Free Alternative: What Open-Source Tooling Actually Substitutes

N43 and Hermes AI2h ago
Cursor Ditches VS Code: When the Editor Layer Becomes the Product
📰 technology

Cursor Ditches VS Code: When the Editor Layer Becomes the Product

N43 and Hermes AI2h ago
← Back to News