The Thirsty Machine
AI runs on machines, machines make heat, and some ways of removing heat evaporate water. This folder set keeps engineering facts, government guidance, operator disclosures, projections and scriptural themes in separate drawers — and refuses to file a site claim without a document behind it.
Side Quest — Infrastructure File
Quest Briefing
Every calculation a machine performs leaves the building as heat, and heat has to be put somewhere. Some cooling designs put it into evaporated water. The core question of this quest: what happens when a rapidly expanding industry needs water and power in places that were already short of both — and how much of what circulates about it is actually sourced? Facts and scripture are filed separately here, and stay that way.
- Files
- 0/13
- Clue links
- 27
- Case status
- OPEN
Evidence reviewed — 0%
Fact File — Why AI Uses Water
Computation becomes heat. Heat has to go somewhere.
Before any argument about scarcity, the mechanism: what a data center actually does with water, which cooling designs evaporate it, which do not, and why 'water used' means two different things depending on whether you are standing at the site or at the power plant.
From Compute to Heat to Cooling
Essentially all electrical energy drawn by servers leaves the building as heat. Removing that heat is the entire reason a data center touches water at all.
Direct Onsite Water vs Indirect Water
Two different quantities are both called 'data center water use'. Comparing a figure of one type with a figure of the other is the most common error in this debate.
Documented Water Footprint
Sourced figures only, each with a publisher and a date.
Government guidance and operator disclosures, quoted as published. Where a figure is self-reported it is stamped COMPANY DISCLOSURE, because self-reporting is a different evidence class from an independent measurement.
WUE — The Standard Measure
U.S. Department of Energy (FEMP / Better Buildings) data center guidance
Water Usage Effectiveness is the accepted metric for data center water performance: annual site water use divided by annual IT equipment energy, expressed in litres per kilowatt-hour.
Microsoft — Published WUE and Zero-Water Designs
Microsoft environmental sustainability reporting and datacenter blog, 2024
One of the few hyperscale operators publishing a fleet-level water metric and a specific zero-water-for-cooling design commitment. Self-reported, and stamped accordingly.
Where Water Stress Matters
An intake board. Deliberately empty until entries are sourced.
The interesting question is siting: whether the largest campuses are being built in basins that were already short of water. This folder holds the field template for logging a site. It contains no site-specific claims, because none have been sourced into the archive yet.
Field form — site intake
No entries filedLog a data center site
Blank by design. A site entry is only admitted when every field below can name its document. Until then the board stays empty rather than carrying an unsourced claim.
- Site / campus
- Operator and facility name as it appears on the permit.
- Location
- County and state or region. Municipality if the supply is municipal.
- Watershed / aquifer
- Named basin or aquifer the supply is drawn from.
- Cooling method
- Evaporative tower, chilled water, air-cooled, closed loop, direct-to-chip.
- Supply type
- Potable municipal, reclaimed non-potable, groundwater, surface water.
- Withdrawal (annual)
- Volume taken. State the unit and the reporting year.
- Consumption (annual)
- Volume not returned. Do not substitute withdrawal here.
- Drought status
- Official classification for the basin at the reporting date.
- Community dispute
- Permit challenge, moratorium, council decision — with the record cited.
- Source document
- Publisher, title, date. No entry is complete without this field.
The Energy-Water Nexus
The meter at the fence line is not the whole bill.
A site can run near-zero onsite water and still carry a substantial water footprint through the electricity it buys. Thermoelectric generation withdraws and consumes water; wind and solar photovoltaics essentially do not. The power mix decides the indirect figure.
Future Growth
Intensity per unit of compute versus total demand.
These are two different trend lines and they can point in opposite directions at the same time. Water per unit of compute can fall while total water rises, if the number of units rises faster. Most public arguments confuse the two.
Two Trend Lines, Often Confused
Water intensity per unit of compute and total water demand are separate quantities. Efficiency gains do not automatically reduce total draw.
Reclaimed Water, Closed Loops and Heat Reuse
The mitigation options are known and mostly unglamorous. The question is deployment rate, not invention.
Prophecy Connections
Possible connection only. No fulfilment claims on this board.
Scripture speaks repeatedly about drought, dried rivers, scarcity and stewardship. Noticing that a modern infrastructure story rhymes with those themes is legitimate. Asserting that AI causes or fulfils them is not, and nothing in this folder does so.
Possible connection only — no fulfilment claims
Key Question Board
What the archive would need answered to move any of this.
Each question is stated so that a specific document could settle it. Anything that cannot be settled by a document is marked as such.
Source Cards
Primary material only.
Publisher, document and date. Secondary aggregation and unattributed internet figures are kept out of this folder on purpose.
Source Card — U.S. Department of Energy
U.S. Department of Energy — data center energy and water efficiency guidance
Federal guidance on data center cooling, water efficiency and the WUE metric, issued through the Federal Energy Management Program and the Better Buildings data center programme.
Source Card — Microsoft
Microsoft datacenter blog, December 2024; Microsoft environmental sustainability report
Operator disclosure of fleet WUE and of a closed-loop, zero-water-for-cooling datacenter design.
Source Card — Entry Rules for This Folder
What may and may not be filed here.
Evidence stamps
How to read a file
- DOCUMENTED FACTEstablished engineering or physical fact, not in dispute. (6)
- COMPANY DISCLOSUREReported by the operator itself. Self-reported, dated, rarely audited. (2)
- GOVERNMENT SOURCEPublished by a government agency or national laboratory. (2)
- ESTIMATEA modelled or projected figure. Depends on assumptions. (2)
- POSSIBLE CONNECTIONA thematic link only. Not a claim of cause or fulfilment. (1)
- OPEN QUESTIONUnanswered on the evidence currently on file. (4)
Final screen
Case Status — OPEN
The mechanism is settled. The scale, the siting and the net effect of new cooling are not.
0 of 13 files reviewed on this device
Established
- — Data center cooling load, and therefore water use, is a direct function of heat rejected.
- — Evaporative and adiabatic cooling consume water; sealed closed-loop and air-cooled designs largely do not.
- — WUE (litres per kWh of IT energy) is the standard onsite metric, per federal guidance.
- — Onsite water and electricity-embedded water are separate accounts; so are withdrawal and consumption.
- — Zero-water-for-cooling hyperscale designs exist and are being deployed, per operator disclosure.
Still open
- — Whether the largest AI campuses cluster in water-stressed basins.
- — Actual withdrawal and consumption at named sites, reported separately.
- — Whether falling water intensity outruns rising total compute demand.
- — How much closed-loop water saving is offset by additional electricity and its embedded water.
- — How widespread local allocation disputes actually are, on the record rather than by anecdote.
Closed / refuted
- — That AI water use is a hoax — the cooling mechanism is basic thermodynamics and is not in dispute.
- — That every data center evaporates large volumes of potable water — cooling design and supply type vary by site.
- — That any biblical passage names industrial computing as a cause of drought — none does.