Muskeology
Frontier tech, minus the hype

AI & Compute

Data Centre Cooling And The Water Question

Cooling dense computing racks means moving heat out of a building continuously, and the method chosen determines whether the facility consumes water, electricity or both.

Detailed image of a circuit board featuring capacitors and intricate pathways, ideal for tech-related content.
Detailed image of a circuit board featuring capacitors and intricate pathways, ideal for tech-related content. · Photo via Pexels

Every watt a server consumes becomes heat that has to leave the building. How a data centre removes that heat shapes its electricity use, its water use and where it can be built.

Compute density changed the problem

Racks packed with accelerators draw far more power in the same floor area than conventional servers did. The heat produced per square metre has risen sharply as a result.

Air can only carry so much heat before the volume required becomes impractical. Fans grow larger, noise increases, and the temperature difference across the room becomes hard to control.

That limit is why liquid cooling has moved from a specialist option to a mainstream requirement for high-density installations, changing how buildings are designed.

Evaporation trades water for electricity

Cooling towers work by evaporating water, which removes a large amount of heat for relatively little pumping energy. The technique is efficient and long established.

The cost is consumption. Evaporated water leaves the system permanently, and additional water is discharged periodically to prevent minerals from concentrating in the loop.

Closed-loop alternatives use chillers instead, which recirculate a refrigerant and consume no water. They draw considerably more electricity to do the same job.

Climate decides which trade-off is available

Evaporative cooling works best in dry air, because dry air absorbs moisture readily. In humid conditions the same equipment removes much less heat.

Cool climates allow outside air to be used directly for much of the year, avoiding both water and mechanical cooling during large parts of the calendar.

Operators therefore site facilities partly on weather. A location that is cool and dry lowers running costs in a way no equipment choice can match.

Local supply matters more than national totals

Water use by data centres is often discussed as a national figure, which obscures the point. The pressure is local, falling on a specific catchment or utility.

A facility drawing from a stressed aquifer creates a different problem from one drawing from an abundant river, even if both consume identical volumes.

Some operators now use treated wastewater or industrial supply rather than drinking water, which addresses the competition directly without reducing the volume used.

Heat leaving the building is still useful

The warmth rejected by a data centre is low-grade, meaning it is not hot enough to drive industrial processes. It is warm enough for buildings and greenhouses.

Capturing it requires a nearby customer and a distribution network, which is why the practice is more common where district heating already exists.

Where the infrastructure is absent, retrofitting it usually costs more than the heat is worth, so the energy is rejected to the air instead.

Tobias Nkemelu
AI & Compute, Muskeology

Tobias builds and breaks machine learning systems for a living, which makes him a difficult audience for benchmark announcements.

More from Tobias →

Also by Tobias Nkemelu

AI & Compute

What to actually worry about with AI

A great deal of the risk discussion is about scenarios, and a shorter list of problems is already causing measurable harm.

Tobias Nkemelu··3 min read

AI & Compute

Where AI systems actually fail in production

Not in the model. In the data pipeline, the distribution shift, the feedback loop and the assumption that the world stays still.

Tobias Nkemelu··3 min read

AI & Compute

Evaluating an AI product claim

A short checklist for reading announcements, which mostly consists of asking what was measured and against what.

Tobias Nkemelu··3 min read

Robotics

Autonomous mobile robots outside the warehouse

Hospitals, hotels, factories and pavements — where wheeled autonomy has spread, and the specific reasons each environment is harder than a warehouse.

Lena Brandt··3 min read