Cooling, Water and the Physical Cost of an AI Boom

The environmental conversation about artificial intelligence has focused on electricity consumption, which is measurable and easy to illustrate. Water consumption is harder to see and, in some places, more contentious.

Run water over a heat exchanger and let it go

The traditional approach to cooling a data centre is evaporative: run water over a heat exchanger and let it evaporate, carrying heat away. It is efficient and cheap, and it consumes water that is not returned to the source.

• Evaporative cooling: low capital cost, high water consumption.
• Closed-loop with chillers: near-zero water use, substantially higher electricity use.
• Direct-to-chip liquid cooling: efficient, requires new hardware and building retrofit.
• Immersion cooling: very efficient, complicates servicing and hardware sourcing.

Residents asked to conserve next to a campus using millions of litres a day

Facilities are frequently sited where land and power are cheap, which often means water-scarce regions. Local opposition has followed: residents who are asked to conserve water while a new campus consumes millions of litres a day tend to notice.

Operators have responded by shifting new builds to closed-loop designs, publishing consumption figures, and funding local water infrastructure. The most sophisticated approach reuses waste heat for district heating, turning a cost into a contribution.

Different methodologies, few audited figures

Comparisons between operators are difficult because reporting methodologies differ and few publish audited figures. Regulators in several jurisdictions have begun requiring disclosure, which will make the comparison possible and is likely to make some of the numbers uncomfortable.

A decision made once, at construction

Choosing a cooling approach is a multi-decade decision made at construction time. An evaporative system is cheap to build and expensive in water; a closed-loop system is the reverse. Switching later means rebuilding the plant.

This is why the siting decision determines the environmental profile more than any operational choice made afterwards. A facility built in a water-rich region with evaporative cooling may be more efficient overall than one in a dry region using a less efficient closed loop.

Heat that is worth more next door

The most promising development is treating waste heat as a product rather than a problem. Several European facilities now supply district heating networks, converting an operating cost into a revenue stream and a public benefit.

The obstacle is proximity: heat cannot be transported far without losses, so recovery only works where a data centre is near the buildings that could use it. That is an urban planning question as much as an engineering one.

A practical point for anyone planning a build: the environmental choice is usually made by the site selection team before the engineering team is involved. By the time the cooling architecture is being debated, the water availability and the climate have already determined most of the answer.