Almost all the electricity a data centre consumes is converted to heat. Removing that heat, rather than supplying the power, is what shapes the building and much of its cost.

Every watt in becomes a watt of heat

Computing does not consume energy in any lasting sense. The energy arrives as electricity, performs work in the silicon, and leaves as warmth in the air and water.

A facility drawing tens of megawatts is therefore also a heat source of comparable size, concentrated into a footprint far smaller than any conventional building of that output.

Cooling capacity must match that output continuously, because the temperature inside a densely packed hall rises to damaging levels within minutes if cooling stops.

Airflow separation does most of the work

Racks are arranged so that all equipment draws cool air from one aisle and exhausts warm air into another, keeping the two streams from mixing.

Mixing is the main inefficiency in older halls, because recirculated warm air forces the cooling plant to supply colder air than the equipment actually requires.

Physical containment, with barriers enclosing one aisle or the other, allows supply temperatures to be raised considerably, and every degree of increase reduces the energy the plant consumes.

Raising the target temperature saves the most

Equipment tolerances are wider than intuition suggests, and operating standards have been revised upward as manufacturers confirmed reliability at warmer inlet temperatures.

A higher target means mechanical refrigeration runs for fewer hours a year, with outside air providing cooling directly whenever conditions permit.

This is why climate influences site selection so strongly. A cool location can operate with minimal refrigeration for most of the year, which changes the operating cost fundamentally.

Water and power availability decide the site

Evaporative cooling is efficient but consumes water, which has made water availability a planning consideration in dry regions and a subject of local objection.

Grid connection capacity is the other constraint, and in several regions the wait for a connection of the required size now exceeds the time needed to construct the building.

Sites are therefore chosen for power and cooling conditions first, with proximity to users a secondary factor except where latency genuinely governs the application.

Density is pushing toward liquid

Processors designed for intensive workloads dissipate far more heat per rack than earlier generations, beyond what moving air can remove economically.

Liquid cooling, carrying heat away through pipes to a cold plate or by immersing components entirely, handles far higher density in the same floor area.

It also produces waste heat at a temperature useful for district heating, which turns a disposal problem into a supply opportunity where a suitable network exists nearby.