The Power Implications Of Liquid Cooling
Liquid cooling does not pipe the problem away; it relocates power and rewrites the reliability model. On the efficiency side it swaps big fan loads for smaller pump loads and can idle the compressor, but every pump, exchanger, and rejection device is a watt the power plant must still feed, so the parasitic-power ledger must be kept honestly or the efficiency claim is fiction. On the reliability side it fuses the failure domains: a dense rack has so little thermal mass that losing cooling overheats it in seconds, so protecting compute power without protecting pump power buys only a hotter failure. Power and cooling must now be backed up, sized, and evaluated as one coupled system, because the site is only as available as the weaker of the two.
Knowledge check
Why must the cooling pumps in a dense liquid-cooled facility be placed on the same protected, backed-up power as the servers?
Show the answer
Because a dense rack has very little thermal mass relative to its heat output, so if a power event keeps the servers running on backup while the pumps lose power and the coolant stops, the rack overheats within tens of seconds, far faster than the backup can bridge to a generator. Protecting compute power without protecting pump power does not buy ride-through; it guarantees a hotter, faster failure, so the two must be backed up together.