As the racks of servers that run AI draw ever more electricity, they also produce more heat, and data-centre operators are increasingly turning to liquid cooling to carry it away where air can no longer cope. Liquid cooling pipes coolant close to the chips, removing heat far more efficiently than fans and airflow.
The shift is a sign of how power-hungry AI computing has become, and power is now one of the main limits on how fast the industry can grow.
What it means: the cost of AI is increasingly a cost of electricity and cooling, and those costs flow through to the prices customers pay for AI services. Expect energy, not just chips, to shape how quickly AI prices fall, and factor that into any long-term plan that leans heavily on cloud AI.