What happened
A United Nations University report reveals data centres' environmental footprint, driven by AI, will double by 2030, rivalling some of the world's largest countries. Last year, global data centres consumed 448 trillion watt-hours of electricity, producing 208 million tons of carbon dioxide and 1.2 trillion gallons of water, comparable to Argentina's total emissions. By 2030, data centres are projected to use 935 trillion watt-hours, ranking sixth globally in power use, with AI accounting for 40% of this energy, up from 20% currently. The report also highlights that 90% of AI's power consumption stems from operational requests, not training.
Why it matters
The escalating energy and water demands of AI-driven data centres create significant operational and regulatory risks for CTOs and infrastructure architects. The projected doubling of electricity use by 2030, reaching 935 trillion watt-hours, will strain grid capacity and increase carbon emissions, impacting procurement teams facing rising energy costs. This follows recent research indicating AI compute spending outpaces GDP metrics. The report's focus on operational requests, which account for 90% of AI power use, shifts the priority for platform engineers towards optimising inference efficiency rather than solely training processes.
