Society, Environment & Big Picture
Energy & Water Consumption
Data centers need electricity for computing and water for cooling; the AI boom turns this into location policy — new power plants, strained grids, dry regions with thirsty neighbors. A single query weighs little, billions a day weigh heavily. An honest balance includes both: the real footprint and the real savings where AI replaces something else.
An example
The circulating figures contradict each other by orders of magnitude, and the reason is rarely intent: some count only the computation, others the cooling, others still the manufacture of the hardware; some refer to a large model, others to a small one. What holds up is the direction, not the number: a query to a large model uses clearly more than a web search, and the total consumption of data centers is measurably growing.
A closer look
Consumption occurs once during training and continuously in operation, water mainly in cooling. Reliable figures are rare because vendors draw boundaries differently: with or without hardware manufacture, with which power mix.
Why it matters
Three things can be controlled in practice, and, happily, they coincide with cost optimization: use the smallest model that solves the task; shorten the context; cache recurring queries. Whoever matches model choice to the task rather than to the brochure lowers consumption and the bill at the same time. For reporting obligations the usual caution applies: demand per-query figures from the vendor rather than group-wide annual balances, and write down what you do not know.
Commonly misunderstood
A single query is little, the sum is not. It becomes meaningfully comparable only per completed task, measured against the process it replaces.