Artificial intelligence has an appetite for buildings. Data centres — the warehouses that store and process everything from search results to model training — are going up faster than almost any other property type. Cities compete for them, because they bring investment and jobs, and because they look like the future.

A new study, however, suggests that this future may arrive thirsty. About a quarter of the world’s data centres stand in regions already under water stress, where demand from farms, factories and households is close to what rivers and underground supplies can provide.

The reason is cooling. The machines inside produce heat without stopping, and the cheapest way to remove it is to let water pass into the air as steam. That water is not returned; it leaves the site for good. A single large campus can use millions of litres a day in summer, and the study expects worldwide demand for cooling water to double by 2030 as AI use grows.

For dry regions the arithmetic is uncomfortable. In several areas where new centres have been approved, farmers have already been told to cut back. “The question is not whether computing matters,” one of the authors said. “It is whether this is the right place to put it.”

Industry groups say their share of total water use remains small, and that newer designs recycle most of the water ina closed loop. Some operators now publish water-efficiency figures, and a few are turning to dry cooling, which uses air rather than water.

Both sides agree on one point: the difficulty lies less in computing than in siting and design. A data centre in a wet, windy country may cost the local supply almost nothing; the same building in a drought-prone basin can matter a great deal.

原创编写 版权所有 侵权必究 每日更新 个性化阅读 英语飙升