AI’s growing appetite for water and power draws sharper scrutiny
As data centre construction accelerates to keep pace with AI demand, electricity and water consumption have become a bigger part of the public conversation around AI's costs — pushing operators toward efficiency measures like waste-heat recovery and more power-efficient chip designs.
As AI data centre construction accelerates worldwide to keep pace with demand for training and inference capacity, the electricity and water these facilities consume has become a more prominent part of the broader public conversation about AI’s real-world costs, alongside longer-running debates over jobs, safety and misinformation.
New hardware generations are being marketed partly on efficiency gains rather than raw performance alone — Nvidia’s new Rubin platform, for instance, is being pitched partly on a claimed five-times improvement in power efficiency for its networking systems, not just faster chips. Separately, some operators are pursuing more creative offsets: Helsinki’s district-heating partnership with a local data centre operator, which redirects waste heat into the city’s heating network rather than simply venting it, is one of the more concrete examples of turning an operating cost into a community benefit.
Whether efficiency gains and offset projects like these can keep pace with the sheer scale of new data centre construction remains genuinely uncertain — global AI compute demand has grown so quickly that even meaningful per-chip efficiency improvements may simply be outpaced by the number of new chips being deployed.
