WhichAI / AI Debates / Energy, water and data centers

Energy, water and data centers

Is AI's footprint a real problem or a manageable cost? · updated 2026-07-30

Training and running AI needs vast data centers, and their electricity and water use has become the loudest environmental debate in tech. The growth is real and fast; so are the efficiency responses. Both things are true at once, which is why the argument does not end.

415 → ~945 TWhglobal data-center electricity, 2024 to 2030 (IEA projection)
4.2-6.6 bn m³projected global AI water withdrawal by 2027 (UC Riverside study)
~5th largest 'country'where data centers would rank by power use on aggressive 2026 estimates

The concern

The concern: demand is doubling in six years, grids and local water basins feel it first, and efficiency gains keep getting eaten by scale (Jevons paradox). Communities near new builds report higher bills and water stress.

The counter-view

The counter-view: data centers are still a single-digit share of global electricity; the industry is moving to closed-loop zero-water cooling (Microsoft) and 120% water-replenish pledges (Google), and AI itself optimizes grids and materials science.

Where it stands (July 2026): disclosure is improving, regulation is starting to ask for per-site numbers, and the honest answer is that growth is currently outpacing efficiency.

Sources: IEA, Energy and AI · Making AI Less Thirsty (UC Riverside, arXiv) · Consumer Reports on data centers · Brookings on AI energy policy