Overview
- The latest NBER working paper by economist Nicholas Muller estimates that pollution from U.S. data centers imposed about $25 billion in damages last year, including $3.7 billion tied to AI workloads, by converting power-plant emissions into health and climate costs using PM2.5 risk and the social cost of carbon.
- Harms are uneven across the map, with Virginia and Texas responsible for about 30% of the health costs, and Northern Virginia studies and residents pointing to tens of millions in annual damages, persistent noise, and steep local electric-bill spikes near dense server clusters.
- Energy demand is surging as the IEA reports data centers used roughly 415 TWh in 2025 and could reach about 945 TWh by 2030, while experts warn that a large facility can consume up to five million gallons of water per day to cool high‑performance chips.
- Construction is still accelerating as analysts track 76 U.S. projects starting in 2026, even as debate deepens with an AEI summary calling the damages modest relative to sector output and more than 230 environmental groups urging a national moratorium.
- Proposed fixes include cleaner power such as on‑site solar with batteries, closed‑loop water recycling and alternative sourcing, greater efficiency in AI models, and wider use of small language models that can run on devices to cut compute and grid load.