If the US Has to Build Data Centers, Here’s Where They Should Go (www.wired.com)

A new Nature Communications analysis maps the environmental footprint of the US data‑center buildout driven by the AI boom — fueled by colossal corporate commitments (Meta’s reported $600 billion and OpenAI’s $1.4 trillion plans) — and recommends where future hyperscale facilities should go to minimize harm. Using chip‑demand forecasts, state electricity mixes, and water‑scarcity data, the study models multiple scenarios through 2030 and finds that siting matters: Texas, Montana, Nebraska and South Dakota strike the best balance of cleaner grids and water availability, while traditional hubs like Virginia, California and Arizona risk worsening local water stress or undermining clean‑energy goals. The paper warns that tech firms’ net‑zero promises are unlikely to hold if buildout outpaces decarbonization and model‑efficiency gains. Technically, cooling is a major energy and water sink, and the study’s worst‑case projection—if demand outstrips efficiency and renewable deployment stalls—adds up to ~44 million metric tons CO2e per year, comparable to emissions of medium‑sized countries. Outcomes hinge on variables such as advances in model efficiency, cooling and onsite power (solar, batteries, gas, or nuclear), grid decarbonization, and policy choices that currently sometimes favor fossil fuels. The authors urge proactive siting, technology improvements, and transparency on resource use so the AI infrastructure expansion can scale without overwhelming regional water supplies or jeopardizing climate targets.