AI rollout bumps into power and water limits

A data center used to be a real-estate story. In April 2026, it is also a utility story, because new artificial intelligence campuses can be approved on paper and still stall if the local grid cannot feed them or the local water system cannot cool them. (therealdeal.com) That is why more than a dozen investors are pressing Amazon, Microsoft, and Alphabet’s Google before annual meetings to publish harder numbers on electricity use, water use, and local community impact instead of broad climate promises. Reuters reported those investors as environmental scrutiny forced the companies to defend how fast they are expanding computing capacity. (journalrecord.com) The water fight is not abstract. Reuters said North American data centers used nearly 1 trillion liters of water in 2025, mostly for cooling, at the same time that many of the fastest-growing data-center markets were already dealing with drought stress and population growth. (journalrecord.com) The power problem is even less flexible. A graphics processing unit cluster can be bought and installed in months, but a new transmission line or substation often takes years, so the bottleneck has shifted from chips inside the building to wires outside it. (therealdeal.com) That mismatch is already changing projects on the ground. The Real Deal reported that Amazon, Microsoft, and Google have each scrapped multibillion-dollar United States data-center developments after community opposition tied to power, water, pollution, or land use. (therealdeal.com) India is now treating the same issue as a national planning problem, not just a property boom. Mint reported on April 8, 2026 that the Central Electricity Authority, India’s top power-planning body, has gone back to redraw demand maps because data centers are clustering in a few states and could distort load patterns across the national grid. (magzter.com) Those clusters are large enough to move national numbers. Outlook Business reported that India’s installed data-center capacity is expected to rise from about 1.2 gigawatts in 2025 to almost 10 gigawatts by 2030, with electricity use rising from roughly 10 to 15 terawatt-hours to about 40 to 45 terawatt-hours. (outlookbusiness.com) India’s federal government is also using an even higher planning figure. In March 2026, Minister of State Jitin Prasada told Parliament that electricity demand from data centers could reach 13.56 gigawatts by 2031-32, which is why planners are now asking states and power distributors to account for server-farm loads in resource plans. (economictimes.indiatimes.com, (communicationstoday.co.in) The old way to judge an artificial intelligence buildout was to ask who had the most chips and the most capital. The new question is which company can secure 24-hour electricity, enough cooling water, and local political consent before the hardware arrives. (therealdeal.com, magzter.com) That is why this story keeps showing up in shareholder filings, zoning fights, and national power forecasts at the same time. The artificial intelligence race is no longer just about faster models inside server racks; it is also about transformers, pipelines, reservoirs, and whether neighbors will let the building plug in. (journalrecord.com, therealdeal.com, magzter.com)