Data centres vs. the grid
AI data‑centre growth is colliding with local power systems and community resistance as demand for kilowatts, cooling and transmission rises, forcing fights over permits, noise, and subsidies ( ). Frontier AI firms are responding by renting capacity and moving generation closer to compute — Anthropic, for example, agreed multiyear capacity from CoreWeave to handle Claude demand, highlighting that compute access is now a physical bottleneck (bloomberg.com).
Anthropic agreed on April 10 to rent data-center capacity from CoreWeave so it can keep up with demand for Claude, which is a blunt sign that artificial intelligence companies are now shopping for electricity-heavy buildings the way airlines shop for gates. (bloomberg.com) That shift is happening because an artificial intelligence data center is not just software on a screen; it is a warehouse full of chips, cooling gear, and power equipment that has to be built, permitted, and connected to the grid. (epri.com) The electric power industry’s own researchers said in 2026 that data centers could rise from about 4% to 5% of United States electricity use today to 9% to 17% by 2030, which is why utilities now treat a single new campus like a new factory town arriving all at once. (datacenterknowledge.com) Nevada shows what that looks like on the ground: lawmakers there heard hours of complaints about noise, water use, and power bills as residents in Boulder City fought a proposed project near Hoover Dam. (pbs.org) The fight is no longer confined to one state, because Axios reported on April 5 that lawmakers in 11 states — including Georgia, Maryland, New York, Oklahoma, and Virginia — were weighing bans, moratoriums, or new restrictions on data-center development. (axios.com) Residents are not mainly arguing about chatbots in the abstract; local opposition is aimed at diesel backup generators, constant fan noise, heavy water demand for cooling, giant land footprints, and tax breaks for facilities that often employ far fewer people than a factory of similar size. (kmph.com) Utilities have a second problem behind the neighborhood fights: even when a company is ready to spend billions, transmission lines, substations, and generation projects take years to approve and build, so the bottleneck is increasingly the physical delivery of power rather than the availability of venture capital. (publicpower.org) That is why some artificial intelligence firms are renting existing capacity instead of waiting for new campuses, and why cloud providers are trying to put generation closer to compute so the power plant and the chips arrive together. (reuters.com) Data-center companies say they are not just adding strain: the industry’s trade group told the Associated Press in April that data centers accounted for half of all corporate clean-energy procurement in 2024, even as renewable supply has not expanded fast enough to cover the new load everywhere. (abcnews.com) So the new artificial intelligence race is starting to look less like a contest over better models and more like a contest over who can secure megawatts, permits, transmission access, and community consent before the next surge in demand arrives. (bloomberg.com)
