Power is the new constraint

A year ago, the bottleneck in artificial intelligence looked like chips. In 2026, the bottleneck looks a lot more like substations, transmission lines, and whoever can deliver firm electricity first. (iea.org, brookings.edu) The reason is simple: a modern artificial intelligence data center is not just a warehouse of servers. It is a factory that turns electricity into computing, hour after hour, and the grid has to be ready before the building is useful. (brookings.edu, epri.com) The International Energy Agency said on April 10, 2025 that electricity demand from data centers worldwide is set to more than double by 2030 to about 945 terawatt-hours, which is slightly more than Japan uses in a year. The same report said electricity demand from artificial intelligence-optimized data centers is projected to more than quadruple by 2030. (iea.org) In the United States, the Department of Energy said data centers used about 176 terawatt-hours in 2023, or 4.4% of total U.S. electricity. It projected that share could rise to 6.7% to 12% by 2028, with total use reaching 325 to 580 terawatt-hours. (energy.gov) That is why the site-selection checklist has changed. Cheap land and fiber links still matter, but utilities now ask a harder question first: can this project actually get power in the year it wants to open. (epri.com, brookings.edu) The Electric Power Research Institute said in its 2024 utility survey that supplying power to data centers is becoming a growing challenge both because there are more of them and because each one is getting larger with cloud and artificial intelligence workloads. That changes a data center from a normal commercial customer into a giant single load that can reshape local planning. (epri.com) Virginia shows what that looks like on the ground. A filing summarized in February 2026 said data center power requests in Dominion Energy territory had reached about 70,000 megawatts, compared with Dominion’s all-time peak load of 24,678 megawatts on January 23, 2025. (virginiabusiness.com) Texas is seeing the same rush from a different angle. Electric Reliability Council of Texas planning material from late 2025 said it was tracking about 226 gigawatts of large-load interconnection requests, and roughly 73% of that total was tied to data centers. (ercot.com) Once those numbers get big enough, regulators stop treating data centers like just another economic-development project. In Maine, lawmakers advanced a bill on April 7, 2026 that would pause new data centers with loads of 20 megawatts or more until November 2027 while the state studies grid and environmental impacts. (usatoday.com, mainepublic.org) Federal regulators are moving too. On December 18, 2025, the Federal Energy Regulatory Commission ordered PJM Interconnection, the grid operator for 13 states and the District of Columbia, to create clearer rules for putting large loads like data centers next to power plants without undermining grid reliability. (ferc.gov) That is why power price, power timing, and power quality are turning into competitive advantages. A region with fast interconnection, spare transmission capacity, and enough generation can win projects even if its land costs more, while a region with cheap land but a five-year wait for grid upgrades can lose them. (brookings.edu, iea.org, gridstrategiesllc.com) The next phase of the artificial intelligence race will not be decided only by who has the best model or the most advanced chip. It will also be decided by who can secure a few hundred megawatts, keep them cheap enough to run every day, and convince regulators and neighbors that the lights will stay on for everyone else. (brookings.edu, dbrs.morningstar.com, iea.org)