AI demand is straining power
Cloud and internet giants are ordering vast GPU capacity while energy and datacentre access are starting to limit how much compute can actually be used, forcing some firms to ration capacity and raise prices. Investors point to large Nvidia order books and reports that the AI build‑out is moving scarcity from chips to electricity and infrastructure. (cnbc.com, enterpriseai.economictimes.indiatimes.com)
Artificial intelligence companies can buy more chips than they can plug in, as power hookups, substations and data-center space start to cap how much new computing actually goes live. (cnbc.com) Nvidia said in March it had more than $1 trillion in graphics processing unit orders through 2027, and its stock was up more than 18% over a 10-day winning streak on April 14. Data-center revenue now makes up 88% of Nvidia’s business and was up 75% from a year earlier. (cnbc.com) A graphics processing unit is the chip that does the heavy math for training and running artificial intelligence models, but the chip is only one piece of the build. Each new cluster also needs a building, cooling equipment, transformers, transmission lines and a utility willing to deliver steady power around the clock. (epri.com) The electricity draw is rising fast because artificial intelligence queries use more power than older internet tasks. The Electric Power Research Institute estimated a ChatGPT request at about 2.9 watt-hours, roughly 10 times a traditional Google search at 0.3 watt-hours. (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. In the United States, data centers are on course to account for almost half of electricity-demand growth through 2030. (iea.org) The strain is most visible in places that already host giant server farms. The Electric Power Research Institute said 15 states account for 80% of United States data-center load, and Virginia alone represented about one-quarter of the state’s electric load in 2023. (epri.com) In Virginia, Dominion Energy told regulators in a February 2026 filing that data-center customers had requested nearly 70,000 megawatts of capacity, almost three times the utility’s record peak load of 24,678 megawatts. The utility said it is now receiving about 10 new large-load requests a month, typically totaling 2,000 to 3,000 megawatts. (virginiabusiness.com) Utilities and regulators are responding by changing how they charge these projects. Virginia’s State Corporation Commission said in November 2025 that Dominion’s new GS-5 class will cover customers demanding 25 megawatts or more, including many data centers. (scc.virginia.gov) That shift moves the bottleneck from semiconductors to infrastructure. Nvidia can still show a full order book, but the next constraint is whether utilities, grid operators and builders can deliver enough electricity, land and interconnection capacity to turn those orders into working machines. (cnbc.com)
