The AI boom is colliding with America’s aging power grid

PG&E entered 2026 expecting a year’s worth of new electricity demand. Barely two months later, nearly all of it was already spoken for. Interconnection requests were piling up faster than planners had expected, overwhelming a regulatory system built for an era when electricity demand barely moved. That world is gone. Load growth that historically ran below 1% annually hit 4% at some grid operators last year, according to a report by the Lawrence Berkeley National Laboratory . Bain and Company projects that AI data centers alone could consume up to 9% of total U.S. electricity by 2030, adding more than 150 terawatt-hours of demand that the current grid was never really built to handle. A third of that new demand is concentrated in Virginia, Texas, and California, according to Pew Research Center , putting extraordinary pressure on regional systems already straining to keep up. AI may dominate the headlines, but it is only part of the story. EVs , new factories, and industries shifting off diesel and gas are all pulling from the same aging grid at the same time. At Southwest Power Pool, which oversees electricity across 17 states, officials compared last year’s surge in demand to two large nuclear plants suddenly appearing on a grid with roughly 56 gigawatts of capacity. By late 2024, more than 2,600 gigawatts of proposed generation and storage projects were waiting to connect to the grid, according to Lawrence Berkeley National Laboratory, more than twice the country’s entire installed capacity. David Sawaya, PG&E’s director of rate-reducing load growth , says virtually every utility he knows has seen its interconnection queue swell by 50% to 150% in just two years. “The process does not move at the speed of business,” he tells Fast Company . “And right now, the business is moving very fast.” A decade behind and racing to catch up The utilities trying to manage this transformation are not exactly starting from a position of strength. Carlos Elena-Lenz, who leads digital enablement at Hitachi Energy, recalls a colleague who joined the company after decades in oil and gas offering a blunt assessment: Utilities globally are about a decade behind that industry in adopting AI. The culture inside many utilities has historically been built around what Elena-Lenz calls a “break-fix” model, waiting for equipment to fail rather than predicting and preventing failure. Many utilities, he says, still cannot tell you with GPS-level precision where their own assets sit on the grid. Then there is the data problem. Yuriy Yuzifovich, CTO of AI at GlobalLogic , a Hitachi subsidiary that builds digital systems for energy companies, says utilities often want AI systems their infrastructure simply cannot support. Equipment across the grid is already generating useful data, but much of it never reaches the enterprise systems designed to use it. And even when the data does make it through, utilities often lack the people or workflows needed to act on it. “Intelligence is just the tip of the iceberg,” he says. The much larger challenge lies beneath it: rebuilding data infrastructure, changing internal processes, and retraining workforces to actually use these systems effectively. Each new AI workload arriving on the grid creates pressure across three dimensions simultaneously: more power, more cooling, and more filtration. For an industry already a decade behind, that’s a significant amount of catching up to do. Some progress is visible, however. Utilities that spent years asking what they should avoid are now asking what they can do. GlobalLogic is deploying AI systems inside utility operations that continuously predict where grid stress will appear hours before it materializes. Hitachi Digital has begun using synthetic data to replicate entire power-grid networks for testing, allowing models to be stress-tested against simulated conditions before touching live infrastructure. THE REGULATORY WALL As is often the case with technology, the tools to modernize the grid are moving faster than the rules written to govern them. In conversation, Sawaya is careful to describe the California regulatory process as cumbersome rather than broken: a deliberate, stakeholder-driven structure built for a different era. Bringing a new proposal to the California Public Utilities Commission, gathering testimony, building a record, and securing a decision can take two years. For a data center developer operating on a business timeline, that can kill a project before it properly begins. Richard Schomberg, special envoy for smart electrification at the International Electrotechnical Commission (IEC), puts an even harder number on the broader bottleneck: Interconnection timelines across the U.S. can stretch to seven years, constrained by transmission capacity, substation readiness, and a queue system designed for far lower volumes. Jesse Jenkins, who leads the ZERO Lab at Princeton University, argues that the industry may be thinking about the problem in