Almost every recent story about artificial intelligence and electricity has been a story about strain: data centers draining the grid, bills climbing, communities pushing back. A report from E&E News points to a quieter and more hopeful possibility, and to the unglamorous reason it keeps failing to happen.
The same technology straining the grid can also help run it. Pacific Gas & Electric, the California utility still recovering from its role in the 2018 Camp Fire, now operates a round-the-clock monitoring program: 1,600 weather stations using AI to forecast fire-risk conditions, 600 infrared cameras trained to spot the faint flicker of a new wildfire, and readings pulled from 5.5 million meters to flag compromised equipment. Other utilities use AI forecasting to position repair crews ahead of storms, sensors to extend the life of transformers and substations, and software to juggle a diverse mix of energy sources. The GridWise Alliance counts eight distinct grid uses for AI. Done well, these "grid-enhancing technologies" squeeze more capacity out of wires that already exist, capacity that would otherwise demand slow, expensive new construction.
The obstacle is not the software. It is who pays for it. Utilities recover their costs by persuading state regulators to fold them into customer bills, and regulators, under intense pressure to hold those bills down, have repeatedly refused. North Carolina rejected Duke Energy's request to recover the cost of voltage-management software in 2018. Hawaii regulators declined to let a utility pass on the cost of microgrid integration software. Illinois regulators in 2020 turned down cloud computing used to manage renewable resources. PG&E says it can demonstrate that a $45 million wildfire-technology program will ultimately lower costs, yet it expects to deploy it slowly anyway.
The irony is sharp. Refusing the upfront expense does not avoid spending; it locks in the older and more expensive model of building new generation and raising rates. As one industry advocate put it, the cheaper path keeps getting kept "out of the hands" of the utilities that need it most. "Where are the incentives in the system to do things differently?" asked Kim Getgen of InnovationForce, which connects technology providers with utilities.
The report also carries a useful correction to the public mood. Households overwhelmingly blame data centers for rising bills, but a widely cited study from Lawrence Berkeley National Laboratory and the Brattle Group found that from 2019 through 2024 the main driver of price increases was the dull work of refurbishing and replacing aging transmission and distribution infrastructure, the poles, towers and wires, along with repairs from hurricanes and wildfires. AI-driven demand is real and growing fast, but it is being layered on top of a grid that was already failing and already costly to fix.
There are signs of movement. The House held a hearing in April on bills to encourage AI on the grid, the Energy Department has opened a $1.9 billion grant program, and federal regulators plan a conference on the subject in July. Maine and Virginia now require regulators to weigh these "non-wires alternatives" before approving costly new construction. The pressure to do so is only mounting: JPMorgan projects U.S. utilities will spend $1 trillion upgrading transmission and distribution over the next decade, and ratepayers will fund all of it. The open question is whether smarter software trims that bill, or whether regulatory caution keeps the savings permanently just out of reach.