Artificial intelligence’s wildly fluctuating power demand is creating technical challenges for data centres, grid operators, and potentially, even energy users at home.
“These loads are extremely dynamic or fluctuating, which causes grid instability and can lead to, if not corrected, potential blackouts or power outages,” Sreemant Roy, a power-quality expert and global offer manager at Schneider Electric in Nashville, Tennessee, told Bloomberg News.
“That has made utility companies globally very worried.”
Traditionally, data centres have drawn vast amounts of power for functions like cloud storage or video streaming, with the power consumption relatively steady and predictable, though they still face pushback from local communities.
But artificial intelligence (AI) processes pose a different problem for data centres and the grids that power them.
Compared to a traditional Google search that would consume about 0.3 watt-hours of electricity, an AI query—such as a ChatGPT- or Google AI-powered response—can use up to 10 times as much, OPAL-RT explains. AI power loads also fluctuate dramatically over short time periods, especially when companies are training new models. Those processes rely on thousands of chips operating in coordinated cycles through varied phases with differing power requirements. Shifting from one phase to another across a range of graphics processing units (GPUs) working in unison causes swings in overall power consumption.
Rapid power fluctuations can also occur during cold starts, planned shutdowns, or sudden load interruptions.
AI power usage can spike by as much as 50% above its design capacity, “so a 1-gigawatt facility may use 1.5 gigawatts for a split second,” Drew Baglino, a former Tesla Inc. executive who started Heron Power Electronics Co., told Bloomberg.
Equipment at the data centres is not made to handle these sudden power changes, the news agency writes, based on interviews with more than three dozen power experts in the United States and Europe. Facilities across the board are seeing damage like gas turbine engine cranks breaking off, or the turbines themselves developing cracks. In some cases, batteries meant to help smooth power flows have needed to be replaced sooner than expected because of the high strain.
The impact can extend beyond data centres. Dramatic fluctuations can affect power quality through “harmonic distortions” to the grid. Grid infrastructure is designed to provide electricity that flows along a 60-hertz wave, and end uses have been designed to operate best when they’re in harmony with that frequency. Bloomberg notes that the expansion of intermittent energy sources like wind and solar is already destabilizing grid functions, but large loads like data centres can distort those harmonics, especially when there are jarring power swings from AI processes. That distortion causes the voltage of electricity being delivered through the grid to change from the 60-hertz. ideal, which is bad for end users.
An earlier Bloomberg analysis in 2024 looked at data from the technology company Whisker Labs, now called Ting Labs, which produces plug-in sensors to prevent electrical fire hazards. It found a correlation between proximity to data centres and worsening harmonics for residential power users, which could lead to everyday appliances failing or burning out, or blackouts.
As noted by the Uptime Institute, correlation does not prove causation. In the original analysis, Whisker Labs did not address a link between harmonic distortions and data centres. But the issue has drawn concern from regulatory bodies and the data industry. One 2026 report prepared for the U.S. Department of Energy calls for improved monitoring.










