Online Reporter
Artificial intelligence’s immense electricity consumption is well known, but less recognised is how rapid fluctuations in data centres’ power demand can damage critical equipment.
Batteries, generators, cooling units and other systems at AI computing facilities are malfunctioning or wearing out prematurely due to repeated power swings.
These technical problems threaten added costs, reliability issues and lost revenue — with even minutes of downtime hitting developers’ bottom lines.
The issues come at a time when investors are already concerned about hyperscalers’ massive spending and whether these facilities may depreciate faster than expected. They also risk destabilising power grids already struggling to maintain stability.
“AI does create very unusual power demand,” said Amber Villegas-Williamson, principal consultant at the Uptime Institute in the UK. “It’s like over-revving your car wears out the engine faster than keeping a constant speed.”
Data centres have long consumed large amounts of electricity, but AI facilities are different: their demand is far larger and swings much more dramatically.
Power increments equivalent to the consumption of factories, towns or cities can appear and disappear within seconds, creating repeated shocks.
A gigawatt data centre is comparable to a city the size of Boston, half of which can flicker on and off every few seconds, said Shannon Miller, founder and president of Mainspring Energy.
Some AI campuses planned in Texas and the US Midwest are more than five times larger, consuming nearly as much power as New York City on average.
AI data centres face particular strain when training new models — a process that mobilises all graphics processing units in unison. Hundreds of thousands of GPUs can power up and down on a millisecond basis, like a swarm of bees changing direction.
AI can see power usage spike as much as 50 percent above design capacity, “so a 1-gigawatt facility may use 1.5 gigawatts for a split second,” said Drew Baglino, founder of Heron Power Electronics Co.
Most equipment isn’t designed for such swings. Jon Parrella, CEO of Terraflow Energy, likens it to driving a Ferrari and shifting straight from sixth gear to first. “You can’t swing that fast,” he said.
Based on interviews with more than three dozen power experts in the US and Europe, the physical stresses on AI facilities are evident. Cranks on natural gas combustion engines have broken off; at xAI’s Colossus facility in Memphis, gas-fired turbines developed cracks. Batteries installed to smooth power swings have sometimes needed replacing within months or even weeks.
These issues are causing delays or curtailing operations at some facilities. A planned 2.67-gigawatt AI campus in West Texas has pushed power delivery from 2027 to 2028 to allow extra engineering time, said Chris James, CEO of Joulent.
“The financial consequence is not primarily replacing a pump or a breaker — it’s the value of that expensive compute capacity not generating revenue because it’s offline,” said Jason Hoffman, chief strategy officer at Switch.
While data centres are built on the assumption of 365-day operation, some are seeing uptime closer to 80 percent, which could hit investors in certain projects within 12 to 24 months, said a person involved in financing.
AI data centres also risk destabilising the wider grid. “These loads are extremely dynamic or fluctuating, which causes grid instability and can lead to, if not corrected, potential blackouts,” said Sreemant Roy, a power-quality expert at Schneider Electric.
The North American Electric Reliability Corp has repeatedly warned that data centres are among the greatest risks to grid stability. It evaluated more than 33 gigawatts of operational data centres in the US and found about three-quarters of their load models insufficient.
The industry is working on solutions. Nvidia is collaborating more closely with power experts on chip development and data centre deployment.
The National Laboratory of the Rockies has set up a test-bed for the US Department of Energy to help developers integrate AI safely onto the grid.
“We have the opportunity right now to get it right,” said Martha Symko-Davies, the lab’s program manager. — Bloomberg.
