According to a Bloomberg investigation released on Thursday based on interviews with more than three dozen power experts in the US and Europe, the fast power swings caused by artificial intelligence workloads are damaging vital infrastructure at data centers and prematurely destroying batteries, generators, and cooling systems in ways the industry did not anticipate.
Hardware Broken By Millisecond Shocks
As hundreds of thousands of graphics processing units synchronize during model training, AI facilities experience power demand surges and collapses within milliseconds, in contrast to traditional data centers, which have relatively stable electricity use. A gigawatt data center, according to Mainspring Energy founder Shannon Miller, is like a city the size of Boston, “half of which can flicker on and off every few seconds.”
An individual acquainted with the situation claims that gas-fired turbines at xAI’s Colossus computer center in Memphis, Tennessee, have cracked. The batteries used to cushion the swings had to be replaced within weeks or months. According to Andrew Cunningham, CEO of GeoPura Ltd., similar turbine cracking has happened in smaller data centers in the UK.
AI power consumption can increase by up to 50% over design capacity, according to Drew Baglino, a former Tesla executive who created Heron Power Electronics Co., “so a 1 gigawatt facility may use 1.5 gigawatts for a split second.” His business is creating power management hardware for Nvidia’s upcoming servers, which are scheduled for release in 2027.
Income At Risk
The cost goes much beyond simply replacing the equipment. According to Jason Hoffman, chief strategy officer at data-center operator Switch, “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.”
The cost of downtime varies from thousands to hundreds of thousands of dollars per minute. Financial models imply year-round continuous operation, although some facilities are experiencing uptime closer to 80%. This might affect investors in 12 to 24 months if it’s not fixed, a data center financing expert cautioned.
Issues With Grid Stability
Large data centers must respond to a rare level-three notice from the North American Electric Reliability Corp. earlier this year by August 3 in order to mitigate imminent threats to grid stability. For more than 33 gigawatts of active US data centers, NERC discovered that over 75% of load models were “insufficient to represent data-center dynamic behavior.”
Sreemant Roy of Schneider Electric cautioned that the variations “can lead to, if not corrected, potential blackouts or power outages,” adding: “That has made utility companies globally very worried.”