AI model training tied to 50% power spikes damaging equipment
Technology
AI data centers are hitting unexpected power spikes, sometimes 50% over what they were built for, especially when training new models.
This is frying batteries, turbines, and cooling systems, leading to project delays and higher downtime costs for operators.
AI demand swings raise blackout risk
Some AI centers now use as much electricity as a city like Boston, with usage jumping up and down in seconds.
As Shannon Miller, founder and president of Mainspring Energy Inc. put it, these swings have already pushed back big projects (like a huge Texas facility delayed to 2028) and made blackouts more likely since most grids were not designed for this kind of demand.
While some fixes are being tried out, the industry still has a long way to go before things get stable.