AI’s insatiable thirst for electricity is expected to surge in the coming years, potentially leading to power shortages for data centers. New servers last year demanded 195 terawatt-hours of electricity, according to a new report from Gartner. That’s as much as 18 million households use in a year. But by 2027, new servers could command 500 terawatt-hours, or 46 million households’ worth.
That’s over and above the juice used by existing data centers, which already consumed 349 terawatt hours in 2022, according to a Goldman Sachs estimate. Without additional sources of carbon-free power, pollution from AI training and use could skyrocket. Sam Altman’s big bet on fusion power — over $375 million — is starting to make more sense.
This article was originally published by Techcrunch.com. Read the original article here.