Running high-end artificial intelligence hardware can trigger millions in annual utility expenses
Chips & ComputeAI Buzz! · Aug 14

Running high-end artificial intelligence hardware can trigger millions in annual utility expenses

Data from market research firm Liftr Insights shows that operating a setup of 2,000 Nvidia graphics processing units carries immense ongoing electricity bills. In Dallas, running 1,000 H100 and 1,000 A100 chips costs about $2 million in power annually, compared to $2.1 million in Houston and $1.6 million in Austin. Despite high operational costs, major cloud providers continue to expand their use of these processors.

NvidiaLiftr InsightsTab SchadtAWSMicrosoft Azure
Read the original