The Defense Advanced Research Projects Agency is funding a $1.83 million project led by the University of Michigan to reduce the electricity consumption of artificial intelligence. Led through the E-CALC program with the University of Illinois and BAE Systems, it aims to track and optimize power use throughout the model lifecycle, from data preparation to deployment. Its research target is to cut consumption by as much as 50%, while a separate $1 million award to Stony Brook University and Ohio State University seeks to forecast energy use without sacrificing accuracy. Data centers could consume more than 840 terawatt-hours of electricity by 2030, compared with about 420 in 2024, and mismatches between software and hardware could waste up to 30% of energy during training.
Source: Read the original article


