Key facts
- AI energy consumption is projected to more than double by 2030.
- Up to 30% of electricity can be wasted during AI model training due to software-hardware misalignments.
- The E-CALC project aims to enhance energy tracking for AI workflows.
- DARPA is funding the project for two years with $1.83 million.
The rapid integration of artificial intelligence is driving a significant increase in global energy demand, with data center consumption expected to more than double by 2030. This surge in energy use, even with advancements in clean energy, is straining power grids and potentially hindering climate goals. Researchers at the University of Michigan Engineering, in collaboration with the University of Illinois Urbana-Champaign and BAE Systems, are developing a new software system to improve AI energy efficiency and tracking. The project, funded by DARPA, aims to address the current opacity in AI's resource usage by enabling developers to estimate energy consumption across all stages of AI workflows.