DARPA Backs Project to Measure AI Energy Use at Every Stage of the Workflow

The artificial intelligence (AI) boom is driving up energy consumption, with data center energy use projected to double by 2030 compared to 2024 levels. While AI proponents argue it will eventually save more energy than it consumes, the sector's current energy and water use is unsustainable. Despite global efforts to increase clean energy production, it cannot keep up with the rapid growth in overall energy demand.

AI is already contributing to the expansion of natural gas-powered plants and could prolong the dominance of fossil fuels in the global energy mix.

Researchers at the University of Michigan Engineering propose a new system to design more efficient AI software, addressing the significant amount of electricity wasted during AI model training. By improving energy tracking capabilities, the system aims to increase efficiency, lower electricity demand, and extend battery life for AI-enabled devices. The project, named E-CALC Energy Calculus for ML Lifecycle Optimization, will be funded by the Defense Advanced Research Projects Agency (DARPA) for two years at $1.83 million.

One of the main challenges in mitigating AI's energy consumption lies in the lack of transparency and insufficient measurement of resource use. The proposed system will enhance the ability to track energy use throughout the entire AI workflow, from data processing and model training to real-world deployment. This technology could be particularly valuable in military and defense settings, where AI algorithms must run on complex field equipment and remote computing systems with strict power and battery limits.

University of Michigan associate professor of computer science and engineering and principal investigator for E-CALC, Mosharaf Chowdhury, emphasizes the project's goal of achieving widespread AI benefits with minimal energy consumption.