The Defense Advanced Research Projects Agency (DARPA) has allocated $1.83 million to the E-CALC project, led by the University of Michigan. The initiative aims to develop a mathematical toolkit for tracking and optimizing energy consumption throughout the AI lifecycle, including data processing, training, and deployment. Partners in this effort include the University of Illinois Urbana-Champaign and BAE Systems, with the award announced in late October 2026.
Researchers believe the E-CALC framework could potentially reduce AI energy usage by up to 50%. The project focuses on addressing inefficiencies where software and hardware mismatches can waste up to 30% of energy during training. This research is particularly relevant for military applications, where drones and field equipment operate under power constraints and cannot rely on hyperscale data centers.
Additionally, DARPA awarded $1 million on October 20, 2026, to a team from Stony Brook University and Ohio State University under the ML2P program. This initiative is developing tools to forecast the energy consumption of machine learning models without compromising their accuracy. These combined investments reflect a broader strategy to integrate energy awareness into AI development.
The urgency for these initiatives is driven by projections that data center electricity consumption will rise from approximately 420 TWh in 2024 to over 840 TWh by 2030. While the total funding of $2.83 million is small compared to commercial data center investments, DARPA has a history of funding foundational technologies that eventually transition into widespread commercial use.
