New AI Energy Tracking Tool Secures DARPA Funding
The two-year project aims to optimize software-hardware alignment and cut electricity waste during AI model training.
Updated on Oct. 31, 2026 in Artificial Intelligence
Live Poll
Do you believe the rapid energy consumption of artificial intelligence is worth its technological benefits?
University of Michigan researchers have received $1.83 million in DARPA funding to develop an AI energy tracking system called E-CALC. This research-stage project aims to help developers measure electricity consumption across all stages of AI workflows.
Why it matters
Extreme opacity in AI resource usage is straining power infrastructure and increasing costs for the sector. This project seeks to mitigate these issues by identifying and reducing inefficiencies in software and hardware interactions.
The project addresses a 30% electricity waste rate during model training caused by hardware-software misalignment. It will function over a two-year research period to track power usage across data processing, training, and deployment.
The players
DARPA
The Defense Advanced Research Projects Agency, a branch of the U.S. Department of Defense that funds experimental research and emerging technologies.
University of Michigan
A public research university leading the E-CALC project with a focus on engineering and energy efficiency.
BAE Systems
A global defense, aerospace, and security company providing technical partnership for this research project.
University of Illinois Urbana-Champaign
A public research university contributing to the energy tracking project through its engineering departments.
The details
The E-CALC system is designed to provide visibility into energy consumption by analyzing how software tasks map to specific hardware components. Hardware-software misalignment occurs when code processes are inefficiently scheduled for the underlying compute architecture, leading to cycles where power is consumed without proportional computational output. By providing granular data across the AI lifecycle, the researchers intend to give developers the metrics needed to optimize software design for energy efficiency.
Timeline
2024: Data center energy consumption baseline year of 420 terawatt-hours.
- 2026-10-31
DARPA announces the $1.83 million research project grant.
2030: Projected year for data center energy consumption to double from 2024 levels.
The Tech Race
This project follows a pattern of government-backed research aimed at quantifying the environmental footprint of large-scale AI infrastructure. It directly counters the current trajectory of energy demand, which is projected to double by 2030.
This is a two-year research project and is not currently a product available for commercial software development or deployment. Developers should monitor the project's progress for future open-source tools that could assist in identifying energy inefficiencies within AI training pipelines.
The takeaway
The E-CALC project highlights the industry's pivot toward energy transparency as power infrastructure constraints become a bottleneck for deployment. Stakeholders should track progress on the researchers' development milestones over the next 24 months for potential efficiency improvements in AI training.
Further reading
For broader trends in research, visit Artificial Intelligence.
Live Poll
Do you believe the rapid energy consumption of artificial intelligence is worth its technological benefits?
