Research project to boost sustainable AI
The environmental impact of AI is significant and growing. The researchers behind the new research project, DEEP, have received a 5,6 million DKK grant for a project focused on reducing the cost and the carbon footprint of AI solutions.
Written 10 February, 2026 12:54 by Jari Kickbusch
When the United Nations Environment Assembly convened in December 2025, the mounting environmental impact of artificial intelligence was a key topic for discussion. Currently, data centres consume about 1.5% of global electricity, of which AI is responsible for about 15%. However, AI models are getting larger, the applications broader, and adoption more widespread. The International Energy Agency projects that energy demand will double by 2030 as a result of AI.
AI consists of software and hardware. The software processes include data collection and preparation, model development, training, validation, deployment, inference, maintenance and retirement. The hardware includes the production of computer chips, such as graphical processing units (GPUs) essential to training and inference, and the construction and operation of data centres. The new research project, DEEP, launched by the University of Applied Sciences and Arts of Western Switzerland (HES-SO) and the IT University Of Copenhagen focuses on the hardware usage by AI.
DEEP
DEEP, short for Deep Learning Resource Efficient GPU Orchestrator, aims at improving GPU utilization. GPUs are processors designed to speed up computer graphics and image processing on various devices. This makes them an essential enabler of machine learning (ML) and artificial intelligence (AI).
A recent study shows that many GPU clusters often operate below 50% capacity. This waste of hardware resources is exacerbated by the high price of GPUs, while contributing to an unsustainable carbon footprint of AI systems.
“Based on experience, we know that reaching above 80% GPU utilisation is not good for energy efficiency and request latency. But staying around 50% or less is a waste of hardware resources. The long-term aim of this project is to establish a hardware resource manager for more sustainable deep learning," says Associate Professor at the IT University of Copenhagen, Pınar Tözün.
The grant
Pınar Tözün collaborates with Associate Professor at School of Engineering and Management Vaud HEIG-VD (part of HES-SO), Pamela Delgado, who is the Principal Investigator of DEEP. Together, the two researchers have secured a DKK 5.6 million grant from the Swiss National Science Foundation to develop novel resource managers and schedulers that account for the resource needs of deep learning workloads and of the characteristics of modern hardware.
“Through proper hardware and workload characterisation, we can improve resource management and thereby reduce the hardware needs and carbon footprint of deep learning," says Pamela Delgado.
The project runs from August 1, 2025, to July 31, 2029.
Learn more about the project
Jari Kickbusch, phone 7218 5304, email jark@itu.dk