Behind every artificial intelligence application lies an energy bill that remains invisible to the end user. As data centers multiply and strain electrical grids, Christina Delimitrou, an Associate Professor at MIT, is focusing her research on a striking contradiction: a vast portion of available computing power remains unutilized, even as systems continue to consume significant energy.
The Challenge of Sustainable AI
According to Delimitrou, the sustainable development of AI is one of the most critical technological challenges of the coming years. Her research group at the Massachusetts Institute of Technology (MIT) leverages machine learning to improve the efficiency, security, and reliability of large-scale computing systems. The goal is to support more applications with fewer resources, thereby reducing the need for constant infrastructure expansion.
The 15% Paradox
The Greek scientist's research, which began during her doctoral studies at Stanford University in collaboration with Professor Christos Kozyrakis, revealed that many large-scale computing systems operated at only 15% of their capacity. Much of this inefficiency is rooted in software management. To address this, her team developed tools like Seer, which uses deep learning to predict and prevent issues in online services—such as video conferencing and streaming—before they manifest as delays or outages.
From Athens to MIT
Christina Delimitrou’s journey began at the National Technical University of Athens (NTUA). Following her studies at Stanford and an academic tenure at Cornell University, she joined the MIT faculty in 2022, where she recently received tenure. She emphasizes the high quality of scientific talent in Greece, noting that the work produced in Greek universities deserves greater international recognition.
