MIT’s Christina Delimitrou Uses AI to Lift 15% Data Center Utilization
Updated
Updated · MIT News · Oct 8
MIT’s Christina Delimitrou Uses AI to Lift 15% Data Center Utilization
3 articles · Updated · MIT News · Oct 8
Summary
MIT associate professor Christina Delimitrou is using machine learning to redesign cloud systems, manage shared hardware and streamline server architectures so data centers deliver more computing from existing equipment.
About 15% utilization in many large computing systems, which Delimitrou identified in earlier research, means operators waste power and push grids toward more fossil-fuel use as demand grows.
Tools from her group target both efficiency and reliability: Seer uses deep learning to predict web-application failures, while Ditto clones proprietary systems so researchers can test fixes that work beyond the lab.
At MIT since 2022, Delimitrou is also adding explainability to AI tools, arguing that better-audited models could curb the need for new data centers as the sector’s power demand keeps rising.