Presentation @ ACM E-Energy 2025

Last week, I had the honor of presenting our latest research paper — "Sustainable, Situation-Aware Multi-Objective Spatial Load Scheduling for Data Centers" — at the Energy Efficient Data Centers Workshop, co-located with the 16th ACM E-Energy Conference in Rotterdam.

Together with my co-author, Prof. Marco Pruckner we investigate how to make the operation of LLM inference in data centers more sustainable by jointly optimizing for cost efficiency, carbon emissions, and water consumption.

Our contributions include:

• A new multi-objective load shifting model that incorporates a spatial and temporal drought risk indicator to weight water consumption more accurately
• A novel framework, IntensityLib, which automatically calculates water intensity factors for the European electricity grid and includes a flow-tracing algorithm to reflect cross-border electricity flows
• A case study that evaluates how to spatially shift LLM inference loads across ten European countries to reduce environmental impact

A key insight from our study is that optimizing for water scarcity rather than only for total water consumption leads to very different results – with the potential to avoid further stress on already vulnerable regions.

Attending ACM E-Energy was a fantastic experience: many new contacts in the fields of energy systems and optimization, stimulating discussions, and inspiring challenges. I'm looking forward to continuing research in this exciting domain.

If you're interested in the paper or in IntensityLib, feel free to get in touch – I’m happy to share and discuss.