Energy Management System for AI Datacentres
AI datacentres create a new class of operational problem: dense compute demand, shifting workloads, thermal constraints, and energy costs that need to be understood as part of one control system. We are interested in software and modelling approaches that make these environments more observable, more efficient, and more strategically manageable.
Problem spacePower, thermal load, workload scheduling, and infrastructure visibility
AimBetter control and decision-making in AI-heavy compute environments
MethodMonitoring, modelling, orchestration, and system design
Research Questions
- How should energy signals inform orchestration decisions for large AI workloads?
- What kind of visibility is needed to balance power budgets, thermal constraints, and throughput?
- Which control loops belong at the hardware layer, and which belong at the software or platform layer?
- How can operators reason about efficiency without losing sight of service quality and research goals?
Why This Direction Matters
As AI infrastructure scales, energy management stops being a secondary concern. It becomes part of the core architecture. We are interested in systems that treat energy as an operational variable that can be measured, modelled, and acted upon.