AI is forcing us to rethink what data-centre efficiency actually means. As rack densities move toward hundreds of kilowatts, compute, power, cooling, heat rejection and the grid are becoming increasingly coupled. Yet much of the industry still evaluates infrastructure primarily through facility-level metrics such as PUE.

This keynote argues that the next frontier is broader: how much useful intelligence can we create from every unit of electricity, water, carbon, land, capital, infrastructure and time?

Drawing on NUS research in high-density computing, tropical liquid cooling and real-world data-centre testbeds, Prof. PS Lee will examine why AI infrastructure must increasingly be engineered from silicon to grid; why warm-water cooling, higher-voltage power architectures and workload-aware infrastructure are becoming important; and why traditional boundaries between IT, electrical and mechanical engineering are beginning to disappear.

The keynote will also explore why Singapore’s constraints, high temperature and humidity, limited land, careful water management and finite grid capacity, can become a strategic advantage. Rather than competing only on how many megawatts of data-centre capacity Singapore can build, the greater opportunity may be to demonstrate how much useful intelligence can be produced from every megawatt.

The destination is a shift from sustainable data centres to sustainable intelligence.