Powering intelligence: How AI is redefining data centre design

Artificial intelligence (AI) is rapidly increasing data centre power densities, changing cooling requirements, and making adaptable infrastructure essential. Drawing on my insights from Data Centre World Asia, this article examines what developers should be prioritising as AI demand grows globally. 

Rack densities have increased substantially over the past five years, with some current AI deployments exceeding 100 kW per rack. At these densities, air cooling alone is no longer enough, and the industry is rapidly moving towards direct-to-chip and immersion liquid cooling. In tropical markets like Singapore, Johor and Batam, the challenge extends beyond heat to humidity, water consumption and resilience. Developers can reduce future retrofit risk by making facilities ready for liquid cooling from the outset, even if they do not deploy it immediately, treating cooling, electrical systems and digital controls as one integrated platform. 

As I mentioned during a panel discussion, "the challenge is no longer keeping servers cool, it's how efficiently we can turn electricity into intelligence."

This shift means sustainability and resilience are no longer competing objectives. The biggest risk today is designing for current technology rather than future technology: data centres built now will operate for twenty years, yet AI hardware could change dramatically within the next five. Developers should prioritise:

• Flexible electrical infrastructure
• Readiness for liquid cooling
• Allowance for additional plant capacity
• Modular, scalable design.

These measures can reduce the risk of costly retrofits while improving operational resilience. Where supported by project evidence, explain how they can also reduce operating costs and environmental impact.

"The most sustainable data centre is not the one with the best metrics on day one. It's the one that can successfully adapt for the next twenty years."

Power usage effectiveness (PUE) indicates how efficiently a data centre uses energy, but it does not capture carbon emissions, water consumption or the value of the computing work delivered. Developers should therefore consider a broader set of measures, including:

• Carbon usage effectiveness (CUE)
• Water usage effectiveness (WUE)
• Percentage of energy supplied from renewable sources
• Energy reuse and waste-heat recovery
• AI output per unit of energy.

"PUE tells us how efficiently we run the building. The next generation of metrics will tell us how efficiently we create digital value."

Data centres are increasingly becoming part of regional ecosystems rather than standalone assets. Singapore, Johor and Batam illustrate this complementary model with Singapore providing connectivity, specialist talent and financial capability; Johor offering greater scale and power availability; and Batam supporting potential for future growth and diversification.

"Data centres will no longer compete as individual facilities. They will compete as integrated regional ecosystems."

Underpinning all of this is power, which has become the defining constraint. The first question used to be, ‘Where is the land?’ Today, it is, ‘Where is the power?' AI is significantly increasing electricity demand from data centres, placing additional pressure on utility capacity in constrained markets. Developers must become active participants in the energy ecosystem through renewable energy procurement, cross-border power imports, battery energy storage and grid-support solutions, ensuring reliability and sustainability are achieved together rather than traded off. 

"In the AI era, access to power may become a greater differentiator than access to capital."

Speed and adaptability have also become equally important in commercial strategy and project delivery. Demand for AI capacity is moving faster than traditional delivery cycles, and clients want capacity online sooner than ever before. This is driving standardised designs, prefabrication and modularisation, regional delivery platforms, and earlier collaboration across the supply chain. The real challenge is avoiding stranded assets by designing for future technologies, which means flexibility is becoming as important as speed. 

"The race is no longer about who can build the biggest data centre. It's about who can adapt the fastest."

For developers, the priority is clear: secure a realistic power strategy, plan for changing cooling and computing requirements, and design infrastructure that can scale across regional markets. Early collaboration with utilities, operators, designers and the supply chain will help reduce the risk of stranded or inflexible assets.