Can AI keep its cool?

01

How big is the AI infrastructure opportunity?​

Around 60 per cent of today’s live data centre capacity was added in the past five years. Industry forecasts suggest the world may add roughly the same amount again by 2030, with Asia and other emerging markets becoming increasingly important growth hubs.

The scale of required investment is significant. Some estimates suggest that enabling the AI-driven build-out may require up to USD9 trillion across IT hardware, physical infrastructure, power generation, grid capacity, cooling systems and enabling supply chains. Enabling data centre growth during the next three years may require external, debt and equity funding of more than USD1 trillion.

02

Could AI accelerate energy transition investment?

The scale of AI infrastructure investment could create a powerful new source of demand for clean power, grids and energy infrastructure. Amazon, Google, Meta and Microsoft have already contracted more than 140GW of renewable power, including almost 35GW in 2025.

Where operators make credible, long-term commitments to efficient infrastructure, additional low-carbon generation and system flexibility, that demand can improve investment visibility and help support the development of new energy capacity.

If AI is used to optimise energy consumption in buildings, industry, transport and power systems, the resulting efficiency gains in those sectors may also perhaps offset a significant part of the additional electricity demand from data centres.

Data centre-related power consumption is low relative to other sectors on a global level, however, near-term local challenges can be significant where rapid growth is already testing power, grid and water systems.

03

Can data centre growth limit added power demand?

Forecasts suggest global data centre electricity demand could reach around 1,100TWh by 2030, roughly twice 2025 levels, with AI potentially accounting for almost half of that demand. 

The challenge is not only how to meet this rising demand, but how to support digital growth without placing unnecessary pressure on global and local electricity systems, energy costs or decarbonisation efforts. 

The report identifies several levers that could help leading data centres reposition themselves from large, largely passive electricity users into more efficient and flexible participants in the energy system. 

These solutions include more efficient hardware and software, higher utilisation, heat reuse, renewable power, 800V DC architecture, battery storage and microgrids. Next-generation AI infrastructure could also dramatically reduce cooling requirements.

For example, new liquid-cooled architectures can operate efficiently at temperatures of around 45°C, substantially reducing the need for separate mechanical cooling and the associated power and water use. 

04

Will low-water data centres become the norm?

Water use remains an important local consideration, particularly in water-stressed regions, but there is substantial variation according to cooling technology, location and power source.

Several levers could help moderate water intensity, including more efficient hardware and software, higher utilisation, heat reuse, renewable power, 800VDC architecture, battery storage and microgrids.

Some leading operators are already improving water efficiency substantially: reported Water Use Effectiveness (WUE) at Amazon, Meta and Microsoft has fallen by 43–52 per cent since 2020. Liquid and dry cooling, higher operating temperatures, recycled water, renewable power and replenishment programmes could reduce future water intensity and may make low to near-zero water intensity a feature of future data centres.

05

What does good look like for data centres?

Data centre sustainability cannot be judged through a single metric or portfolio-wide average. Power and water impacts vary according to location, grid mix, climate, technology and utilisation, making facility-level assessment important.

Leading facilities optimise energy, carbon and water together, taking account of local grid conditions, climate and water availability, while reporting both absolute impacts and intensity metrics.

Yet the report’s assessment finds that while more than 60 per cent of the operators reviewed disclose overall energy or water use, fewer than 10 per cent provide facility-level data. More consistent, facility-level disclosure could help make data centres comparable, strengthen planning and investment decisions, and direct capital towards better environmental performance.