A data centre cannot wait for the weather to improve. Whether temperatures soar above 45°C or heavy rain sweeps across the UAE, the computers supporting digital services must continue operating, with electricity and cooling systems running around the clock.
As the UAE builds increasingly powerful artificial intelligence infrastructure, operators are preparing for more than the demands of computing. Extreme heat, flooding, water shortages and pressure on electricity systems are now being considered years before new facilities begin operating.
Khazna Data Centers, one of the companies developing the country's AI infrastructure, has assessed climate risks across its existing facilities and development sites, according to its first environmental, social and governance (ESG) report, published this month.
Stay up to date with the latest news. Follow KT on WhatsApp Channels.
For the initial 200MW development at the UAE-US AI Campus in Abu Dhabi, those assessments examined possible exposure to extreme heat, flooding and water stress over the medium and long term. The findings informed decisions about site layout, drainage, cooling systems and equipment.
Elisabetta Baronio, ESG Director at Khazna, told Khaleej Times that the environmental demands of a data centre are determined long before the first computer is switched on.
"A data center's impact starts well before it begins operating. It is shaped by where you build, the materials you use, how you design cooling and power systems, the suppliers you work with, and ultimately how the facility is operated over its lifetime," she said.
The challenge is growing as AI systems require increasingly powerful computing equipment concentrated within relatively small spaces. More computing power produces more heat, placing additional demands on the infrastructure keeping equipment at safe operating temperatures.
"Power distribution needs to support higher loads, cooling systems need to remove more heat, and the infrastructure still has to maintain the reliability and redundancy expected of a mission-critical facility," Baronio explained.
Managing water and cooling
Water is becoming an important consideration in how the UAE designs its data centres.
Khazna's report shows that its owned facilities consumed approximately 393,000 cubic metres of water in 2025, including around 305,116 cubic metres for cooling. Of that cooling water, approximately 244,374 cubic metres was potable water, while 60,576 cubic metres came from recycled sources.
That means roughly four in every five litres used for cooling were potable, illustrating the importance of finding alternative water sources as computing capacity expands.
At Khazna's DXB1 facility in Dubai, treated sewage effluent supplied by Dubai Municipality is already being used for cooling, reducing dependence on desalinated water. Across its portfolio, recycled water accounted for 19.8 per cent of cooling water requirements in 2025.
"The climate fundamentally shapes the engineering approach," Baronio said. "In the UAE, we have to manage high ambient temperatures while being conscious of the pressure that large-scale infrastructure can place on water resources."
Khazna primarily uses air-based cooling and reported that 19 of its data centres were equipped with adiabatic free cooling systems in 2025. These use outside air conditions to assist cooling, reducing reliance on conventional systems, although performance depends on the climate.
For higher-density AI facilities, the company is also incorporating liquid cooling, which brings coolant closer to the computer chips producing heat. Its planned 100MW AI-optimised facility in Ajman incorporates air-assisted, direct-to-chip liquid cooling into its design.
However, Baronio cautioned that circulating the same cooling fluid does not eliminate a facility's wider water requirements.
"Closed-loop liquid-cooling systems can continuously recirculate cooling fluid, but it's important to distinguish that from the water requirements of the wider facility. You still need to consider how heat is ultimately rejected, and the implications for both energy and water consumption."
The difficulty is not unique to the UAE. An October 2026 International Energy Agency report found that nearly 40 per cent of identified AI-dedicated data-centre capacity worldwide is located in areas with high or extremely high water stress. It also identified a trade-off between cooling systems that conserve water and those that consume less electricity.
Can AI help reduce its own energy demands?
Khazna is also examining whether artificial intelligence itself can help operate data centres more efficiently.
Through a pilot announced with the UAE Ministry of Energy and Infrastructure and Agility, the company is evaluating Phaidra's AI-driven control technology to optimise cooling performance, including in high-temperature environments.
Such systems can continuously analyse equipment performance, computing demand and surrounding temperatures to adjust cooling operations. Baronio said the potential is considerable, although Khazna has not yet established a verified savings figure.
"For me, the value of AI in this context will be demonstrated through measurable operational improvements, not simply by the sophistication of the technology being deployed."
The company's report also shows that 27.9 per cent of the electricity used for its mechanical load was covered by carbon-free electricity in 2025, through on-site solar generation and certified clean-energy instruments. Khazna aims to raise that figure to 50 per cent by 2030 and 100 per cent by 2035.
Yet Baronio acknowledged that greater efficiency does not necessarily mean a facility will consume fewer resources overall.
"A data center can become more efficient while its total energy consumption increases because it is supporting more computing capacity. The same applies to water. That's a reality our industry needs to be transparent about."
For the UAE, the preparations being made today will influence how the next generation of AI infrastructure operates through decades of changing weather and growing computing demand.
"The infrastructure we build today will still be operating ten years from now," Baronio said. "So my hope is that, by then, we can look back and demonstrate not only how much computing capacity we've delivered, but how much better we've become at managing the resources required to support it."