Thirsty for data: water scarcity challenges Gulf’s AI ambitions

• Limiting data centre capacity
• Innovation offers alternatives
• Waterless cooling costly

Water scarcity could complicate the Gulf’s ambitious plans to build ever-larger and more powerful data centres, experts say.

“Water accessibility can significantly limit data centre capacity, particularly as cooling demands increase with higher density workloads and rising ambient temperatures,” Rasheed Ahmad, technical director at American Water and Environmental Solutions, tells AGBI.

Saudi Arabia’s new AI company Humain has pledged to build artificial intelligence “factories” with a capacity of 500 megawatts, for which US technology company Nvidia will provide “several hundred thousand” of its most advanced graphics processing units (GPUs). 

For its part, the UAE, led by Abu Dhabi, announced plans for a 5 gigawatt joint AI campus with the US. 

But the desert countries of Saudi Arabia and the UAE, whose combined population of 45 million is growing, already face acute water challenges. Ambitious plans to expand AI infrastructure and build larger data centres risk exacerbating this.

“Water availability is becoming a critical constraint,” says Ahmad.

Water desalination, powered primarily by natural gas, accounts for between 60 and 90 percent of water needs in the Gulf, a region marked by extremely low rainfall of less than 100 millimetres per year and high rates of evaporation due to intense heat.

Data centres often consume vast amounts of water as they cool the microchips required to process data and the buildings in which they are housed.

The Gulf and wider Middle East host at least 280 data centres, with more planned. Of these, 42 are in Saudi Arabia and 56 in the UAE, according to Data Center Map.

Alamy via Reuters

“In water-stressed regions, the ability to scale may ultimately depend on reducing or fully eliminating water use in cooling systems,” says Ahmad.

Air cooling is one option for some data centres and it “might be adequate for AI inferencing workloads,” according to research by McKinsey & Company. 

AI inference is the process of using a trained machine learning model to make predictions or decisions based on new input data. It is the deployment phase of an AI system. 

But these systems cannot be used for training AI as the chips doing this work generate even more heat and must be cooled with more efficient liquids.

In AI, the training phase is where the model learns patterns from a large dataset. Training uses even more power than inference.

Saving water should be a priority when building new data centres, Jay Dietrich, research director of sustainability at data centre-focused New York-based Uptime Institute, tells AGBI.

Many regional data centres still rely at least partially on water for their cooling needs. The AD1 Abu Dhabi IBX Data Center, for example, which is newly built by Equinix, uses water-cooled chillers to extract the heat generated by processors. 

Other approaches are under development, including those that do not use any water at all, says Dietrich, but these are more expensive, adding to already high costs.

• UAE and US to build largest AI campus outside America
• Water scarcity: The GCC’s desalination challenge
• Dubai’s hottest new job title: chief AI officer

“If I was in the Middle East, I’d figure out how to do waterless,” says Dietrich. “They’re more expensive systems; the exchange is you get rid of water treatment.”

Estimated to cost from $30,000 to $40,000 each, the 18,000 Nvidia GB300 Grace Blackwell chips ordered so far by Saudi Arabia could cost over $600 million. 

Cooling costs are over and above that. On average, a GPU uses between 300 and 700 watts of power. Cooling takes up between 30 and 50 percent of the power. 

Herbert Radlinger, managing director of Germany-based data centre designers NDC-Garbe, does not necessarily expect water scarcity issues to fully curtail the Gulf’s data centre ambitions.

Data centres have the option of using less water but consuming more energy to achieve the same level of cooling, he says. However, this introduces energy inefficiencies in hot climates like the Middle East, as well as environmental consequences.

“They’re building data centres in regions that need extreme amounts of cooling energy to achieve this, either with power, or water, or both,” he says.