The Middle East and North Africa holds 6.3% of the global population and 1.4% of global renewable freshwater. Eleven of the seventeen most water-stressed countries on earth are in MENA. Data center investment is accelerating into the region anyway, drawn by sovereign AI ambitions, low-cost power, and favorable tax structures, without a credible answer to what cooling those facilities will cost the water table.
The problem is not unique to MENA, but it is most acute there. Conventional evaporative cooling towers in a data center draw between 1 and 5 liters of water per kilowatt-hour of IT load. A 100 MW facility running at 50% utilization consumes tens of millions of liters annually. In Phoenix or Austin, that number creates political friction. In Riyadh or Abu Dhabi, it is competing directly with drinking water infrastructure that already depends on energy-intensive desalination to exist.
MENA region: 6.3% of global population, 1.4% of global renewable freshwater. 11 of 17 most water-stressed countries globally are in MENA. MENA accounts for 42% of global desalination capacity. Kuwait derives 90% of drinking water from desalination; Oman 86%; Saudi Arabia 70%; UAE over 40%. Projections: nearly the entire MENA population could face acute water scarcity by 2050 without sustained intervention.
The standard response from data center developers operating in MENA is that the region's desalination capacity makes freshwater scarcity less relevant. That framing misses the energy cost of the substitution. Desalinating seawater requires 3–10 kWh per cubic meter. Using desalinated water to cool a data center means burning power to make water to cool a facility that is burning power. The circular inefficiency compounds at scale.
Kuwait runs 90% of its drinking water through desalination. Oman runs 86%. Saudi Arabia runs 70%. These are not margins of capacity that can absorb the water draw of a serious AI data center buildout without either building far more desalination infrastructure or accepting real degradation in supply reliability for everyone else. The political economy of that tradeoff will not resolve itself quietly.
Training a single large AI model consumes millions of liters of water across the full lifecycle, including the indirect water consumption of the power generation feeding the facility. Answering 20 to 50 AI queries consumes the rough equivalent of a 500 ml water bottle when accounting for the full chain. Neither number is large in isolation. Aggregated across the inference demand that every major MENA sovereign AI initiative is targeting, the numbers become a credible infrastructure constraint.
The indirect water draw, meaning water consumed by the power plants supplying the data center, is typically larger than the direct facility cooling draw and is almost never reported. A gas-fired power plant supplying a data center in Saudi Arabia consumes water in its own cooling processes. That water consumption is attributed to the power sector, not the data center, which is why operator-reported water figures consistently understate total consumption.
Closed-loop liquid cooling with dry heat rejection is the technically correct answer for water-scarce environments. A direct-to-chip liquid cooling system circulating a non-evaporative fluid and rejecting heat through dry coolers or air-cooled heat exchangers can operate with near-zero consumptive water use. Recovery rates of 96% in closed-loop configurations have been demonstrated commercially. At those figures, the water draw becomes a maintenance quantity rather than an operational cost.
The barriers are familiar: capital cost, workforce capability to design and commission liquid systems, and the legacy air-cooled installed base that operators are trying to run commercially while also upgrading. MENA data centers built from scratch for AI workloads have no excuse for designing evaporative cooling into the facility. The brownfield facilities retrofitting into AI density have fewer good options, but closed-loop liquid cooling should be the specification target regardless. The water constraint in MENA is not going to ease. The AI workload density is going to increase. The gap between those two curves closes on one side only.