According to a Morgan Stanley report, artificial intelligence (AI) data centres are expected to increase global water consumption in the coming years. The report states that annual water use for cooling and electricity generation by AI data centres is expected to reach about 1,068 billion litres by 2028.
Additionally, while water used for cooling data centres is widely recognised, the indirect water use from electricity generation is often underappreciated. Meaning, water consumption estimates depend on assumptions around water intensity, consumption factors, cooling technologies, and the regional energy mix. As more efficient solutions are adopted, these figures could change. To reflect this uncertainty, the report presents three scenarios under which AI’s water use could range from 637 billion litres to 1,485 billion litres annually by 2028.
The AI water footprint spans three scopes: scope 1 covers on-site data centre cooling, scope 2 relates to electricity generation, and scope 3 covers semiconductor manufacturing. Further, electricity generation (scope 2) typically accounts for the largest share of water use, followed by cooling (scope 1) and semiconductor production (scope 3).
Similarly, more than half of the leading global data centre hubs are located in regions already facing medium risks from water scarcity, drought, flooding, and declining water quality. Many secondary markets and emerging hubs are situated in areas with high or very high-water stress.
