Philip Robotham | Head | Intermediary | Schroders South Africa | mail me |
While Artificial Intelligence’s (AI) mammoth energy requirements have been widely publicised, its water footprint receives far less attention. World Water Day, celebrated on 22 March 2025, provides an opportunity to highlight AI’s lesser-known environmental impact – its water usage.
A study by the University of Massachusetts Amherst found that training a single generative AI model consumes 284,000 litres of water. That equals the amount of water an average person consumes over 27 years.
Can AI ever quench its thirst?
Although AI is a digital technology, it depends on physical processes that have real environmental impacts.
Data centres consume enormous amounts of energy and water, and manufacturing semiconductors requires incredibly sophisticated equipment and facilities. According to a Forbes article, AI’s projected water usage could reach 6.6 billion cubic metres by 2027. This alarming projection signals the urgent need to address AI’s growing water footprint.
The water footprint of data centres includes water used for electricity generation and cooling processes. This is a significant problem because AI adoption is rapidly expanding across industries.
Recent innovations by DeepSeek are fast-tracking the democratisation of AI, further accelerating this widespread use. Consequently, demand for data centres and their associated water consumption will continue to increase. But there are ways for data centres to reduce their operational water footprint.
The electrical energy entering a datacentre must eventually be ejected as waste heat through a cooling system. Data centre cooling accounts for 33-40% of total data centre energy usage. Additionally, cooling consumes hundreds of billions of litres of fresh water annually.
Innovative solutions to quench AI’s thirst
On the positive side, steps can and are being taken to reduce data centre water requirements. Recently, significant advances in cooling technologies have improved efficiency and reduced operational costs. Furthermore, the location of data centres can make a huge difference in their environmental impact.
About 70% of freshwater is used in food production, increasing competition for water resources. Building datacentres in regions with pre-existing water scarcity indirectly impacts agriculture. This may worsen droughts, which are already becoming more frequent and severe.
By locating data centres in areas with abundant renewable power or cooler climates, significant emissions savings are achievable. Additionally, innovative methods exist for repurposing heat generated by datacentres. Among these methods is using excess heat to support district heating systems.
High-temperature water can be channelled into households and buildings to provide heating. Alternatively, datacentre thermal energy may be harnessed for agricultural applications. This includes providing year-round heating for greenhouses and warming water in fish farms or public swimming pools.
Sustainable food and water systems require massive investment to meet future challenges. Our food and water systems will face unprecedented pressures over the next 30 years.
In conclusion
Aside from AI’s environmental impact, we must also reduce greenhouse gas emissions and adapt to a changing climate. At the same time, we need to improve biodiversity, reduce pollution and waste and encourage healthier global diets.
Early estimates suggest this structural shift will require around $30 trillion of capital reallocation by 2050. This represents a significant step change in investment rates across the food and water value chain.
We believe these changes will create opportunities for investors in food and water sustainability.
