AI’s Water Footprint: Data Centre Water Use is More Complex Than Metrics Show

Data centres are expanding rapidly to meet demand for artificial intelligence and cloud services, increasing pressure on water, energy, and other infrastructure in the regions that host them. But standard data centre metrics do not always tell the full story. Alongside water used for on-site cooling, significant water consumption takes place upstream through the electricity generation that powers data centres. In contexts where water scarcity, both for human consumption and industrial operations, is a key concern, especially during droughts which are becoming more frequent and severe with climate change, water consumption of data centres is critical. 

Data centre operators, businesses, investors, and host communities need to consider the full water footprint of data centres – including both on-site and off-site consumption – when assessing the risks and impacts of new capacity. 

The True Water Footprint of AI 

The International Energy Agency projects that data centre electricity consumption will more than double by 2030 in its Base Case scenario. 

Water use is an important, and often overlooked, part of the picture. The most widely used water efficiency metric, water usage effectiveness (WUE), measures only the water consumed on site (for example, to cool), not the water used to generate the electricity that powers a facility. Research on African data centres finds this type of water use is typically the larger share of the total water footprint.  

For data centres, and the businesses that rely on them through the digital services they build and buy, this could mean underestimating their exposure to water-related hazards and the ecological footprint of their business practices. 

Read on to explore the scale of the risks and recent developments around the world—from South Africa, Malaysia, and Brazil—and practical steps businesses are taking to understand and manage these risks. 

Why This Matters

In June 2026, the United Nations University Institute for Water, Environment and Health  published an assessment of AI’s environmental footprint. It estimated that 9.3 trillion litres of water will be needed to generate enough electricity to meet the projected demand from data centres by 2030. This does not include on-site water usage, such as cooling. That is almost enough water to meet the minimum annual domestic water needs [20 litres per person per day] of all 1.3 billion residents of Sub-Saharan Africa for a full year.

The same report estimates that data centres consumed 448 TWh of electricity in 2025 and puts the associated land footprint at more than 14,500 km². Its central finding is that assessing carbon alone overlooks the water and land impacts of AI, which are felt locally. 

For businesses, these impacts can mean competition for water with agriculture and households, pressure on local power grids, and regulatory and reputational exposure in host regions. (Learn more about the climate risks to technology and data security.) 

Who Invests and Who Supplies the Water

Cape Town: Approval Without a Water Figure

On 14 July 2026, Cape Town’s Municipal Planning Tribunal approved a land use application for two data centres covering 120,000m² to be built by Equinix near the airport, despite the application failing to state how much water or energy the facilities would use. 

The Legal Resources Centre, representing objectors, lodged an appeal on 14 September 2026, stating that this omission made the application “impossible for it to be properly considered”. Attorney Kimal Harvey, presenting for the objectors, estimated the facilities’ electrical demand at about 174 MW and cited research suggesting a data centre using traditional cooling consumes about 25.5 million litres of water a year per MW. These are the estimates presented by the objectors, not disclosed figures. 

South Africa has an arid to semi-arid climate; the country’s Department of Water & Sanitation’s National Water Resource Strategy reported annual rainfall averaging 465 mm (half the global average), and states that the country “is fast approaching full utilisation of available surface water yields”. 

The approval of the data centre is a first step: the Site Development Plan and building plans still require scrutiny. The decision now rests with Cape Town’s executive mayor, who acts as the appeals authority. See how businesses are managing their compliance and legal exposure.

The researchers attribute much of this to the electricity mix. Most data centre water consumption is indirect, occurring during electricity generation. In Algeria, Tunisia, Egypt, and Libya, this process uses a combined-cycle (combining gas and steam) using an average of 0.795 litres of water per Kwh, compared to the global average of 1.58L/Kwh for gas-only generation. 

The picture changes where electricity is dominated by hydropower. 

Ethiopia and the Republic of the Congo draw 95.7% and 99.6% of their electricity from hydropower and were the only two countries analysed where water consumption for data centres exceeds the global average.  Evaporation from reservoirs makes hydropower a water-intensive energy source.  

Malaysia: Rules Before the Largest Facilities Are Built

In the southern Malaysian state of Johor—located just across the strait from Singapore—the data centre sector is expanding rapidly, bringing growing demand for both water and electricity. 

In the state, data centres use around 9.07 million litres a day, or 0.6% of the state’s total consumption, and demand is expected to grow. Across peninsular Malaysia, peak electricity demand for data centres reached 849 MW in December 2025 and is projected to reach 4,811 MW by 2030. 

The thresholds are guidelines rather than an established enforcement record, but they provide published standards for investors and communities before the largest facilities are built. 

For businesses planning new capacity in Johor, this means water and energy availability, efficiency, and compliance need to be considered early in the development process. 

Brazil: A Strict Limit on Cooling Water 

Brazil is taking a different approach to managing data centre water use. 

On 15 September 2026, Brazil’s president signed REDATA, a special tax regime for data centre services that provides tax incentives for qualifying equipment purchases and imports. 

 To qualify, a facility must: 

• source all its electricity from renewable or low-emission generation; 
• publish a sustainability report covering its water use and power sources; 
• direct at least 10% of its capacity to the Brazilian market; 
• invest 2% of acquired asset value in domestic research with universities or research institutes; and 
• keep WUE at or below 0.05 litres per kilowatt hour, measured annually. 

This WUE limit is about 44 times lower than Malaysia’s national guideline and 36 times lower than Johor’s threshold. It is also below the average on-site WUE reported for Microsoft Azure (0.30 L/kWh) and AWS (0.18 L/kWh).  

As with Malaysia’s guidelines, the WUE limit covers on-site water only.  

REDATA’s energy condition allows low-emission generation, which can include hydropower—a water-intensive source. Businesses assessing facilities under either regime should therefore look at both on-site water use and the power supply. 

What Businesses Can Do 

Procurement and IT leaders 

• Ask providers for both measures. Request on-site WUE and the source and water intensity of the electricity supplying the facility. 

• Treat model choice as a resource decision. The African dataset estimated that writing a ten-page report used about 0.66 litres of water with Llama-3-70B and about 59 litres with GPT-4. The authors describe these as 2024 reference values, not current performance. 

• Consider where models run. A 2025 benchmarking study of 30 models estimated that DeepSeek-R1 used 29.1 Wh and over 200 ml of water on its own infrastructure, compared with 7.4 Wh and 34 ml on Microsoft Azure. These are modelled estimates, not metered measurements. 

Sustainability and reporting leads 

• Report on-site and electricity-related water separately, so that improvements in cooling are not mistaken for reductions in the total footprint. 

• Check water sources. Drinking water, treated wastewater, and groundwater carry different risks. Research notes that data centres commonly consume around 80% of the freshwater they withdraw, compared with roughly 10% for typical households and offices. 

Site selection and supply chain leaders 

• Include the electricity mix in site selection, as it can affect the water footprint more than local climate. 

• Secure supply before approval. Follow Malaysia’s reported practice of confirming power and water supplies before projects proceed, and use treated wastewater where available. Learn more about how businesses are keeping essential utilities and process inputs stable and reliable. 

• Agree drought priorities in advance. Establish, in writing, how water will be allocated between the facility, agriculture and households if supply is constrained. 

Operations and continuity teams 

• Plan for disruption to digital services. Where customer-facing systems depend on data centres in water- or power-stressed regions, build in redundancy and backup arrangements. See what businesses are already doing to manage nature and climate risks to technology and information systems.

Measure Your Own Digital Footprint 

Many organisations have not yet measured the footprint of their own digital estate. Digital Impact for Species, a free tool built by the Nature & Climate Impact Team at the University of Exeter, estimates the carbon footprint of any web page in seconds. First launched at the World Economic Forum in Davos, it helps organisations test their own websites and gives practical guidance to make them lighter.