Thirsty servers

25 March 2026

Data centres are essential to the functioning of our digital lives but present significant water demand challenges

The world generates over 400 million terabytes of data every day – that’s the equivalent of around 100 trillion smartphone photos, 12,000 for every human on the planet. This data is transmitted between our personal devices and a wide range of industrial sensors and monitoring systems, and is stored and processed at data centres.

Data centres are sometimes perceived as virtual infrastructure. They are in fact highly energy and water dependent facilities, and in many regions the rapid expansion of the digital economy is increasingly constrained not by data storage capacity or computing power, but by physical resources, particularly water availability.

Total global data centre capacity is to nearly double, with nearly 100 gigawatts (GW) of new capacity added between now and 2030. With that increase in capacity will come a significant shift in the scale of data centres. Large and mega-sized data centres will increase their share of total data centres from 28% in 2025 to 43% by 2030. This shift is mainly driven by an increased in demand for high-density AI infrastructure: to represent 50% to 70% of total data centre computing by 2030. Hyperscale data centres operated by giants like Amazon, Microsoft and Google, are expected to grow from 567 in 2025 to 738 by 2030.

Data centres play a critical and rapidly expanding role across countries and regions by supporting economic competitiveness, national security, public services and technological sovereignty. The global market for data centres was valued at US$269.79 billion (£197bn) in 2025 and is expected to grow to US$699.13bn by 2034. Global investment in the construction and upgrading of data centres is expected to reach approximately US$3 trillion to US$7tn between 2025 and 2030.

However, their rapid expansion also presents significant challenges, including high energy and water demand, pressure on local infrastructure and environmental impacts. There’s the embedded water associated with the materials used in data centre construction, plus operational water use that can be broadly classified into two categories: water used directly to prevent the servers from overheating, removing heat through evaporation; and water consumed indirectly through the generation of electricity supplying data centres.

A thirst for power

A data centre with a capacity of 100 megawatts (MW) can consume approximately 2.5bn litres of water per year, while the global data centre sector is estimated to use more than 560bn litres annually. Equivalent to London’s water consumption over roughly seven months, this figure is projected to increase to around 1,200bn litres by 2030.

The reporting of water consumption by data centres is insufficient across the industry, although leading tech companies do report aggregated water and energy use information. The Microsoft Environmental Sustainability Report presented net water consumption of 7,844Ml, while Google reported 24,000Ml for its data centres and offices. We could not find a report on Amazon’s water consumption data.

Data centres in numbers

• 400 million terabytes – data processed by data centres daily
• US$269.79 billion – global market for data centres
• 50-70% – proportion of data centre capacity used by AI by 2030
• 560bn litres – amount of water consumed by global data centres annually
• 1.5% – proportion of global electricity consumed by data centres

In a survey conducted by the New York City-based Uptime Institute in 2021, more than 60% of data centre operators say there is no “business justification” to collect water use information. Clearly, if we want better water use reporting and planning, we’re going to need both more regulation and consumer pressure – and it seems to be happening. Although data centres enhance local economies by bring much-needed jobs to some areas, authorities and the public are becoming increasingly aware of the potential impact on water resources and the environment. In August 2025, council leaders in Tucson, Arizona unanimously voted to shut down a proposal for a massive 290-acre data centre known as Project Blue because of concerns over water use.

The International Energy Agency estimated that in 2024 data centres accounted for 1.5% of global electricity consumption, or 415 terawatt-hours (TWh). This is expected to more than double to around 945 TWh by 2030. Such an increase will undoubtedly impact on water resources, but it’s unclear exactly how, given the changing nature of how electricity is generated.

The majority of new data centres are located in the water-stressed southeast of England

One of the major challenges is the mismatch between the locations of data centres, which are based on a range of technical and economic metrics, and the local availability of critical resources such as water and power. The majority of new data centres, classified as critical national infrastructure by the UK government, are located in the Southeast surrounding London, where a water deficit of 1,213Ml per day is projected for 2034 to 2035. This raises further concerns regarding their substantial water demand, which may necessitate stronger consideration of data centre water use within regional and water resources management plans.

Designing for scarcity

The future of data centres is increasingly oriented towards sustainability, with a strong emphasis on reducing water demand through innovative design and emerging technologies. This is being driven in part by the growing demand for more sustainable AI infrastructure. Back in 2011, The Green Grid, a trade association for the global data centre industry, introduced the concept of ‘water usage effectiveness’ (WUE) which evaluates how data centres use water. WUE measures the “total site water usage” for each unit of energy consumed by IT-related hardware. WUE can be as high as 2.5 m3/MWh for evaporative cooling systems and as low as zero for other technologies which do not rely on water, such as air cooling.

But while opting for a less water intensive cooling technology might decrease a data centre’s WUE, this might have unintended consequences: if the alternative technology is more power hungry, than there will be a proportionate increase in the amount of water consumed generating the electricity to power it. Furthermore, WUE does not account where the water is coming from or the resilience of the surrounding environment and ecosystems. All of which limits its usefulness as a primary metric for assessing data centre water use.

That said, we are now seeing a range of technological innovations that are helping to reduce data centre water consumption. Direct to chip and immersion cooling, higher allowable operating temperatures, AI driven cooling optimisation, improved closed loop designs and waste heat recovery all reduce cooling demand and associated water use. Where water-based cooling remains necessary, increasing emphasis is being placed on non-potable water sources such as treated sewage effluent, reclaimed wastewater, seawater and harvested rainwater. We’re also seeing advances in water recycling through advanced treatment and closed loop systems that reduce reliance on drinking water supplies and improve resilience during supply constraints.

Increasingly, thoughtful decisions about where sites are located complement these technological measures. Locating data centres in cooler climates, or close to alternative water sources and renewable energy infrastructure, can substantially reduce both direct and indirect water demand while accounting for the sensitivity of local water resources and thereby limiting environmental and ecological impacts. Together, these approaches support a transition towards more water efficient and resilient data centre design, shaped by local environmental and regulatory conditions.

Public engagement also has an important role in managing demand, particularly given the limited public awareness of the water costs associated with AI use. For example, one widely used large language model is estimated to consume approximately 0.52 litres of water to generate an average 100-word email. In the UK, around 8.5bn emails are sent each day, and approximately 17% of adults report using AI tools to assist with email writing. This indicates that user behaviour at scale can have a measurable impact on AI related water consumption, and that informed public choices could contribute to reducing overall demand.

In conclusion, addressing the growing water footprint of AI driven data centres demands a holistic, systems-based approach that moves beyond isolated efficiency measures. Water consumption must be assessed across the full lifecycle of data centre development and operation, from site selection and cooling technologies to energy sourcing and supply chain impacts. These assessments should be closely integrated with technological choices, regulatory frameworks, and local water availability, recognising that solutions effective in one context may be unsustainable in another.

Equally important is the need for policy support to ensure that data centre water use is systematically monitored, reported and explicitly considered within future water resource projections and planning frameworks. Such policies should be accompanied by suitable incentives that actively encourage the adoption of more water efficient and environmentally sustainable technologies and operational practices.

Broader socioeconomic considerations must also be embedded within decision making, including community water security, environmental protection and long-term resilience under climate change. By adopting an integrated and forward-looking approach, water and environmental professionals can help ensure that the rapid expansion of AI infrastructure is aligned with sustainable water management and the wider public interest.

--

For more on how the UK should be planning for water scarcity now and in the future, see CIWEM’s policy statement on drought management.

This article originally ran in the Spring 2026 print issue of The Environment magazine. Become a member of CIWEM today to gain access to the quarterly magazine, as well as digital access via MyCIWEM. Non-members can also access the monthly The Environment digital newsletter.

Reza Ahmadian is a professor of water and environmental engineering at the School of Engineering at Cardiff University
Man-Yue Lam is a lecturer in fluid mechanics and hydraulics at the School of Engineering at Cardiff University
Roja Ahmadi is a lecturer in data science at Ada, the National College of Digital Skills