It might sound a bit far-fetched, but the future of AI growth and where we build our next data centres is increasingly tied to something as fundamental as water and local resources. Simply put, as AI gets hungrier for processing power, the data centres that house this power also need more of everything – electricity, land, and crucially, water for cooling. When these resources are scarce, particularly in areas already grappling with drought or population growth, it creates a significant hurdle for expansion, forcing a rethink of where and how we build.

Let’s be blunt: AI needs a lot of water. It’s not drinking it, of course, but it’s a critical component in keeping its powerful processors from overheating.

Cooling Challenges for Data Centres

Modern data centres are packed with powerful servers and graphics processing units (GPUs) that generate an immense amount of heat. Keeping these machines cool is non-negotiable for their efficiency and longevity.

Air Cooling – The Old Standard Under Strain

Traditionally, data centres have relied on air cooling. Think of giant air conditioning units blasting cold air through the rows of servers. This works well to a point, but as processing power density increases, so does the heat. Extracting this heat efficiently with air alone becomes more difficult and energy-intensive. Plus, very dry air can lead to static electricity issues, while very humid air can cause condensation – both bad news for electronics.

Water Cooling – The Efficient, Thirsty Alternative

To combat the rising temperatures, many data centres are turning to water cooling. This can involve anything from chilled water pipes flowing near the server racks (indirect cooling) to direct-to-chip liquid cooling where coolant flows directly over the hottest components. Water is incredibly efficient at transferring heat, far more so than air. This efficiency is great for performance and energy use within the data centre, but it shifts the resource demand from electricity for fans to water for the cooling towers.

Evaporative Cooling – A Double-Edged Sword

Many large data centres utilise evaporative cooling towers. This is where water is sprayed into the air to cool it down through evaporation, and then this cooled air or water is used to chill the equipment. It’s very effective and often more energy-efficient than traditional chillers, especially in dry climates. However, the clue is in the name: evaporation. A significant amount of water simply evaporates into the atmosphere, requiring constant replenishment. In drought-stricken regions, this becomes a major problem.

The Growing Water Footprint of AI

Estimates vary, but some reports suggest that a single query to a large language model (LLM) like ChatGPT can consume several millilitres of water for cooling. Multiply that by billions of queries daily, and you start to get a sense of the scale.

Individual Queries and Aggregate Consumption

While a few millilitres per query might seem negligible, the sheer volume of AI computations adds up astronomically. Training a large AI model can consume millions of litres of water over its training period. This is often an overlooked aspect of AI’s environmental impact, as most discussions focus on energy consumption.

The Indirect Water Cost

Beyond direct cooling, there’s also the embedded water cost in generating the electricity that powers the data centres. Many power plants, especially thermal ones (coal, gas, nuclear), also rely on water for cooling their own operations. So, even if a data centre uses a seemingly “waterless” cooling method, its energy source might have a significant water footprint further up the supply chain.

Dwindling Resources and Rising Tensions

When you combine the growing demand for water with increasing environmental pressures, you get a recipe for conflict.

Drought and Water Scarcity

The UK might be known for its rain, but even here, some regions experience water stress, particularly in the south-east. Globally, many areas where tech companies prefer to build due to infrastructure and talent availability are perennial drought zones.

Global Drought Hotspots

Regions like California, the American Southwest, parts of China, and many Mediterranean countries are increasingly battling prolonged and severe droughts. These are also prime locations or desirable areas for building advanced technological infrastructure, creating a direct conflict between technological expansion and fundamental human needs.

Competition with Agriculture and Domestic Use

When water is scarce, data centres aren’t just competing with each other; they’re competing with farmers who need water for crops and livestock, and with local populations who need it for drinking, sanitation, and everyday life. In a drought, the optics of a massive tech facility guzzling water while local residents face water restrictions are, understandably, very poor and can lead to significant public backlash.

Electrical Grid Strain

It’s not just water. Data centres are ravenous consumers of electricity. As AI models grow in complexity, so does their power hunger.

Peak Demand Challenges

Data centres operate 24/7, creating a constant, high demand for electricity. This demand can put significant strain on local electrical grids, especially during peak times or in regions with older or less robust infrastructure. Expanding these grids to accommodate new data centres is a costly and time-consuming endeavour.

Renewable Energy Limitations

While many tech companies pledge to use 100% renewable energy, the reality is more complex. Supplying a constant, massive load like a data centre solely from intermittent renewables (solar, wind) often requires significant battery storage or reliance on traditional backup power during periods of low generation. This adds another layer of cost and resource demand.

Local Resistance and Planning Hurdles

This resource competition isn’t going unnoticed by local communities and planning authorities.

Public Outcry and Negative Perceptions

When a new data centre project is announced in a water-stressed or energy-constrained area, it frequently sparks public opposition. Local residents often perceive these facilities as ‘water hogs’ and ‘energy guzzlers’ that offer little direct benefit to the community beyond a limited number of jobs, while straining essential communal resources.

NIMBYism (Not In My Back Yard) for Data Centres

While some might dismiss it as NIMBYism, residents often have legitimate concerns about noise pollution (from cooling systems and power generators), increased traffic during construction and operation, and the visual impact of large, often windowless, industrial buildings in their vicinity. The resource consumption aspects only amplify these concerns.

Environmental Activism

Environmental groups are increasingly scrutinising the water and energy footprints of the tech industry. They highlight the ecological impact, particularly in sensitive regions, and advocate for more sustainable practices or even moratoriums on new data centre builds in vulnerable areas.

Planning Permission Delays and Denials

Local councils and planning authorities are becoming more cautious about approving new data centre developments, especially in areas facing resource scarcity.

Stricter Environmental Impact Assessments

Developers are facing more rigorous environmental impact assessments (EIAs) that delve deeply into proposed water sources, consumption rates, wastewater discharge, and energy supply. Demonstrating minimal environmental impact and sustainable resource use is becoming a much higher bar to clear.

Infrastructure Demands as a Barrier

A common reason for planning delays or denials is the data centre’s profound impact on local infrastructure. Councils often demand proof that the existing water supply, sewage systems, and electrical grid can handle the increased load without negatively affecting current residents or future development plans. This proof can be hard to provide, or require significant, expensive, and time-consuming upgrades that developers may be reluctant to fund entirely.

Shifting Strategies for Sustainable Growth

Facing these growing challenges, the tech industry and policy-makers are being forced to innovate and adapt.

Advancements in Cooling Technologies

The race is on for more water-efficient and even waterless cooling solutions.

Closed-Loop Liquid Cooling

Moving away from evaporative cooling, closed-loop systems circulate a coolant repeatedly without significant loss. This drastically reduces water consumption, though it might increase initial capital costs. These systems are becoming more prevalent, especially in high-density installations.

Immersion Cooling

Another cutting-edge technique involves submerging server racks directly into a non-conductive dielectric fluid. This fluid is very efficient at dissipating heat, and because it’s a closed system, water consumption is minimal, usually only for an external heat exchanger if needed. This is highly efficient but complex to implement and maintain.

Adiabatic Cooling

This method involves pre-cooling air using fine water mist before it enters traditional air conditioners, but without direct evaporation of the water to cool the air inside the data centre itself. It’s more water-efficient than traditional evaporative cooling but still requires some water.

Decentralisation and Strategic Siting

Instead of clustering data centres in traditional tech hubs, a more distributed approach might be necessary.

‘Cold and Wet’ Locations

Companies are starting to look at regions with naturally cooler climates and abundant water resources, even if they are more remote. Places like the Nordics, for example, offer natural low temperatures and access to ample hydropower, reducing both cooling needs and reliance on water-intensive energy sources.

Edge Computing – Closer to the Source

The rise of edge computing, where processing is done closer to the data source rather than in massive central data centres, could also alleviate some pressure. Smaller, distributed edge data centres might have lower individual resource footprints and could be placed in areas where resources are less strained, though the aggregate still needs managing.

Regulatory and Policy Interventions

Governments and international bodies will play an increasing role in managing these challenges.

Water Efficiency Standards and Mandates

Expect to see stricter regulations on water usage for industrial facilities, including data centres. This could involve mandatory water recycling, limits on water consumption per unit of computing power, or even levies on excessive water use.

Incentives for Sustainable Practices

Conversely, governments might offer incentives for data centres that adopt cutting-loop cooling, use renewable energy, or contribute to local district heating schemes (using waste heat from the data centre to warm homes or businesses).

Integrated Resource Planning

A more holistic approach to resource planning at local and national levels will be crucial. This means considering the cumulative impact of industrial development on water, energy, and land, rather than reviewing each project in isolation.

The Future Landscape of AI Infrastructure

Location	Drought Severity	Resource Competition
California, USA	Severe	High
Western Cape, South Africa	Extreme	Medium
Madhya Pradesh, India	Moderate	Low

The challenges of drought and local resource competition are not insurmountable, but they demand a fundamental shift in how we approach AI infrastructure.

The Rise of ‘Sustainable Data Centre’ as a Benchmark

Sustainability will move from a ‘nice-to-have’ marketing point to a fundamental requirement for data centre development. Metrics around water usage effectiveness (WUE) and power usage effectiveness (PUE) will become even more critical, with low scores being a barrier to entry.

Innovation as a Driver for Progress

The necessity to conserve resources will spur innovation in cooling, energy efficiency, and operational management. Companies that lead in these areas will gain a significant competitive advantage.

A More Distributed and Diverse Ecosystem

We’ll likely see a more geographically diverse distribution of AI infrastructure, with data centres located based on resource availability rather than just connectivity or historical tech hubs. This might mean AI processing centres springing up in unexpected places, leading to new economic opportunities in regions previously overlooked.

In essence, the age of building massive, resource-guzzling data centres without considering their environmental context is rapidly drawing to a close. The future of AI, a technology promising incredible advancements, will be shaped not just by silicon and algorithms, but by something as ancient and essential as water. If we want AI to continue its exponential growth, we need to ensure its foundational infrastructure is built on principles of sustainability and respect for local resources, securing not just the future of AI, but the liveability of our planet.