How AI's Water Thirst is Reshaping the Climate Equation

In the dry outskirts of Phoenix, Arizona, a sprawling data-centre campus hums day and night, powering the training of large-language models, generative AI services and massive cloud workloads. It looks like the future of technology, but if you dig beneath the servers and cooling towers, you will find a less-celebrated reality: freshwaterbeing pumped in bulk to cool machines, compete with other users, and quietly accelerate ecological risk. A 2025 analysis projected that in the Phoenix region alone, data-centre water demand could surge nearly 900% in six years, from ~385 million gallons/year to ~3.7 billion gallons.

Water stress in the Phoenix Region of Arizona, USA (Source: “Drained by Data: The Cumulative Impact of Data Centers on Regional Water Stress”, CERES, 2025)

That single projection encapsulates the broader challenge: as AI workloads proliferate, infrastructure expands, and climate-driven water stress intensifies, the hidden resource cost of AI is increasingly visible. It is no longer a question of just electricity, latency or compute. It is a question of water. And not just anywhere — where the water sits, who needs it and whether it is being managed for a changing climate matter profoundly.

A Global Footprint of Water Demand and Potential Risk

More data now show the scale of consumption and the urgency of strategic action. According to the Environmental & Energy Study Institute (EESI), large data centres can use up to 5 million gallons (~19 million litres) per day for cooling — roughly the annual water use of a small town. In Europe, the European Parliament referenced a figure of 6.6 billion m³ of water by 2027 tied to data centre operations under AI growth. Elsewhere, in Virginia’s “Data Centre Alley” in the U.S., water consumption jumped by nearly two-thirds between 2019 and 2023 to 1.85 billion gallons (~7 billion litres).

While not all water is equal (some is non-potable, some is recycled), the more telling metric is location + competition. Two-thirds of new data centres built since 2022 in the U.S. are located in “high water-stress” regions (California, Arizona, Texas, Illinois, Virginia). Internationally, investigations show major players building in regions already facing water scarcity: e.g., Latin America’s drought-prone zones.

These are not idle numbers. They illustrate three intertwined trends:

• Spatial risk: Infrastructure sited in water-scarce regions means competition between AI infrastructure and other essential uses (agriculture, municipal, ecosystems).

• Embedded water cost: AI’s water footprint isn’t just onsite cooling — it includes energy generation, evaporative losses, supply-chain manufacturing. As one article notes, indirect water use (via power generation) may dominate the footprint.

• Climate feedbacks: In hotter, drier futures, cooling demands rise, water becomes scarcer, and the “when” and “where” of AI infrastructures create system-level vulnerabilities.

  
  

If AI infrastructure is treated purely as a digital problem (compute-first) with water treated as incidental, companies and policymakers may overlook a critical resilience boundary.

Boiling liquid carries away heat generated by computer servers in two-phase immersion cooling system (Source: Microsoft)

Deeper Analysis: Why the Water-AI Nexus Matters

Resource Competition & Societal Risk

When data centres withdraw large volumes of water, especially in stressed basins, the ripple effects matter. In some cases, local households and agriculture feel the squeeze. For example, reports from Bengaluru (India) show data-centre demand (~8 million litres/day) adding to an already severe municipal water crisis. In Latin America, communities near planned data centres have raised alarm that tech infrastructure “is draining water from cities that need it most.”

Hence, AI infrastructure is not just an industrial consumer — it intersects with water justice, public-resource allocation, and climate resilience. From a business lens, every client building or expanding data-centres must ask: Is my water use competitively depriving other users? Are we locking ourselves into risk by placing high-demand infrastructure in water-scarce basins? What is the economic implication under different scenarios of climate change?

Hidden Cost & Resilience Exposure

Many firms focus on energy efficiency (PUE: power usage effectiveness). But water usage effectiveness (WUE) remains under-tracked. For example, some 2010s era data centres reported less than one-thirdtracked water consumption at all. In 2024‐25, the lack of transparency persists. Without a full understanding of water withdrawals, evaporative losses, indirect embedded water, firms are exposed to governance, reputational and liability risk. As the Lawfare analysis observes: “A small 1-MW centre may use up to 26 million litres/year; given scale, the water footprint rivals cattle and textiles industries.”

We can also observe scale effects: larger-scale data centres are often more cost-efficient in water use per unit compute than smaller ones. A 2025 EU study found that centres > 10 MW used up to three times less water per unit of processing than smaller centres. From a strategic view: bigger can be greener — if and only if designed well.

Climate-Resilient Water Strategy (or Lack Thereof)

With climate change, many of the assumptions built into data-centre siting and design are shifting. Areas previously considered “safe” for water may become stressed; cooling demands may increase; wastewater treatment capacity may be stressed. The technical design must evolve from “just keep it cool” to “keep it cool in a variable, stressed climate”. The technology choices (liquid cooling, closed-loop, grey-water reuse) thus become integral to risk management — not just operational efficiency.

Thus, whether AI infrastructure becomes a water-resilience assetor a liability really depends on foresight. From the industry perspective, the differential payoff lies in proactive design, robust site selection, and integration with circular water systems.

Strategic Recommendations: For Organisations, Policymakers & Investors

Here are prioritized actions: designed for operational tweaks and strategic positioning, as well as risk mitigation and competitive advantage:

For Organisations Deploying AI Infrastructure:

1. Water-risk screening & strategic siting

   1. Map prospective sites against water-stress indices (e.g., Aqueduct Water Risk Atlas, local drought frequency) and future climate projections.

   2. Prioritize locations with moderate/low water stress, cooler ambient temperatures and access to alternative water sources (treated wastewater, seawater, aquifer recharge).

   3. Avoid being incentivised purely by tax breaks or cheap land if the water risk is high: short-term gain may become long-term cost.

      
      

2. Hydrologic-first cooling and reuse systems

   1. Commit to water usage effectiveness (WUE) targets, e.g., ≤ 0.5 litres/kWh IT load (industry best-practices already report ~0.3 L/kWh).

   2. Use closed-loop liquid immersion cooling, direct-to-chip cooling, geothermal or seawater cooling where feasible. Several facilities report up to 50–70% water withdrawal reduction with such technologies.

   3. Recycle water onsite: use grey-water, captured rainwater, advanced treatment and reuse for cooling towers rather than fresh potable withdrawals.

   4. Monitor and publicly report WUE, water withdrawal and reuse metrics as part of transparency and stakeholder credibility.

   

   Different technology types for immersive cooling. (Source: Cleantech Group, 2023)

3. Integrate energy & water strategy (the water-energy-AI nexus)

   1. Optimise energy sources: renewable energy, low-water power generation pathways, waste-heat reuse. Since power generation often drives large hidden water use, shifting to low-water intensity energy (e.g., wind, PV) reduces the embedded water footprint.

   2. Use AI and digital twins to optimise cooling load, server utilisation and water cycles — early pilots show water-footprint reductions of up to 3X via workload scheduling across sites.

      
      

4. Scenario planning & stress testing water supply

   1. Include drought and groundwater depletion scenarios in infrastructure planning.

   2. Engage local stakeholders and regulators: water licences, permit conditions, community water usage.

   3. Build flexibility: modular data-centre design, ability to switch to air-cooling or non-water modes during drought.

For Policymakers and Regulators:

1. Mandate transparent water-use reporting by data-centre operators — water, energy, withdrawals, reuse. The EU’s new directive is a path-finder.

2. Incorporate water use of data centres in basin-scale water resource planning — not just electricity. Zones with major tech investment should be integrated into regional water-allocation frameworks.

3. Incentivise water-efficient cooling technologies: e.g., grants, low-interest loans for closed-loop cooling or alternative cooling in water-stressed zones.

4. Define and enforce cumulative water withdrawal caps in high-stress basins: data centres ought to be subject to the same resource constraints as other large industrial users.

   
   

For Investors and Corporate Decision-Makers:

1. Include water stress indicators in ESG due diligence: e.g., site in basin with < 0.3 water-stress index, or operator WUE ≤ 0.5 L/kWh.

2. Recognise that water risk is increasingly a structural cost: permits, community opposition, stranded water assets (dry wells) are real.

3. Reward scale and design maturity: larger hyperscale players who optimise water per unit compute may have competitive advantage over smaller operators who rely on older cooling systems and less reuse. The EU study showed large centres can use up to three times less water per unit computing than smaller ones.

Conclusion: From Digital Growth to Resource Intelligence

As AI and large-scale computing continue to transform industries, the narrative cannot remain solely about processing power, latency, or energy consumption. The next frontier is resource intelligence — particularly water. When a facility withdraws millions of litres of freshwater in a basin already stressed by climate change, it becomes part of a broader resilience equation: the future of food systems, public supply, ecosystems, and corporate sustainability. We are witnessing the point where infrastructure meets hydrology.

For organizations, the path ahead is clear: treat water not as a free utility, but as a strategic input. Prioritize smart siting, closed-loop cooling, energy-water integration, reuse, and transparency. For policymakers, the job is to build frameworks that recognize data centres as major water users, not peripheral. For investors, the choice is to back not just capacity but capacity that respects water boundaries.

In the era of AI, the most powerful chip may be the one that uses lesswater and energy. And the resilient company may soon be the one that thinks about water as intelligently as it thinks about algorithms.

[First published on Substack "Ginci Insights" on October 28, 2025: https://gincinno.substack.com/p/how-ais-water-thirst-is-reshaping?r=2cxt8s]