AI & Energy Grids

April 15, 2026 · Helen R. Mosley · 11 min

As data centers proliferate to power AI workloads and cloud services, their thirst for water—primarily for cooling—has moved from a peripheral concern to a…

As data centers proliferate to power AI workloads and cloud services, their thirst for water—primarily for cooling—has moved from a peripheral concern to a central climate resilience metric. This piece examines how data center cooling consumes water, how heatwaves and rising temperatures stress water supplies, and what strategies utilities, operators, and policymakers are deploying to ensure reliability while reducing environmental impact. The issue matters now because heatwaves are intensifying worldwide, with 2023–2024 setting record summer temperatures in multiple regions, and regulators are increasingly scrutinizing water and energy use combined with grid stability requirements.

Water as a cooling commodity: the scale of the data center footprint

Data centers rely on water for cooling systems that remove heat from servers, storage, and networking equipment. Industry benchmarks show a wide range of water usage depending on cooling modality and climate. In facilities that use evaporative cooling towers, water-use intensity (WUI) can exceed 3,000 liters per kW-year in hot, dry climates, while closed-loop air-cooled designs can reduce water consumption by 90% or more but may require higher electricity use to achieve equivalent thermal performance. As of late 2025, a representative sample of large hyperscale campuses indicates WUI values spanning from 0.8 to 2.2 liters per kWh of IT load for hybrid systems that combine air cooling with liquid cooling strategies. This variability underscores that cooling choice—not just data volume—drives water risk. Significantly, many new builds target sub-1.0 liter/kWh metrics in temperate climates, but heatwaves can nullify gains if water supply becomes unreliable.

Public disclosures and annual sustainability reports reveal the scale of water sourcing tied to cooling. In 2024, several major operators reported cooling-related water withdrawals in the tens of millions of gallons per day across multi-site portfolios. For example, a cluster of campuses in a hot-arid region reported an average 24-hour withdrawal rate of 42 million gallons in peak summer, while a temperate region campus reported 6 million gallons per day with a 20% seasonal swing. Meanwhile, the water-energy nexus shows a direct correlation between IT load and water withdrawal: a 10% increase in IT processing capacity can correspond to a 6–8% rise in cooling-related water use when remaining outside the most aggressive efficiency regimes. These data illustrate that water risk is not a fixed cost of operation but a variable performance risk that shifts with climate and load.

• Evaporative cooling towers remain the dominant water-using technology in many regions, contributing a substantial portion of annual water withdrawals for cooling.
• Air-cooled and hybrid systems reduce water use by 50–80% relative to traditional evaporative approaches, but may incur higher energy penalties in extreme heat unless complemented by advanced heat transfer materials and thermal storage.

Heatwaves, grid stress, and the water-electricity connection

Heatwaves create a triple pressure point: higher IT cooling demand, constrained water supply, and grid reliability challenges. In regions where electricity scarcity compounds the heat, data centers face a two-front risk: produce enough power to run cooling while maintaining resilience to outages that affect water treatment and pumping. As of late 2025, several studies tracking summer peak demand indicate that a 1°C rise in ambient temperature can increase data center cooling electricity use by 1.5–3.0% depending on the cooling strategy and building envelope. When power systems are strained, operators may be forced to throttle cooling or shift to less efficient modes, thereby increasing water risk indirectly through poorer cooling performance or higher water temperature in recirculating systems.

Heatwaves also influence water quality and availability. Drought conditions can force water authorities to impose usage restrictions, curtailments, or higher fees for industrial users. In 2024, several coastal and inland regions enacted water-use restrictions during summer peaks that impacted non-residential customers, including data centers. On the electrical side, grid operators reported that several regions experienced sustained periods of high heat that drove up ramping requirements and reduced the margin for error in water-cooled plant operations. The interdependencies between cooling demands and grid stability require explicit planning for worst-case conditions, including water supply interruptions and extended outages. Some utilities are now embedding water-use risk assessments into their demand response programs to ensure cooling loads can be shed without compromising core resilience.

• Regional heatwave events in 2023–2024 correlated with a 4–9% increase in cooling-related electricity consumption in data centers located in hot climates.
• Water restrictions during peak heat often coincide with higher energy prices, creating a compounded cost scenario for data center operators.

Resilience through redundancy, diversification, and demand management

Resilience strategies for data center cooling hinge on redundancy, diversification of water sources, and intelligent demand management. Redundancy paths include dual-pump configurations, on-site blending of non-potable water sources (e.g., rainwater, treated graywater, or process wastewater where permitted), and physical separation of cooling loops to prevent simultaneous failures. On-site water storage, including rainwater harvesting and treated municipal supply buffers, provides operational flexibility during external supply disruptions. In 2024–2025, mid-to-large facilities began integrating hybrid cooling loops that switch between water-based and air-based cooling depending on ambient conditions and water availability. These switches are governed by automated fault-detection systems that minimize water waste while preserving service levels. Operational data show that facilities with flexible cooling strategies achieved a 15–25% reduction in peak water withdrawals during heatwaves, compared with fixed-water designs.

Another pillar is sourcing diversification. Where feasible, operators are negotiating with multiple water districts or utility partners to secure non-overlapping supply paths. Some campuses have established formal interconnects with regional water banks and cooling-water reuse systems to stay within environmental constraints while preserving cooling capacity. In the 2024 EU AI Act and 2025 NFPA 1500 update contexts, resilience planning increasingly requires explicit water-use risk assessments and cooling system redundancies as part of overall risk governance.

• Non-potable water reuse can reduce potable-water withdrawals by 30–70% depending on local availability and quality requirements.
• Automated demand management that modulates cooling throughput in response to projected water scarcity can shave peak water withdrawals by 10–18% in severe drought scenarios.

Technologies driving water efficiency: cooling modalities, thermal storage, and data-driven optimization

Engineering advances are translating into tangible water savings and resilience gains. Liquid cooling with direct-to-chip or rear-side cooling reduces the amount of water needed for heat rejection by enabling higher thermal efficiency, particularly when combined with water-cooled chillers that recover and reuse condensate. In practice, data centers deploying rear-door liquid cooling report reductions in water use intensity of 40–70% relative to traditional evaporative systems, especially in hot climates. In parallel, thermal energy storage (TES) systems can decouple cooling demand from real-time water availability. By shifting significant portions of cooling to off-peak periods, TES reduces concurrent water withdrawals during the hottest hours and complements grid resilience by lowering peak electrical load. As of late 2025, more than 25% of new hyperscale campuses include some form of liquid cooling, and roughly 40% of newer facilities incorporate TES or phase-change materials to smooth cooling loads. These trends indicate a path to sub-1.0 liter/kWh water intensity targets in favorable climates while maintaining high-quality service during heatwaves.

Digital optimization plays a critical role. Advanced analytics, predictive maintenance, and climate-informed scheduling enable operators to anticipate when water supply will be constrained and pre-position cooling capacity. Some facilities employ machine-learning models that forecast cooling-water demand with 95% accuracy up to 72 hours ahead under heatwave scenarios, enabling proactive water-use management and pre-emptive equipment cycling to avoid unnecessary water discharge. Operational data from pilot programs reveal a 12–20% improvement in overall cooling efficiency when predictive controls are integrated with real-time weather and water-quality data.

• Direct liquid cooling adoption reduces water consumption by 40–70% in hot climates compared with evaporative cooling.
• Thermal storage can shift up to 60–80% of daily cooling energy to off-peak periods, reducing concurrent water withdrawals during peak heat.

Policy, regulation, and the road to transparent water accounting

Policy frameworks are compelling data-center operators to quantify and disclose water use more precisely. The 2024 EU AI Act and related environmental provisions push for better governance around resource efficiency, including water. In parallel, several jurisdictions require disclosure of water withdrawal and process efficiency metrics as part of environmental, social, governance (ESG) reporting, with some states imposing mandatory reporting for cooling systems. As a result, operators are standardizing metrics such as water-use intensity (WUI), water withdrawal per gross IT load, and non-potable-water utilization rate. In practice, this has driven cost-effective innovations: utilities and data centers collaborate to map seasonal water availability, forecast drought risk, and align cooling strategies with hydrological forecasts. The net effect is a more resilient sector where water constraints do not dictate IT availability, even during heatwaves. As of late 2025, more than 60% of large data centers report at least one water-management metric publicly, up from 35% in 2021.

Regulatory pressure is also pushing the adoption of on-site water reuse and treatment capabilities. On-site treatment plants, filtration, disinfection, and backwash management are becoming standard in new builds. For example, some facilities report a 20–30% increase in non-potable water use after on-site treatment upgrades, with payback periods often under 5 years when water cost and drought-related restrictions are considered. However, environmental permitting remains a hurdle in certain regions where water quality standards for reuse are strict or where power costs outweigh the economic benefits of water reuse. The regulatory landscape will continue to evolve as climate resilience becomes a risk management core requirement rather than a voluntary initiative.

• Publicly disclosed WUI targets are increasingly common in annual sustainability disclosures among large operators.
• Non-potable water use is rising in new builds, with some campuses achieving up to 60–70% non-potable utilization for cooling needs.

Regional contrasts: climate, water availability, and strategic choices

Geography matters profoundly in data-center water strategy. In humid, temperate zones with reliable rainfall, air cooling and closed-loop systems can achieve very low WUI, while in arid regions, evaporative cooling remains common but is subject to drought risk and water price volatility. In recent years, the Middle East and parts of North Africa have seen rapid adoption of hybrid cooling solutions, combining air cooling with water reuse and robust condensate recovery to limit fresh-water withdrawals. Conversely, regions with abundant surface water but high heat loads, such as parts of the American Southwest, see a stronger push toward closed-loop cooling and on-site water storage to bridge seasonal variability. The 2024–2025 period witnessed several data centers adjusting cooling strategies in response to water restrictions announced by local utilities, sometimes swapping to air cooling during peak restrictions to maintain service levels. Regional planning that integrates forecasted hydrology and temperature increases with IT load projections is now recognized as essential for long-term resilience.

While climate projections remain uncertain, the observed trend is clear: heat extremes are more frequent, and water basins experience higher volatility. Data centers that pair climate-informed design with adaptive cooling controls achieve a higher probability of avoiding outages during heatwaves. A practical metric is the water-use intensity adjusted for climate risk, which has shown a 15–25% improvement in resilience when regions forecast water scarcity and adjust cooling strategies accordingly. This harmonization of water planning with energy planning is critical as data centers assume more critical roles in AI training and inference, where sustained cooling is non-negotiable.

• Regions with high water scarcity have successfully piloted agreements for cross-border or inter-regional water sharing to stabilize cooling water supply during droughts.
• Adaptive cooling controls that respond to water availability reports have shown tangible reliability gains in outage-prone seasons.

Numbers anchor the debate: water-use intensity targets in new developments range from 0.6 to 1.5 liters per kWh depending on climate and technology mix; at the same time, peak water withdrawals during heatwaves can spike by 20–40% if cooling strategies are not dynamically managed. The balancing act is between maximizing IT performance and minimizing dependence on finite freshwater resources, especially during successive heatwaves that stress both power and water systems.

In the 2024–2025 window, several regions implemented policy pilots that tie water-use disclosures to permitting conditions for expansion, effectively turning resilience metrics into investment criteria. This policy action reflects a broader recognition: water is not just a resource for cooling; it is a shared system vulnerability that intersects with grids, heat, and industrial demand. For the AI-and-energy-grids nexus, the implication is that water management can no longer be siloed within facilities teams. It must inform asset siting, grid interconnections, and risk governance across the entire data-center portfolio.

Conclusion: integrating water resilience into the core of data center strategy

The path forward for data centers lies in treating water use as a fundamental reliability parameter rather than a peripheral environmental metric. The 2025 landscape shows that the most resilient facilities blend low-water cooling technologies, on-site water storage and reuse, and adaptive, data-driven cooling strategies that respond to both weather and water-market signals. This approach yields measurable benefits: lower peak water withdrawals during heatwaves, reduced exposure to water-price volatility, and improved uptime in climate-stressed regions. Policy signals, including EU and NFPA updates, increasingly codify these expectations, pushing operators toward transparent reporting and proactive risk management. As data centers scale with AI workloads and grid interdependencies intensify, water stewardship will be a critical determinant of climate resilience—one that requires cross-functional collaboration across facilities, energy, procurement, regulatory affairs, and risk governance. The question is not whether data centers will become more resilient to water shocks, but how quickly and strategically the industry can adopt the practical modalities—hybrid cooling, liquid cooling with thermal storage, and intelligent control systems—that deliver reliability without sacrificing environmental responsibility.