Key takeaways
• Water stress from AI data centre expansion threatens future infrastructure, as rising AI workloads increase cooling demand and strain already limited regional water supplies.
• Adopting strategies to reduce AI-related water use, such as liquid‑cooling and on‑site recycling, can cut consumption while maintaining performance.
• Companies are starting to pursue more selective and nuanced sustainability strategies, according to C-suite executives at the BNP Paribas TIME Conference 2026.
• Policy frameworks for AI water-resource reporting are emerging, urging transparent disclosure of withdrawals, recycling rates, and pricing, which helps investors assess water‑risk exposure.
In their latest report, the BNP Paribas Markets 360 team explains why water stress from AI data centre development is becoming a critical issue for the digital‑infrastructure sector and what it means for long‑term sustainability.
Understanding water stress from AI data centre expansion
The AI boom is already reshaping global water demand. Total water consumption across the AI supply chain is projected to climb to roughly 1,200 billion litres by 2030, up from about 560 billion litres in 2023 (IEA 2025). Two‑thirds of that use is linked to primary energy supply, while cooling accounts for roughly a quarter.
Geographic constraints that limit water supply
Unlike electricity, water cannot be generated on demand; it is bound by geography, hydrology and climate. Many existing hyperscale sites sit in regions where renewable surface and groundwater supplies are already strained. These structural constraints make scaling water supply more difficult than expanding power capacity. As water stress intensifies, Markets 360 analysts expect water governance to tighten across major markets. This could translate into stricter permitting requirements, higher water pricing, and greater transparency requirement around water consumption and recycling practices.
How cooling demands amplify the stress
Data centre cooling is highly seasonal, often peaking in the hottest months when water is scarce. In 2023, US data centres directly consumed about 66 billion litres of water for cooling, roughly the annual usage of the city of San Francisco. As AI workloads become more compute‑intensive, heat output rises and so does the need for water‑heavy cooling solutions. Some regions such as the UAE are relying on desalinated water, as the corrosive properties and mineral content of raw seawater could damage electrical equipment. However, desalination itself carries a significant energy footprint. Thermal desalination technologies still dominate in the Gulf, and these systems primarily rely on natural gas.
Strategies to reduce AI-related water use
Addressing the problem of water stress from AI data centre expansion requires technical and operational shifts that cut water withdrawal without sacrificing performance.
Optimising liquid cooling technologies
Higher AI server rack densities are accelerating the shift away from traditional air cooling toward liquid-based systems that can manage higher heat loads more efficiently. This transition can also help reduce water consumptions, as liquid cooling lowers reliance on water-cooled chillers and evaporative cooling towers.
| Cooling technology | Water use profile |
| Air cooling: Air conditioning, fans, and vents used to expel hot air and intake cool air; can have low water requirements if ambient air temperature is cold but can require cooling towers (see evaporative cooling) if not | Mixed |
| Evaporative cooling: Combines air cooling with open-loop water evaporation to cool air in data centre; about 80% of water drawn lost to evaporation, 20% discharged | High |
| Rear-door heat exchangers: Usually coupled with air cooling: air conditioning cools server room; water-cooled panels on individual servers capture waste heat and route to cooling tower or water chiller | Moderate |
| Direct-to-chip liquid cooling: Liquid pumped through microchannels directly on the chip’s surface to absorb heat; most commonly use pure water or a glycol mix in a closed-loop system to a chiller | Low |
| Immersion cooling: IT equipment submerged in dielectric fluid to dissipate heat | Low |
Water recycling and reuse for semiconductors
Semiconductor fabs can draw as much as 38 million litres of ultra‑pure water per day, the equivalent of 500 swimming pools (Interface). This is why recycling and reusing water could alleviate water stress from AI data centre expansion. For example, some manufacturers are increasing reclaimed water reuse and improving on-site treatment to reduce freshwater dependency. Regulatory pressures (e.g., higher water pricing, stricter permits) are also driving efficiency improvements.
Policy frameworks for AI water-resource reporting
Regulators are increasingly formalising water stewardship requirements, transitioning from voluntary disclosures to mandatory reporting obligations for companies in water-intensive sectors, including AI infrastructure.
Emerging global disclosure standards
The Taskforce on Nature-related Financial Disclosures (TNFD) and other emerging standards now require companies to publish detailed water‑withdrawal accounting, recycling ratios and downstream impact metrics. Aligning with these frameworks not only avoids penalties but also gives investors clearer insight into water risk exposure.
Tightening Permitting, Rising Water Pricing, and Transparency Mandates
Across major markets, permitting processes are tightening, water pricing is climbing and transparent reporting is becoming a legal prerequisite. For example, Taiwan’s seasonal water‑use surcharges was fully implemented in 2025, raising the effective cost of water for high‑volume users. Such measures incentivise firms to adopt water‑positive designs and to disclose their water footprints rigorously.
Contact the BNP Paribas Markets 360 team for the full “Positioning for AI’s water constraints” report or click on the button below if you’re a client.
| FAQ |
| What drives water stress from AI data centre expansion? Rapid AI workload growth increases cooling needs, and many sites are located in regions where renewable water supplies are already limited. |
| Which technologies help lower water consumption? Direct-to-chip liquid cooling, immersion cooling, and hybrid rear‑door heat exchangers can reduce water use compared with traditional evaporative systems. |
| How can organisations report their water use responsibly? By adopting emerging policy frameworks that require detailed water‑withdrawal accounting, recycling rates, and transparent public disclosures. |
| Are there regulatory trends that will affect AI infrastructure? Yes, tighter water‑governance, higher pricing and stricter permitting are expected across major markets, pushing companies toward more water‑positive operations. |
BNP Paribas does not consider this content to be “Research” for the purposes of the recast Markets in Financial Instruments Directive (2014/65/EU) (MiFID II). If you are subject to MiFID II, you should consider making your own assessment as to the characterisation of this content. Please see further our Legal notice for marketing documents, referencing to whom this content is directed.