The massive scale-up of data center capacity now underway has drawn attention to the amount of water these facilities require to operate. Rystad Energy estimates that without water-saving measures in place, global water consumption by data centers could rise to 644 billion liters per year by 2030. With active mitigation, however, demand could be reduced to 388 billion liters. These figures represent worst- and best-case scenarios as data center capacity ramps up, driven largely by the growing adoption of artificial intelligence (AI) in everyday life. The stakes extend beyond water itself: data centers are already a significant and fast-growing consumer of electricity — the International Energy Agency estimated they accounted for roughly 1.5% of global electricity use in 2024 — so the sector's physical footprint is attracting scrutiny on multiple fronts.

Forecasting the water impact of data centers is far from straightforward, since consumption varies significantly depending on cooling technology, climate, and whether one measures water withdrawn or water actually consumed, according to Minh Khoi Le, Global Head of Data Center & Hydrogen Research at Rystad Energy. On Rystad Energy's estimates, data centers consumed 222 billion liters of water directly for cooling in 2025, but that figure could nearly triple to just under 644 billion liters by 2030 in the risked central case. More water-efficient pathways could bring the total down to 543 billion liters in the moderate case, or as low as 388 billion liters in an aggressive water-saving scenario. These figures capture only direct cooling-water consumption; the less visible water footprint embedded in the electricity supply adds another layer of complexity.

Technology variations

Data centers use water primarily to remove the heat generated by computing equipment, but consumption varies considerably with cooling technology and location. While AI servers generate substantially more heat, the rise of AI does not necessarily mean a proportional increase in water consumption: rack-level liquid cooling can reduce the heat burden on facility-level systems and enable greater use of dry cooling. Technologies such as those proposed for NVIDIA's Vera Rubin platform could therefore help limit water use, although the potential savings remain highly dependent on climate, with dry cooling generally more effective in colder locations. This matters because cooling choices made at the design stage effectively lock in a facility's water profile for its operating lifetime.

Beyond on-site water consumption, a useful metric is water-use effectiveness (WUE), the industry measure of water used per unit of IT energy, which can also account for the water consumed to generate the electricity a data center uses. Some cooling technologies reduce on-site water use but require more electricity. Dry cooling, for example, can save around 2.15 liters of water for every kWh of IT load but requires an additional 0.30–0.74 kWh of electricity. The indirect water footprint can be material relative to direct consumption, though its contribution varies with the water intensity of the electricity supply. In the United States, the indirect portion of a data center's water consumption can be more than twice its direct water consumption. A data center's overall water footprint therefore depends on both its on-site cooling system and the water intensity of the electricity it consumes.

Operator dimension adds another complication

Because there is no general requirement for WUE standards in most regions, design choices vary widely among operators, even within the same climate zones. Globally, Amazon Web Services (AWS) reported an average direct site WUE of 0.12 liters per kWh in 2025, Meta reported 0.19 liters per kWh for 2024, and Microsoft reported 0.27 liters per kWh for fiscal 2025. Digital Realty reported 0.59 liters per kWh and Equinix 0.91 liters per kWh in 2025. Google's disclosed data imply a materially higher intensity, although the company does not publish an official WUE figure.

These values should not be read as a simple ranking, given differences in how leased and colocation facilities are treated and variation in regional portfolios. The underlying calculations also differ: AWS and Meta define site WUE as water withdrawal per IT capacity, as opposed to water consumption. AWS's own 2025 regional values ranged from 0.02 liters per kWh in Stockholm to 2.85 liters per kWh in Jakarta — a spread of more than 100 times within a single operator's reporting framework, underscoring how strongly geography and cooling architecture influence the metric.

Major data center operators, including the hyperscalers AWS, Google, Microsoft and Meta, have committed to using water more sustainably. All four have set targets to become "water positive" by 2030, meaning they aim to replenish more water than they consume. They are working on projects to restore freshwater supplies and improve water efficiency, including in sectors such as agriculture. These efforts vary by location and watershed, however, with greater focus often placed on areas facing high water stress, so the impact on local water resources will vary.

Regulation is emerging, but reporting still leads enforcement

WUE regulation has started to emerge, but often stops at reporting requirements rather than performance mandates. The EU Commission is planning a Data Center Energy Efficiency package that would introduce a labeling system and performance standards, while Singapore's Green Data Center Roadmap aims for less than 2 cubic meters per megawatt-hour (MWh), equivalent to 2 liters per kWh. Regulations are developing in the United States but vary widely by state. Texas, for example, recently imposed a moratorium on new approvals pending audits of the tax breaks, power, water, and cooling use associated with data centers.

Water stress will also be an increasingly important consideration for data center regulation, as water demand is set to grow in regions already facing significant constraints on availability. Regions with high and extremely high water stress are projected to account for 34% of the data center sector's total global direct water consumption by 2030. Requiring the adoption of the least water-intensive cooling technologies could reduce consumption in these regions by 45%, highlighting the potential impact of technology standards in water-stressed areas. Some of the most exposed areas globally include Jamnagar and Thane in India and Reeves County in Texas, where data center water consumption is high relative to local water stress. The direction of travel for the industry will be visible in upcoming milestones: whether the EU's efficiency package translates labeling into binding performance standards, whether Singapore's WUE ceiling becomes a template for other jurisdictions, and whether operators' 2030 water-positive commitments are met as AI capacity expands.

By Rystad Energy