The artificial intelligence race has created an infrastructure expansion unlike anything the digital economy has experienced since the birth of hyperscale cloud computing. Companies now compete to deploy increasingly powerful computing clusters, while governments encourage investment through incentives designed to attract billions of dollars in new facilities. Most discussions surrounding this unprecedented construction wave revolve around electricity demand, semiconductor supply chains, and grid modernization because those subjects dominate headlines and investor conversations. Water, by comparison, remains largely invisible despite serving as an essential operating resource behind nearly every advanced computing workload. The result is a growing disconnect between how AI infrastructure is discussed publicly and how it actually functions once servers begin processing increasingly complex models. Understanding that hidden dependency has become essential because water is emerging as one of the defining constraints shaping the industry’s long-term expansion.

AI Infrastructure Extends Beyond Electricity

Electricity represents only the visible portion of AI’s infrastructure requirements because every megawatt consumed eventually produces heat that operators must remove safely. Modern graphics processors generate far higher thermal loads than traditional enterprise servers, forcing operators to deploy increasingly sophisticated cooling strategies capable of maintaining stable operating conditions around the clock. Those cooling decisions determine how much water facilities consume directly, yet they also influence how much electricity the buildings ultimately require throughout their operating life. Moreover, every additional unit of electricity introduces another layer of water demand because large portions of power generation still depend on water-intensive cooling processes. Looking only at water flowing inside a data center therefore captures only part of the industry’s environmental footprint. The larger story begins outside the building, where power plants quietly expand AI’s water demand far beyond what facility operators report publicly.

Direct water consumption attracts public attention because cooling towers visibly evaporate water into the atmosphere during periods of heavy operation. Yet indirect water consumption linked to electricity production represents an even larger challenge because thermal power stations also require enormous volumes of water to remove excess heat from generating equipment. Fossil fuel plants, nuclear facilities, geothermal generation, and hydropower each carry different water profiles, while wind and solar require comparatively little water once operational. Lawrence Berkeley National Laboratory estimates that US data centers directly consumed approximately 17.4 billion gallons of water during 2023, while electricity generation supporting those facilities required another 211 billion gallons during the same period. That relationship demonstrates why reducing cooling water alone cannot eliminate AI’s broader resource demands. Every new computing cluster therefore influences regional water systems through both its own operations and the energy infrastructure supporting continuous computational activity.

Consumption and Withdrawal Tell Different Stories

Public debates frequently combine water withdrawal and water consumption as though both measurements describe the same environmental outcome. They do not because withdrawal measures the amount of water removed from a source, whereas consumption reflects water effectively lost through evaporation or other processes before returning to the surrounding ecosystem. Water withdrawn but later discharged may become available again after treatment, although temperature changes and chemical alterations can still affect aquatic environments and downstream users. This distinction matters because facilities with similar withdrawal figures may produce very different long-term impacts depending on their cooling technologies and local environmental conditions. Policymakers increasingly recognize that focusing on only one metric risks understating broader ecosystem pressures. Investors evaluating infrastructure projects also need both measurements to understand how facilities interact with local watersheds over decades rather than individual operating cycles.

Every data center ultimately performs the same task by transferring heat away from computing equipment before releasing it safely into the surrounding environment. Operators frequently rely on evaporative cooling towers because they remove heat efficiently during periods of elevated outdoor temperatures while helping facilities control electricity costs. Dry cooling systems reduce or nearly eliminate direct water consumption by using fans and ambient air, yet those systems generally require additional energy and often lose efficiency as temperatures climb. Some operators combine both approaches by relying primarily on dry cooling before introducing evaporative techniques during hotter periods to maintain reliable equipment performance. Site-specific factors including climate, electricity pricing, environmental regulations, and available water resources influence which combination ultimately delivers the most practical outcome. Consequently, no universal cooling strategy exists because every location presents a different balance between operational efficiency, energy consumption, infrastructure cost, and long-term environmental resilience.

Computing Efficiency Does Not Guarantee Water Efficiency

Technological progress continues improving processor performance, rack density, and overall computational efficiency across successive AI hardware generations. Those gains understandably create expectations that future facilities will automatically consume less water while delivering more computing capacity. Research suggests the reality remains considerably more complicated because cooling performance depends on multiple interacting variables rather than server efficiency alone. Lawrence Berkeley National Laboratory found that water use per unit of computing work can vary by more than 10,000 times depending on cooling architecture, weather conditions, operating practices, and equipment configuration. Such variation illustrates why headline efficiency improvements cannot accurately predict water demand across different facilities. Infrastructure planners therefore increasingly evaluate entire system designs instead of focusing exclusively on individual hardware improvements when estimating future resource requirements.

The AI sector has become remarkably sophisticated in reporting carbon targets, renewable energy procurement, and operational efficiency, yet comparable visibility into water use remains inconsistent across much of the industry. Most operators disclose little information about individual facilities, leaving local governments, researchers, and nearby communities with limited insight into how projects may influence regional water supplies throughout the year. Annual sustainability reports provide useful snapshots, but they rarely capture seasonal fluctuations that occur when summer temperatures increase cooling demand at the exact moment local water systems experience their greatest stress. Some municipalities have also entered confidentiality agreements that prevent disclosure of facility-level water consumption, making independent analysis increasingly difficult. Microsoft Corp., Alphabet Inc.’s Google, and Meta Platforms Inc. publish some of the industry’s most detailed site-level water reporting, although even those disclosures remain limited compared with the operational data available internally.

Location Decisions Now Carry Greater Environmental Weight

The environmental consequences of a data center depend as much on geography as engineering because identical facilities can produce very different outcomes under different regional conditions. Projects built in water-abundant areas may integrate into existing resource systems with relatively manageable impacts, while developments in drought-prone regions introduce additional pressure into already constrained watersheds. Research has shown that a significant share of recently announced or developing facilities are located in areas already experiencing elevated levels of water stress, creating new competition among municipalities, agriculture, industry, and digital infrastructure. Alex de Vries-Gao, a researcher at VU Amsterdam, warned that data centers could “make it worse at the worst possible time.” He also noted that increasing competition for finite water resources can raise costs while reducing long-term availability for surrounding communities. Those concerns become increasingly relevant as developers search beyond traditional metropolitan markets for larger sites capable of supporting next-generation AI campuses.

The discussion surrounding AI infrastructure often treats electricity and water as separate challenges even though both remain deeply interconnected throughout the entire computing ecosystem. Every increase in computing capacity raises electricity demand, and much of today’s generation portfolio still relies on water-intensive cooling technologies despite continued renewable energy expansion. BloombergNEF projects that data centers could increase from roughly 5.9% of US electricity consumption today to around 20% by 2035, illustrating how rapidly AI may reshape national energy demand. That projection also suggests indirect water requirements could continue rising even if future facilities become substantially more efficient at cooling their own servers. The industry’s resource footprint therefore extends far beyond the boundaries of individual campuses into transmission systems, generation fleets, and regional water infrastructure. Evaluating AI’s sustainability through only facility-level metrics risks overlooking the much larger environmental relationship connecting computing, electricity production, and water availability.

Communities Are Demanding Greater Accountability

Public acceptance can no longer be assumed simply because AI facilities promise employment opportunities and local tax revenue. Residents increasingly ask how proposed developments will affect drinking water supplies, agricultural activity, environmental resilience, and future infrastructure planning before supporting new construction. Carbon Direct reported that at least 46 US data center projects worth approximately $170 billion experienced delays or cancellations following community opposition between January 2024 and May 2026, demonstrating that local concerns now influence billion-dollar investment decisions. Environmental organizations have also pursued legal action in several cases to obtain greater access to water-use information for proposed developments. Rural communities deserve clear explanations regarding long-term infrastructure impacts before projects move from planning into construction because many lack extensive regulatory resources to perform independent technical assessments. Community engagement has therefore become a strategic requirement rather than a communications exercise for developers seeking durable project approvals.

Regulators have started recognizing that AI infrastructure requires oversight extending beyond traditional planning approvals and electricity connections. Utah now requires developers to disclose projected water usage for new server farms, reflecting growing interest in improving transparency before facilities begin operations. New York Governor Kathy Hochul announced a temporary pause on new data centers while the state develops a regulatory framework addressing electricity costs and water-related impacts associated with future expansion. Southern Nevada has prohibited evaporative cooling in new commercial and industrial buildings because of the technology’s intensive water requirements, signaling that cooling choices themselves may become regulatory considerations. Across Europe, operators must report energy-efficiency metrics including water usage under new reporting requirements, although facility-level information remains confidential and only aggregated national and European Union data becomes publicly available. These developments indicate policymakers increasingly view AI infrastructure as critical public infrastructure requiring more comprehensive governance than previous generations of digital facilities.

Technology Alone Will Not Solve the Challenge

Major technology companies continue investing heavily in solutions designed to reduce operational water demand without slowing AI deployment. Amazon.com Inc. has committed to becoming water positive by 2030 through watershed restoration projects, well refurbishment initiatives, and investments intended to return more water to communities than the company withdraws. Nvidia Corp. introduced a new AI server capable of operating cooling liquid at higher temperatures, reducing cooling requirements while improving infrastructure efficiency for customers deploying advanced computing clusters. Operators also continue expanding recycled water use, dry cooling technologies, improved server architectures, and localized infrastructure upgrades where suitable supplies exist. Those innovations represent meaningful progress, yet they primarily address direct operational consumption rather than the substantially larger indirect footprint associated with electricity production. Sustainable AI infrastructure will ultimately require coordinated improvements across computing hardware, cooling systems, energy generation, and regional resource planning instead of isolated technological breakthroughs.

AI’s future will depend not only on access to chips, electricity, and capital but also on how intelligently developers evaluate water as a strategic infrastructure resource. Selecting locations, cooling architectures, and energy sources requires balancing environmental resilience, operational reliability, community expectations, and long-term economic performance rather than optimizing a single metric. Researchers continue arguing that standardized, continuous facility-level reporting would provide policymakers, utilities, investors, and citizens with the information needed to measure genuine progress across the sector. Microsoft recently joined Google and Meta in providing more granular water-use metrics, reflecting gradual movement toward greater transparency, although researchers maintain that considerably more disclosure remains necessary. Finally, Alex de Vries-Gao summarized the industry’s reporting challenge by observing, “You don’t really know if things are getting better or worse if you’re not considering the full picture, and they’re only showing a tiny part. You don’t know the size of the iceberg.”