The viral claim that every AI query drains "a bottle of water" has made its way from social media into boardroom conversations and even utility planning meetings. The reality is less dramatic at the individual query level, but far more consequential at the system level, especially for the water and sewer utilities that will actually feel the demand.
Here's what utility professionals need to understand about AI, data centers, and water, and what STAline is already seeing show up in the field as this trend accelerates.
Yes: artificial intelligence indirectly uses water. AI systems run in data centers that rely on cooling water to dissipate server heat, and on grid electricity whose generation often consumes water at thermoelectric power plants. AI itself doesn't require water to perform computations; it requires electricity, and both delivering that electricity and cooling the hardware that consumes it carry a water footprint.
Does AI use a lot of water, on a per-prompt basis? No. Recent published estimates put a typical text prompt's water footprint at a fraction of a milliliter to a few tens of milliliters, orders of magnitude below the "one bottle per query" claim that's circulated widely. But AI's share of data center electricity demand is significant and growing. For utilities, the real question isn't how many milliliters a single prompt consumes. It's whether the facility proposed for your service area will need 2 million gallons a day or 5 million, and what that does to your system during a July peak.
Data centers use water for cooling and for the power generation that runs them, not to process AI workloads directly. The water enters the picture at two points: on-site thermal management and off-site electricity generation.
On-site, servers packed into racks produce concentrated waste heat, and AI-optimized chips generate especially intense heat during processing. Common cooling approaches include:
Water-cooled systems can recycle water many times before discharge, but each cycle loses volume to evaporation and concentrates dissolved solids in the remaining stream. For utility operations, this means both increased demand for potable or non-potable supply, and additional wastewater or blowdown flows carrying elevated total dissolved solids and temperature, flows that may require pretreatment before entering the collection system.
Direct on-site water consumption is what a data center withdraws for its own cooling equipment: an estimated 17 to 20 billion gallons of fresh water per year across U.S. data centers in recent years. Evaporative cooling towers account for the bulk of that volume, with higher usage in hot, dry climates.
Indirect water use comes from electricity generation. Data centers account for a meaningful and growing share of U.S. electricity consumption, and thermoelectric power plants use substantial water for cooling and steam production. Because most of a data center's total water footprint traces back to the power plants generating its electricity rather than the facility itself, total water impact, direct plus indirect, runs well above the direct-cooling figure alone.
Renewable generation (wind, solar) carries far lower water intensity than fossil or nuclear thermoelectric generation. So the honest answer to "is AI powered by water?" depends heavily on the local grid mix and the facility's power purchase agreements. A data center running largely on solar has a very different indirect water profile than one drawing from a coal-heavy grid.
Utilities should weigh both sides: local system impacts from on-site withdrawal and discharge, and regional watershed impacts tied to the power mix serving new data centers.
PUE (Power Usage Effectiveness) equals total facility power divided by IT equipment power. Efficient hyperscale facilities target PUE values in the 1.2 to 1.4 range. Lower PUE means less energy wasted on cooling and power distribution, which reduces both electricity use and the indirect water use tied to it.
WUE (Water Usage Effectiveness) measures liters or gallons of water per kilowatt-hour of IT load. Disclosed WUE figures vary significantly by operator and facility, and many disclosures still omit indirect water consumed at power plants, so utilities should ask developers to clarify exactly what a quoted WUE figure includes.
Water transfers heat more efficiently than air, which is why many operators still favor water-based cooling. But AI-optimized facilities are adopting higher rack densities that push beyond what traditional air cooling can handle, driving adoption of newer approaches like immersion cooling, where servers are submerged in a non-conductive liquid. These systems carry different water and sewer implications than legacy air-cooled halls. Utilities should ask developers for PUE and WUE targets by build-out phase, and how those figures may shift as liquid cooling adoption increases.
U.S. data centers' water footprint, direct and indirect combined, is substantial and accelerating, driven largely by AI workload growth. Regional clustering compounds the effect:
A single 50 to 100 MW campus can place demand on a system comparable to a new small town.
The real question isn't "why does AI use so much water." It's "what does high-density AI development mean for my system's capacity, reliability, and rates?"
On the water side, AI workloads demand continuous heavy power, translating to sustained cooling loads. New data centers can drive peak-day demands that interact with existing industrial and residential loads, narrowing operating margins on the hottest days. Utilities may need dedicated mains, additional withdrawal capacity, or non-potable supply infrastructure to serve these facilities without degrading service to existing customers.
On the wastewater side, cooling tower blowdown introduces concentrated streams with elevated TDS, chemical treatment residuals, and higher temperatures, often requiring pretreatment before entering the collection system.
Planning implications are direct: master plans, hydraulic modeling, and capital improvement programs may need revision when a hyperscale data center or AI cluster is proposed. But there's an opportunity here too: long-term, creditworthy customers can support water infrastructure investment when rates and cost allocation are structured to match. State Revolving Fund (SRF) financing is one of the tools utilities have available when they need to upsize mains, expand treatment capacity, or add advanced treatment to serve this kind of growth, as we covered in our SRF explainer.
No single federal rule governs data center water reporting. Several states are filling that gap:
Virginia has moved furthest. New reporting rules require waterworks operators to categorize their water sales data, breaking out usage by data centers with state air permits, domestic users, and industrial/commercial users, split into potable and non-potable withdrawals. That data will be aggregated and made public on Virginia DEQ's website starting January 1, 2027. Separately, the state has directed DEQ to identify "Cooling Water Scarcity Areas" by July 2027; data centers located in those areas will need to show they've minimized cooling water use by 2032. It's a real regulatory shift, but a longer runway than the headlines sometimes suggest.
Texas, Georgia, Arizona, Ohio, and Iowa are each at varying stages of considering enhanced permitting thresholds, water availability assessments, or disclosure requirements for large data center projects.
These rules vary widely by state and even by watershed. Utilities should work directly with state agencies and local governments to understand how emerging regulation affects their specific service area, and shouldn't assume a rule that applies in Loudoun County, Virginia will look anything like one in Maricopa County, Arizona.
Consider a realistic scenario: an 80 to 120 MW hyperscale data center campus proposes connecting to your system within a 3-to-5-year window. Key planning questions utilities should be asking:
Standard residential or commercial rate tiers rarely fit a customer that may use more water in a day than a neighborhood uses in a month. Options include industrial large-user tiers, system development charges scaled to projected demand, and infrastructure cost-sharing agreements.
And as utilities plan for the pipe, valve, meter, and distribution infrastructure this kind of build-out demands, having a reliable wholesale partner with local inventory and national reach becomes part of that planning conversation, not an afterthought. STAline works with contractors and municipalities across the country supplying the underground utility and waterworks products these projects require, and we're already seeing this kind of demand show up in what our customers are asking for.
Coordination with electric utilities is essential too. Water and sewer planning should align with grid expansion timelines, since delays on one side can strand investments on the other. The AWWA's "Cooling the Cloud" framework offers structured guidance for these conversations around efficiency, reuse, and shared responsibility for long-term resilience.
Reducing data centers' reliance on clean freshwater is both feasible and increasingly expected of developers. Recycled and reclaimed wastewater use for cooling is becoming more common, particularly in water-stressed regions, and some facilities are shifting toward air-cooled or direct-to-chip liquid cooling designs that minimize evaporative losses.
Opportunities for integrated planning include co-locating new data centers near water reclamation plants, designing blowdown return pipelines, and negotiating data-sharing agreements so utilities can monitor actual water use against projections. Most major developers are open to these conversations when utilities approach them early, before the permit application, not after.
AI-driven data center growth intersects with infrastructure challenges that pre-date the AI boom entirely: aging treatment plants, storage shortages, and distribution constraints. Digital infrastructure growth can help justify upgrades that were already needed, provided the cost burden is allocated fairly between existing customers and new large users.
AI isn't going away, and the water and energy demands of running it will keep growing. Utility leaders who understand how data centers actually use water, how efficiency metrics like PUE and WUE work, and how to plan and rate for this kind of large-load growth will be better positioned to protect their water resources, maintain affordability, and capture the economic benefits headed for their communities. Start the conversation with developers, regulators, and your supply partners before the permit application arrives.
Is AI really "powered by water," or is that just shorthand?AI is electrically powered. The "water-powered" framing is shorthand for the indirect link: more AI compute means more electricity generation, which can mean more cooling water at power plants. The strength of that link depends entirely on the regional grid mix. A utility whose service area draws mostly from natural gas or nuclear generation will see a different upstream water impact than one served by wind and solar.
Does AI's water consumption show up as a noticeable supply problem for utilities?For many systems, a single modest data center won't create a visible supply problem. But clusters of large hyperscale campuses in water-stressed areas can materially affect peak-day demand and storage requirements, as Georgia's Fayette County discovered when a data center's unmetered water use went undetected until residents reported low pressure. Scenario modeling that accounts for plausible data center growth helps utilities anticipate when new wells, treatment capacity, or interconnections will be needed.
How should utilities respond when a developer asks for reclaimed or non-potable water?Serving data centers with reclaimed water can reduce demand on potable supplies, but only if the utility has sufficient treatment capacity, distribution infrastructure, and clear reliability standards in place. Utilities should evaluate seasonal flow patterns, storage needs, and potential impacts on downstream users before committing reclaimed water to a long-term contract, and should spell out backup supply and cost-sharing responsibilities explicitly in any agreement.
Are AI-heavy data centers different from "regular" data centers from a utility perspective?AI-heavy facilities often run higher rack densities with more continuous utilization, changing both cooling strategy and water profile. Traditional enterprise data centers might average 5 to 10 kW per rack; AI-optimized facilities can exceed 40 kW per rack, often requiring immersive or direct-to-chip liquid cooling rather than conventional air cooling. Utilities should request detailed cooling and process descriptions during the planning phase for any large proposed facility.
What role can utilities play in managing this?Utilities can influence outcomes by setting transparent industrial rate structures, offering reclaimed water where feasible, and building water-efficient design expectations into connection policies and development agreements. Participation in state policy discussions about disclosure and siting is equally important, as is working with experienced supply partners who understand the scale and pace this kind of infrastructure demands.
ALABAMA CUSTOMERS: See important notice regarding your state's sales tax.
"Seller has collected the simplified sellers use tax on taxable transactions delivered into Alabama and the tax will be remitted on the customer's behalf to the Alabama Department of Revenue.Seller's program account number is SSU-R012660449."
