The next cooling decision may begin long before a chiller is selected, because the most consequential question for a new data center is increasingly whether the site can afford to consume water at all. Power has traditionally dominated infrastructure discussions because electricity determines how much compute a site can support, how quickly capacity can come online, and how much the operator must spend to keep that capacity productive. Water introduces a different constraint because cooling demand interacts directly with local supply conditions, competing users, utility infrastructure, environmental rules, and the physical characteristics of the surrounding watershed. A site can have attractive power availability and still carry a thermal strategy that becomes difficult to defend when drought conditions tighten or local water systems face competing demand.

Water does not behave like an ordinary utility input in a data center because its availability depends on a physical geography that cannot be expanded as easily as electrical infrastructure. Electrical capacity can be expanded through additional generation, transmission, or grid infrastructure, while water availability remains constrained by the physical conditions and infrastructure of the source serving a site. Cooling towers make that distinction particularly important because their basic heat-rejection mechanism relies on evaporation, which means part of the cooling system’s water inventory leaves the site as vapor rather than simply circulating indefinitely. The U.S. Department of Energy describes evaporation as the primary mechanism through which cooling towers reject heat, while also identifying blowdown, drift, and leakage as additional pathways for water loss.

Why Water Has Quietly Become Data Centers’ Next Balance-Sheet Risk

Water risk can affect a data center through operating, infrastructure, regulatory, and community-related considerations that are increasingly relevant to project planning and site selection. The first is direct operating exposure, where a cooling architecture depends on a continuous supply of water and therefore inherits the reliability characteristics of the local water system. The second is infrastructure exposure, because a project may require utility upgrades, treatment capacity, storage, discharge arrangements, or alternative supply connections before its cooling design can operate as intended. The third is regulatory exposure, since water withdrawals, discharges, treatment systems, and local water conditions can affect permitting and operating requirements. A fourth channel comes through reputation and stakeholder acceptance, particularly when the same water source supports households, agriculture, industry, or ecosystems that face seasonal or structural pressure.

The Numbers Behind the Panic

Water analysis should not rely on a single consumption figure because cooling performance and water requirements vary with climate, operating conditions, heat-rejection equipment, water chemistry, and IT load. WUE provides an operational lens by relating site water use associated with cooling and humidification to IT energy use, which allows operators to track how changes in thermal architecture affect water intensity over time. That metric becomes more informative when the operator separates withdrawal from consumption, identifies the source of the water, and understands what portion returns to a treatment system or local watershed. Cooling-tower operation creates a particularly important distinction because evaporation represents the intended heat-rejection mechanism, while blowdown removes concentrated dissolved solids from the circulating water and therefore creates another water-management requirement.

Regional context then becomes critical because water stress is not a property of the data center itself but of the relationship between local demand, available supply, seasonal variability, water quality, and competing uses. WRI’s Aqueduct platform evaluates indicators such as baseline water stress, water depletion, variability, and other dimensions of physical and regulatory risk, while also warning that broad indicators should be supplemented with local analysis. That limitation matters for data center planning because a regional water-stress classification can identify a warning signal without answering the engineering questions that determine whether a particular site can operate reliably. An operator still needs to understand the source basin, utility capacity, drought response rules, treatment requirements, discharge pathways, groundwater dependence, and the potential for competing demand to change over the asset’s operating life.

From ESG Checkbox to Site-Selection Criterion

Site selection also needs to account for the fact that cooling technology can change the meaning of a water constraint. Direct-to-chip systems can use a closed secondary loop that isolates IT cooling from the facility water system, allowing the heat to move through controlled fluid circuits before reaching the final heat-rejection equipment. Air-side economization can reduce mechanical cooling demand when outdoor conditions fall within the required operating envelope, while dry coolers can reject heat without the evaporative loss associated with open cooling towers. Immersion systems can remove the need for air movement across the IT equipment and can transfer heat through dielectric fluid to a dedicated heat exchanger, creating another route toward low-water operation.

Path One — Direct-to-Chip Liquid Cooling

Direct-to-chip liquid cooling changes the thermal pathway at the point where heat is generated, making it a practical route toward lower-water data center designs when paired with an appropriate facility-side heat-rejection system. Conventional air cooling first moves heat from processors and other components into the surrounding air, after which the facility cooling system must remove that heated air and ultimately reject the heat outdoors. Direct-to-chip cooling instead places a cold plate directly against major heat-generating components and uses a controlled liquid loop to capture heat before it spreads through the room. The approach does not automatically create a zero-water cooling system because the downstream heat-rejection system determines whether the captured heat is ultimately rejected through evaporation, dry cooling, mechanical refrigeration, or another method. .

The appeal becomes clearer when the thermal chain is examined as a series of interfaces rather than as a single cooling machine. A cold plate extracts heat from the processor package, the technology cooling loop transports that heat toward a coolant distribution unit, and the CDU transfers the heat into the facility-side cooling system or another heat-rejection pathway. This architecture gives operators greater control over coolant temperature, flow, pressure, filtration, and isolation because the IT-side loop can remain physically separated from the facility loop. ASHRAE identifies direct component liquid cooling as a technology that delivers cooling media directly to datacom equipment and notes that such systems require dedicated piping, heat exchangers, and related components between the IT cooling system and facility climate-control infrastructure.

Single-Phase vs. Two-Phase Systems

Single-phase direct-to-chip cooling uses a liquid that remains in the liquid state as it moves through the cold plate, absorbs heat, and returns toward the heat exchanger or CDU. The architecture resembles a conventional hydronic loop in its basic physical behavior, but the design places the heat-transfer interface much closer to the semiconductor package and therefore reduces the distance that heat must travel through air before reaching the coolant. This makes the system relatively straightforward to integrate with established pumping, filtration, heat-exchanger, monitoring, and control practices, particularly when the operator already understands closed-loop water or water-glycol systems. Current industry guidance identifies single-phase direct-to-chip cooling as a mature liquid-cooling approach for high-density AI and HPC systems, while two-phase approaches remain a more specialized architecture.

Two-phase direct-to-chip cooling takes a different approach by using phase change at the heat source, allowing the cooling fluid to absorb heat as it transitions between liquid and vapor. That mechanism can produce highly controlled thermal behavior because the phase-change process uses latent heat rather than relying solely on sensible temperature rise across the cold plate. The trade-off comes from the additional complexity of managing vapor, condensation, fluid chemistry, pressure, containment, and component compatibility within the cooling circuit. Two-phase direct-to-chip systems use phase change at the cooling interface and therefore require the system to manage both liquid and vapor states within the cooling architecture. The technology therefore becomes less a question of whether phase change works and more a question of whether the workload and chip architecture justify the additional engineering burden.

Retrofit Economics vs. Greenfield Design

Direct-to-chip cooling can be introduced selectively during a retrofit rather than requiring an existing data hall to abandon its air-cooling infrastructure at once. Existing air-cooled environments can add CDUs, liquid distribution piping, rack connections, cold plates, monitoring, leak detection, and heat-exchange capacity while retaining air cooling for equipment that does not yet require liquid treatment. Vertiv describes this hybrid path as a common deployment model because operators can introduce liquid cooling into existing environments without abandoning the broader cooling infrastructure that already supports the site. The engineering challenge lies in determining whether the existing heat-rejection plant can absorb the new liquid-cooled load without creating a bottleneck elsewhere in the thermal chain. Electrical distribution, floor loading, rack layouts, overhead piping, maintenance clearances, redundancy, water quality, and emergency procedures can all influence the cost and feasibility of the retrofit.

Greenfield design offers a different economic equation because the cooling architecture can influence the building from the earliest engineering stage instead of fitting inside an existing envelope. A new site can reserve space for CDUs, establish liquid distribution corridors, design appropriate service clearances, select compatible heat-rejection equipment, and coordinate electrical and mechanical systems around the expected compute density. The design team can also choose higher supply and return temperatures where the IT equipment permits them, which can improve the conditions under which heat can be rejected without intensive mechanical refrigeration. ASHRAE’s AI guidance highlights the value of warm-water liquid cooling because higher supply and return temperatures can support more efficient heat rejection and create opportunities for useful heat recovery. That flexibility can become important for zero-water designs because a higher-temperature liquid loop can make dry or economized heat rejection more practical across a wider range of operating conditions.

Who’s Deploying This at Scale

The movement toward direct-to-chip cooling is no longer confined to specialized research environments because current AI infrastructure increasingly incorporates liquid cooling into rack-scale system designs. NVIDIA’s GB200 platform, for example, is designed as a rack-scale AI system with liquid-cooled racks, showing how thermal management has become part of the architecture of high-performance compute rather than a downstream facility decision. NVIDIA’s newer reference work around Vera Rubin systems also emphasizes continuity in liquid-cooling architecture, including common CDU and rack-drop interfaces that can support successive generations of hardware. That approach matters because operators need thermal infrastructure that survives hardware refresh cycles rather than becoming obsolete when processor configurations change.

Microsoft provides another important example because its newer AI-optimized data center design connects direct-to-chip cooling with a closed-loop thermal architecture intended to eliminate evaporative water use for cooling. The design recirculates water between servers and chillers rather than continually drawing fresh water to replace evaporation, creating a clear separation between the coolant circulating inside the technical system and the water consumed by conventional evaporative heat rejection. Microsoft’s broader water strategy also shows that the company is combining cooling-system redesign with alternative water sources and operational controls rather than relying on one technology to solve every site-specific problem. That combination is important because direct-to-chip cooling can remove a major source of water consumption while still leaving other water needs elsewhere in the site.  

Path Two — Immersion Cooling and the Case for Zero-Evaporation Infrastructure

Immersion cooling takes the liquid-cooling argument one step further by removing the air interface from the primary heat-transfer process and placing the computing hardware directly inside a thermally conductive dielectric fluid. Instead of moving heat from chips into air, then from air into a facility cooling system, immersion allows the fluid surrounding the equipment to capture heat across a much larger surface area and transport it toward a heat exchanger. That architecture can reduce dependence on fans, room-level air movement, and conventional computer room air-conditioning equipment, which changes the mechanical design of the data hall as much as it changes the cooling method itself. Single-phase immersion keeps the dielectric fluid in a liquid state throughout operation, while two-phase immersion deliberately uses boiling and condensation to move heat through a phase-change cycle.

The attraction becomes particularly clear when the physical limits of air cooling begin to shape the design of high-density computing environments. Air has relatively poor heat-transfer characteristics compared with engineered liquids, so moving more heat through an air-cooled rack requires greater airflow, more heat-transfer surface, tighter thermal management, or some combination of the three. Immersion changes that relationship by allowing the coolant to contact the equipment directly, eliminating several thermal resistances that exist between a processor and the surrounding room air. Intel’s work with Shell, Supermicro, and Submer demonstrates this architecture in a certified single-phase system, with the dielectric fluid carrying heat away from immersed equipment and transferring it through a heat exchanger to another cooling circuit. The resulting system can remove server fans and reduce reliance on room-level air-conditioning infrastructure, but the benefit does not mean that every data center should immediately replace direct-to-chip or air cooling with immersion.

Single-Immersion vs. Two-Phase Immersion

Single-phase immersion places servers or selected computing components inside a dielectric liquid that remains in a liquid state throughout normal operation, with pumps moving the warmed fluid toward a heat exchanger before returning the cooled fluid to the immersion tank. The architecture keeps the thermal cycle relatively intuitive because the fluid does not intentionally boil inside the tank, which simplifies containment and allows operators to use established pumping and heat-exchange principles. Submer describes single-phase immersion as an open-bath approach in which the coolant remains in contact with the hardware and moves through a heat-exchange system that removes the captured heat. Because single-phase immersion keeps the working fluid in a liquid state during normal operation, its service procedures differ from those required by two-phase systems, while fluid handling, equipment removal, contamination control, and component compatibility remain important considerations.

Two-phase immersion uses a dielectric working fluid that boils when it absorbs heat from the immersed equipment, with the resulting vapor rising toward a condenser where it releases heat and returns to liquid form. The phase change creates an efficient heat-transfer mechanism because the fluid can absorb substantial thermal energy while changing state, allowing the system to manage concentrated heat without depending entirely on a large sensible temperature rise through the liquid. The architecture also creates a more enclosed thermal environment because the vapor must remain contained and the condenser must continuously return the fluid to the tank. Submer distinguishes two-phase systems from its single-phase approach by noting that two-phase immersion requires a more complex access and maintenance environment because the fluid operates across liquid and gas states. That architecture requires operators to consider vapor containment, condensation, fluid management, condenser performance, pressure behavior, and the properties of the working fluid.

Density Gains and the AI Workload Connection

AI changes the immersion discussion because the thermal problem increasingly follows the concentration of computing rather than the simple number of servers occupying a room. Accelerators generate substantial heat within tightly packed assemblies, and that concentration can push conventional air-cooling arrangements toward increasingly demanding airflow and heat-removal requirements. Immersion attacks the problem at the equipment boundary by surrounding the heat-generating components with a fluid that can capture thermal energy directly, reducing the need to transport that heat through room air before it reaches the cooling plant. Intel’s validated single-phase immersion solution highlights this capability by describing the dielectric fluid as substantially more thermally conductive than room-temperature air and linking that property to higher rack power density and reduced reliance on fans and air-conditioning equipment.

The density advantage also changes the physical relationship between compute and building infrastructure because reducing air movement can affect the layout of the entire thermal plant. An air-cooled room needs distribution paths that move conditioned air toward equipment and return heated air toward the cooling system, while an immersion environment can concentrate heat transfer within tanks and associated fluid circuits. That shift can reduce the importance of raised-floor airflow patterns and room-scale air distribution, although it introduces requirements for tank placement, structural support, fluid containment, lifting arrangements, and service access. Research examining immersion cooling identifies higher computing density and potential reductions in cooling infrastructure as advantages, while also identifying maintenance and reliability considerations that operators need to evaluate. The implication for AI deployments is not that physical space disappears, but that the space can be reorganized around thermal equipment that behaves differently from conventional server racks.

Barriers to Adoption — Cost, Fluid Supply Chains, Serviceability

The largest barrier to immersion adoption is not whether the fluid can remove heat, because commercial demonstrations have already established that basic technical capability, but whether operators can integrate the technology into a service model that matches the expectations of a mission-critical computing environment. Conventional servers are designed around human access, standardized rack procedures, air cooling, familiar component replacement, and broad availability of technicians and spare parts. Immersion changes that workflow because a technician may need to remove equipment from a tank, manage fluid exposure, control contamination, drain or isolate part of a system, and follow specific procedures before a component can return to service. The fluid itself becomes part of the operational environment, which means operators need to evaluate availability, storage, handling, material compatibility, maintenance, replenishment, and end-of-life management as part of the cooling-system design.

Fluid supply represents another technical consideration because the coolant becomes a critical material within the thermal system and must remain compatible with the hardware and cooling architecture. Different immersion fluids have different chemical properties, material compatibility requirements, environmental considerations, and handling characteristics, so operators need a procurement strategy that extends beyond the initial installation. Submer notes that dielectric fluids can include different chemical families and that fluid selection is fundamental to achieving the intended thermal and sustainability outcomes, which means the choice should follow the equipment and operating environment rather than a generic preference for one fluid category. Operators also need to examine what happens when fluid must be removed, filtered, replaced, or transported, particularly when a deployment scales across multiple locations and requires consistent service procedures.

Path Three — Closed-Loop and Air-Side Economization

Closed-loop and air-side strategies offer a less radical route toward zero-water cooling because they can reduce or eliminate evaporative demand without requiring every server to sit inside a dielectric fluid. Air-side economization uses favorable outdoor conditions as the heat sink, allowing the cooling system to reduce reliance on mechanical refrigeration when the ambient environment falls within an acceptable operating envelope. Closed-loop mechanical systems take a different approach by circulating cooling fluid through heat exchangers and rejecting heat through dry or mechanically assisted equipment without intentionally consuming water through evaporation. Both strategies depend on climate, equipment operating limits, controls, filtration requirements where applicable, redundancy, and the operating temperatures of the cooling system. Microsoft’s recent explanation of its cooling portfolio illustrates the range of approaches, including direct-air systems that can operate with little or no cooling water under suitable outdoor conditions and hybrid systems that shift between evaporative and dry modes.

Climate-Dependent Free Cooling Strategies

Air-side economization works by treating outdoor air as a useful thermal resource when its temperature and moisture characteristics allow the data center to maintain acceptable IT conditions without relying fully on mechanical refrigeration. The approach can dramatically reduce water dependence because direct air cooling does not require an evaporative cooling tower to remove heat during periods when the outdoor environment already provides an effective heat sink. The challenge lies in the variability of that resource, because outdoor temperature, humidity, airborne contaminants, smoke, dust, and seasonal conditions can change the operating envelope from one location to another. Microsoft describes direct-air cooling as a strategy that can use little or no water, while also noting that hotter outdoor conditions can require a different cooling mode, which demonstrates why the architecture must be designed around transitions rather than ideal conditions.

The most robust air-side strategy combines economization with a cooling envelope that gives the facility enough flexibility to use outdoor conditions without compromising equipment reliability. ASHRAE’s thermal guidelines establish environmental classes and operating recommendations that help designers determine the temperature and humidity conditions equipment can tolerate, creating the foundation for assessing how often economization can operate effectively. Higher allowable coolant or air temperatures can also improve the conditions for dry heat rejection because the heat exchanger has a larger temperature difference available against the outdoor environment. AI workloads make this more complicated because dense accelerator systems increasingly favor liquid cooling, meaning air-side economization may remain useful for residual loads, building systems, or hybrid configurations rather than serving as the sole thermal strategy.

Closed-Loop Chiller Systems Replacing Cooling Towers

Closed-loop chiller systems can remove a major source of water dependence by replacing the open evaporative heat-rejection process with a contained fluid circuit and a dry or mechanically cooled final heat sink. In a conventional cooling-tower arrangement, water circulates through the tower and loses heat as a portion evaporates, which means the system must continually manage makeup water, concentration, blowdown, and water treatment. A closed-loop design keeps the primary cooling fluid within the circulating system and transfers heat through heat exchangers toward equipment that can reject it without routine evaporative water consumption. Microsoft’s AI-oriented data center design demonstrates the principle through chip-level cooling connected to a closed loop, where the same cooling fluid circulates between servers and chillers rather than being continually consumed through evaporation.

The trade-off appears in the electrical system because non-evaporative heat rejection can require additional fan, pump, or compressor energy depending on ambient conditions and system design. That relationship explains why water and power cannot be optimized independently, particularly when the operator moves away from an evaporative system specifically to eliminate water consumption. ASHRAE’s AI data center framework recognizes this interaction and points toward warm-water liquid cooling and efficient heat rejection as ways to improve the overall balance between thermal performance, energy use, and water consumption. Higher coolant temperatures can make it easier to reject heat without aggressive refrigeration because the facility-side system can operate closer to the outdoor heat sink. Variable-speed pumps, optimized controls, larger heat-transfer surfaces, and appropriate redundancy can further reduce the electrical penalty associated with non-evaporative heat rejection.

Trade-Offs — Power Draw vs. Water Savings

Cooling architectures that eliminate routine evaporative water consumption can shift more of the thermal-system burden toward electricity, equipment capacity, and mechanical infrastructure. Evaporative systems exploit the thermodynamic advantage of phase change to reject heat efficiently, while dry systems must rely more heavily on sensible heat transfer and mechanical equipment when ambient conditions are unfavorable. That difference can make a zero-water design more electrically demanding in some operating conditions, particularly when the facility cannot take advantage of economization or elevated coolant temperatures. The correct comparison therefore needs to consider the complete operating profile rather than comparing cooling equipment at a single design point. ASHRAE’s current AI data center guidance emphasizes system-level optimization because liquid cooling, heat rejection, supply temperatures, and facility energy performance interact closely.

The most effective zero-water designs therefore seek to avoid paying the full electrical penalty of dry cooling across every operating condition. Economization can carry the thermal load when outdoor conditions allow, while liquid cooling can capture heat close to the processor and deliver it at temperatures that improve the performance of downstream heat rejection. Closed-loop systems can then use variable-speed equipment and intelligent controls to adjust pumping and refrigeration according to actual workload demand rather than maintaining maximum cooling capacity continuously. Microsoft’s newer AI data center design illustrates this layered approach by combining chip-level cooling with a closed-loop system and warmer cooling conditions that can support more efficient heat rejection. The result is a design philosophy in which water savings come from reducing or eliminating routine evaporation while power performance depends on the thermal path, operating temperatures, equipment efficiency, and control strategy.

Path Four — Alternative Water Sourcing and Recycled-Water Loops

Not every data center will eliminate evaporative cooling, particularly where existing infrastructure, climate conditions, or operating requirements make water-assisted heat rejection technically and economically attractive. In those cases, the engineering objective shifts from eliminating water use entirely to ensuring that cooling does not compete unnecessarily with higher-value freshwater demand. Reclaimed water and other appropriately treated non-potable sources can separate cooling demand from potable supplies while allowing facilities to retain evaporative cooling where that architecture remains technically appropriate. This approach becomes especially relevant when an operator can connect the data center to an established reuse network or develop treatment infrastructure capable of producing water suitable for the site’s cooling requirements. The strategy requires more than identifying an alternative source because cooling systems impose water-quality requirements that can affect scaling, corrosion, biological growth, filtration, blowdown, and equipment life.

Non-Potable, Greywater, and Reclaimed Water Use

Reclaimed water can reduce dependence on potable supplies by using appropriately treated wastewater or other non-potable sources for cooling applications that do not require potable-quality water. The distinction is important because cooling towers do not need water for human consumption, but they do need water with characteristics that allow the system to operate without excessive scaling, corrosion, biological growth, or fouling. Water treatment can therefore become part of the cooling architecture, with filtration, chemical management, monitoring, and controlled blowdown supporting the thermal equipment. The U.S. Environmental Protection Agency recognizes water reuse as a means of substituting treated water for applications that do not require potable quality, while emphasizing that treatment and local requirements determine whether a particular reuse application is appropriate.

The technical challenge lies in the chemistry of the source because reclaimed water can contain constituents that behave differently from conventional potable supplies inside cooling equipment. Higher dissolved solids, organic matter, nutrients, residual treatment chemicals, or biological activity can alter the behavior of the circulating water and increase the risk of scaling, corrosion, or fouling if the system lacks appropriate treatment and monitoring. Cooling towers already require careful management of concentration cycles, blowdown, chemical treatment, and makeup water, and a non-potable source can make that management more important rather than less important. The Department of Energy’s cooling-tower guidance emphasizes water chemistry, cycles of concentration, blowdown control, and treatment as core elements of efficient tower operation. Operators considering reclaimed water therefore need a source-specific chemistry assessment and should design treatment equipment around the actual composition of the incoming supply rather than assuming that a utility reuse label guarantees suitability.

On-Site Water Recycling and Treatment Loops

On-site recycling can extend alternative sourcing by treating water within or near the data center’s operating boundary and returning suitable treated water to the cooling process rather than relying entirely on new supply. The treatment train can include filtration, softening, reverse osmosis, or other processes depending on the source-water chemistry and the requirements of the cooling system. Such a loop can reduce the volume of fresh water required to maintain cooling, but it introduces its own energy use, maintenance requirements, membrane management, concentrate handling, and equipment footprint. The operator therefore needs to evaluate the treatment system’s energy, chemical, maintenance, and residual-management requirements alongside the water savings rather than assuming that recycling automatically minimizes every resource impact. Cooling-tower optimization guidance from the Department of Energy highlights the importance of controlling blowdown and maintaining appropriate water chemistry, both of which directly influence how much makeup water the system needs.

The strongest treatment loops treat water quality as a continuously managed engineering variable rather than a periodic laboratory check. Sensors can monitor conductivity, pH, temperature, turbidity, and other parameters, while automated controls can adjust treatment or blowdown according to the actual condition of the circulating water. This becomes particularly useful when the incoming source varies because reclaimed-water systems may receive water with changing chemistry depending on upstream treatment and seasonal conditions. A well-designed loop can respond to those changes while keeping the cooling system within its specified operating envelope, reducing the risk that water conservation efforts will compromise equipment reliability. The same philosophy applies to closed-loop liquid cooling, where coolant quality can affect heat exchangers, cold plates, pumps, and other components even when the loop does not consume water through evaporation.

Partnerships With Municipal Utilities

Utility partnerships can make alternative water sourcing more practical because the operator does not have to create every element of a reclaimed-water system independently. A municipal utility may already have wastewater-treatment, distribution, monitoring, and operational capabilities that can support a dedicated non-potable supply for industrial cooling. The data center can then become one participant in a broader reuse system rather than relying entirely on standalone treatment, although the project still needs water-quality management and cooling-water treatment appropriate to its thermal equipment. Microsoft’s work in Quincy, Washington, provides a useful example of how an operator and local utility can develop a reuse system that treats and recirculates water for data center cooling rather than relying on the same local groundwater source used by other users.

A utility partnership also changes the risk model because the operator becomes dependent on another infrastructure system whose capacity, treatment performance, pricing, and operating rules can change over time. The agreement therefore needs to address source reliability, quality specifications, outage procedures, storage requirements, emergency alternatives, discharge responsibilities, and the conditions under which the supply can be interrupted. The cooling system should not rely on a reclaimed-water connection without considering what happens during maintenance, treatment failure, drought restrictions, or upstream changes in wastewater characteristics. A resilient design can combine reclaimed water with storage, limited backup supply, improved cycles of concentration, or a dry-cooling mode so that a temporary water interruption does not immediately become a computing interruption. This layered approach mirrors the broader zero-water philosophy because resilience comes from reducing dependency at the system level rather than simply replacing one water source with another.

The Investment Lens — What CXOs and Investors Should Be Underwriting For

The investment case for a zero-cooling-water data center begins with treating cooling-water dependence as part of the site’s broader infrastructure risk. Water should be evaluated alongside operating and sustainability considerations because the availability, quality, and reliability of cooling water can influence the feasibility of operating and expanding a data center. Recent research has highlighted a less visible constraint in which community water systems may struggle with the peak withdrawals associated with data center cooling, particularly during the same hot periods when cooling demand reaches its highest level. That interaction creates a direct connection between thermal architecture and infrastructure capacity because a project can have sufficient electrical capacity while still encountering limitations in the water system serving the site. The investment question therefore becomes whether the cooling design gives the asset enough flexibility to operate through water constraints without creating disproportionate power, capital, or maintenance burdens.

Water Risk as a Due-Diligence Line Item

Due diligence should begin by identifying the physical water dependency created by the proposed cooling architecture rather than accepting a generic statement that the facility uses an efficient cooling system. Investors need to know whether the system depends on evaporation, whether the cooling loop remains closed, where makeup water originates, and what happens if the primary supply becomes unavailable. The analysis should also distinguish between water used inside a closed technical loop and water consumed through evaporation, because the two create fundamentally different resource exposures. The Department of Energy’s WUE framework provides a useful starting point for measuring site water performance, but the metric cannot by itself establish whether the underlying water source remains resilient or whether local infrastructure can support future demand. A credible diligence process therefore needs to connect operational water performance with local basin conditions, utility capacity, water-quality requirements, drought response, discharge obligations, and expansion plans.

Technical diligence should then examine the thermal architecture at the level of individual heat-transfer stages, because the phrase “water efficient” can conceal important dependencies between IT cooling and facility cooling. A direct-to-chip system may reduce the need for room-level air cooling while still relying on an evaporative heat-rejection system, whereas a closed-loop direct-to-chip system can remove routine evaporative consumption from the cooling process. An immersion system can create another contained thermal pathway, but the final heat-rejection equipment still determines whether the overall architecture consumes water. Air-side economization can eliminate water use under favorable outdoor conditions while requiring a separate mechanical strategy when the climate moves outside the economizer operating envelope. Investors should therefore request a system diagram that follows heat from the processor through the coolant distribution system, heat exchanger, chiller or heat-rejection equipment, and final environmental sink.

Where Capital Is Actually Flowing

New AI infrastructure is increasingly incorporating cooling architectures that address higher compute density while also controlling water and thermal performance. Direct-to-chip cooling has gained attention because it places the heat-transfer interface close to the processor and allows operators to pair liquid cooling with closed-loop, non-evaporative heat rejection. Oracle’s current AI data center approach explicitly uses direct-to-chip, closed-loop, non-evaporative cooling and describes the system as requiring an initial fill rather than continuous cooling-water consumption. Microsoft has similarly described an AI data center design based on direct-to-chip cooling and a closed loop intended to eliminate evaporative water use for cooling. These deployments matter to investors because they show that water reduction is moving into the core engineering architecture of new AI infrastructure rather than remaining an operational optimization applied after construction.

Data center operators are also deploying technologies that reduce water dependency without requiring every workload to use the same cooling architecture. Dry cooling, air-side economization, reclaimed-water systems, water treatment, heat recovery, and hybrid cooling can each address different parts of the water problem depending on climate and workload requirements. Brookfield’s Data4 operation provides an example of an operator pursuing low-water cooling through an architecture that avoids cooling towers and adiabatic systems while also examining gray water, rainwater, and industrial wastewater as additional sources. The value of these investments comes from matching the technology to the physical conditions of each site rather than applying one universal solution across a portfolio. A water-constrained location may justify more capital-intensive dry cooling, while another site may obtain a better overall result by combining efficient liquid cooling with reclaimed water and carefully controlled evaporative equipment.

Toward the Zero-Water Standard

The zero-water data center should not be understood as a claim that every facility can operate without water entering the property for any purpose, because the practical objective concerns cooling-related consumption and the architecture that creates it. That distinction keeps the discussion grounded in engineering rather than turning water reduction into an abstract sustainability label. Direct-to-chip cooling can capture heat close to where it is generated, immersion can use dielectric fluid as the primary thermal interface, closed-loop and dry systems can reject heat without routine evaporative water consumption, and reclaimed-water systems can reduce dependence on potable supplies when water-based cooling remains necessary. Each path solves a different part of the problem, and the strongest designs will often combine them instead of choosing one technology as a universal answer.

The next stage of data center development will make that discipline harder to avoid because AI workloads concentrate heat in ways that expose weaknesses in conventional cooling architectures. Operators can no longer evaluate compute capacity separately from the system that removes the heat generated by that compute, particularly when the local water network may face its own capacity constraints during periods of high cooling demand. Recent research has framed peak water capacity as an emerging infrastructure bottleneck, while current operator designs from Oracle and Microsoft demonstrate that closed-loop, non-evaporative cooling can already remove routine cooling-water consumption from new AI-oriented facilities. Those examples do not establish a single industry-standard solution, but they demonstrate that routine cooling-water consumption can be eliminated from some AI data center designs when the thermal architecture is engineered from the compute equipment through to final heat rejection.