That is how the sector can move beyond energy-efficient data centres towards genuinely climate-responsive, water-resilient digital infrastructure.

As AI accelerates, compute capacity will continue to be measured in megawatts. The quality of India’s digital growth should also be measured in the litres those megawatts require—and increasingly, in the litres they no longer require.The second principle is site-specificity. India is too climatically and hydrologically diverse for a single cooling template. No single approach minimises data centre water consumption. A coastal, humid site, a plateau city, and a hot inland location may therefore require different heat-rejection systems.The first principle is design integration. Architectural, electrical and mechanical infrastructure can no longer be optimised separately because high-density computing makes their interactions too significant. The same applies to sustainability: energy, water, refrigerants, heat rejection and IT performance must be considered together from concept design onwardsIndian operating examples show how incremental changes add up. STT GDC India reports that between 2020 and 2024 it reduced WUE by 34.5 per cent to approximately 0.73 L/kWh. It attributes the improvement to liquid-cooling technologies, rainwater harvesting, treated wastewater and water monitoring. At Pune DC1, it reports reducing withdrawals through higher cycles of concentration, conductivity changes and high-TDS water reuse; at Chennai DC2, it integrated air- and water-cooled chiller piping to reduce dependence on water-based cooling. These are company-reported results, but the lessons- meter first, optimise chemistry, increase reuse and maintain cooling-mode flexibility- are broadly applicable.Rear-door heat exchangers offer a useful middle ground, particularly for retrofits. They remove much of the server exhaust heat directly behind the rack, allowing lower-density hardware to remain air-cooled while selected high-density rows receive liquid-assisted cooling. This type of segmentation between AI/HPC and lower-density computing is recommended.India’s AI boom is creating a new challenge for data centres: how to expand computing capacity while managing the water needed to keep these facilities running. As compute becomes denser, the heat generated by each rack rises and so does the pressure on cooling systems, local water supplies and the resilience of the sites that host them. India is adding data centre capacity at speed, but the next phase of growth will be judged not only by how many megawatts can be delivered, but by whether those megawatts can operate through heatwaves, droughts, and tighter resource constraints. That makes water resilience an engineering priority, not a sustainability afterthought. The opportunity is to rethink cooling, siting, water sourcing, and operations together, so that AI-ready infrastructure can scale without straining potable water supplies.

India’s digital growth meets a physical constraint

India’s data centre story has, understandably, been told largely in megawatts. But artificial intelligence is changing the physical character of the infrastructure being built. The Ministry of Electronics and Information Technology reported in March 2026 that national data centre capacity had increased from about 375 MW in 2020 to about 1,500 MW by 2025. The same statement noted that approximately 38,231 GPUs had been onboarded through 14 empanelled service providers, while electricity demand from data centres is estimated to reach 13.56 GW by 2031–32.

AI and high-performance computing change the thermal equation inside a data centre. ASHRAE’s current AI Data Centre Energy Performance Framework identifies direct-to-chip liquid cooling as increasingly important for high-density AI and HPC deployments. Its guidance cites AI thermal loads of 60–120 kW per rack and above.

Virtually every kilowatt of power entering IT equipment ultimately becomes heat that must be dissipated. How that heat is removed determines electricity consumption and, in many conventional systems, water consumption. Cooling and environmental control account for roughly 7 per cent of total electricity consumption in highly efficient hyperscale facilities and more than 30 per cent in less-efficient enterprise data centres, according to the International Energy Agency.

India faces a more nuanced question than whether it can add another gigawatt of data centre capacity. It must ask where capacity should be located, what cooling architecture it should use, and how it will operate during heatwaves, droughts, and local supply constraints.

India’s 2025 Dynamic Ground Water Resources Assessment identified 6,762 groundwater assessment units nationwide. Of these, 730, or 10.8 per cent, were classified as over-exploited; another 201 were critical and 758 semi-critical. Nationally, the stage of groundwater extraction was 60.63 per cent, but that figure masks significant local differences—which is why water resilience cannot be assessed using a national average alone.

For data centres, national availability versus local stress matters. Water is a basin- and aquifer-specific resource. A litre consumed in a relatively water-secure catchment does not have the same environmental or social consequence as a litre consumed during a dry season in a highly stressed aquifer. The Lawrence Berkeley National Laboratory review of data centre water consumption reinforces this point: workload-level water use can vary by more than four orders of magnitude depending on server efficiency, electricity-related water consumption, utilisation, cooling system, infrastructure efficiency and climate. India has no single recipe for minimising water use in data centres.

This is why the industry should resist framing the issue as a simple contest between “water cooling” and “air cooling.” Evaporative systems can reduce compressor energy but consume water; dry systems can cut direct water use but may require more energy. Liquid cooling can move heat efficiently without meaningful water use at the server, yet water may still be consumed if heat is rejected through a cooling tower. Climate and system boundaries matter.

Water Usage Effectiveness, or WUE, is an essential starting point. The US Department of Energy defines site WUE as annual site water use (in litres) divided by annual IT equipment energy use (in kilowatt-hours). Its energy counterpart, Power Usage Effectiveness (PUE), is the total facility energy divided by the IT equipment energy. These metrics should be examined together.

Consider the scale involved. If 1,500 MW of IT capacity were hypothetically operated continuously at full load for a year, it would represent approximately 13.14 billion kWh of IT energy. At an illustrative site WUE of 1.0 L/kWh, that would correspond to 13.14 billion litres of annual site water use. But it shows why small changes in WUE become material as capacity moves into gigawatt territory.

The objective should not be “zero water at any cost” or the lowest possible PUE, regardless of local water quality. It should be to minimise the combined burden of heat rejection while maintaining reliability: less energy, less potable water, less dependence on stressed sources and greater ability to operate when weather or water availability deviates from norms.

Water therefore moves from the sustainability report into the engineering brief.

Water Resilience Starts at the Site Boundary

Some of the most important water decisions for a data centre are made before a cooling system is selected.

Historically, site selection has focused on power, fibre connectivity, land, latency, security and regulatory considerations. For the next generation of AI facilities, water availability, source quality, climate, and catchment conditions must be included in due diligence alongside power. We should evaluate energy and water availability, regional resource constraints, temperature, humidity, flooding and other hazards during site selection.

A serious water-risk assessment should go beyond asking whether the local utility can supply a particular volume. The assessment should also distinguish legally available water from water that remains socially acceptable when neighbouring communities face prolonged water shortages.

The first question is the condition of the catchment and aquifer. Central Ground Water Board information can establish whether a location falls within a safe, semi-critical, critical or over-exploited assessment unit. WRI’s Aqueduct can provide broader water-risk screening, but engineering decisions should be grounded in local hydrology, seasonal availability, permits, and utility conditions rather than in a single global risk score.

The second question is source resilience. Two pipelines are not necessarily resilient if both depend on the same stressed reservoir or aquifer. A source map should identify potable municipal supply, treated municipal wastewater, on-site wastewater, harvested rainwater and permitted groundwater, then test each source under drought conditions.

Reclaimed and fit-for-purpose water can reduce both operating and reputational risk.

The third question is climate. Cooling equipment should not be selected against generic descriptions such as “hot and humid.” Teams should examine hourly dry-bulb and wet-bulb temperatures, humidity, seasonal extremes and heatwave behaviour.

Design teams should test future climate conditions rather than simply sizing against historical weather files. A plant that is efficient in a typical year but needs substantial evaporative assistance above a narrow threshold may become increasingly water dependent as extreme heat becomes more frequent.

An engineering brief should define four operating conditions: normal annual operation, peak summer operation, drought or water-curtailment operation, and loss of a principal water source.

This approach changes equipment selection. A facility may use dry heat rejection for most operating hours, with controlled adiabatic assistance during extreme heat. Another site with reliable tertiary-treated municipal wastewater may reasonably use evaporative cooling because the energy savings and use of a non-potable source yield a better outcome. A high-density AI campus may use warm-water direct-to-chip loops connected primarily to dry coolers, with limited trim cooling for extreme conditions. These are different answers because local constraints differ. Cooling strategy should balance water and energy rather than optimise one in isolation.

Site planning should also identify potential heat sinks. Heat recovery is easier to justify when a suitable user is nearby. Evaluating heat reuse where there is a viable sink such as a building, campus or process load, noting that warmer liquid-cooling loops provide more useful heat than low-temperature server exhaust air. In India, domestic hot-water preheating, selected commercial or institutional loads and nearby industrial processes may warrant evaluation.

Finally, master planning matters. A 20 MW first phase may be predominantly air-cooled, while later AI halls require liquid distribution and dry-cooler capacity. Reserving space for coolant distribution units, piping, heat-rejection equipment, water treatment and reclaimed-water storage is cheaper than retrofitting a constrained campus. ASHRAE similarly advises providing sufficient land and infrastructure for evolving density and cooling requirements.

A water-resilient data centre is therefore not simply a conventional facility with a rainwater tank attached. Resilience begins with land due diligence, catchment understanding and a mechanical master plan that treats water as a finite design input.

The Leadership in Energy and Environmental Design (LEED) rating system also provides a framework for addressing these issues at the building level. The United States Green Building Council (USGBC) has specifically adapted LEED for data centres to address their high energy and water demands, enabling project teams to consider energy efficiency, water performance and resource use as part of an integrated approach to building design and operation.

Cooling architecture for an AI-density world

The biggest opportunity created by AI is also the most obvious: as rack heat densities exceed what conventional air cooling can handle efficiently, designers must rethink how heat is captured. That disruption can also improve water performance.

Traditional data halls cool servers indirectly, moving chilled air through equipment before warm exhaust returns to cooling systems. In chilled-water facilities, heat passes through additional stages before reaching a cooling tower or dry heat-rejection device. Each additional fan, pump and temperature difference adds energy.

Before replacing infrastructure, existing facilities should correct the basics. Hot- or cold-aisle containment, blanking panels, airflow management, rack-inlet sensing, and improved temperature control can eliminate significant inefficiencies. The US Department of Energy (DOE) notes that improved air management can enable higher chilled-water temperatures and lower airflow, with referenced guidance indicating up to 20 per cent less chiller energy in relevant applications.

Overcooling creates similar inefficiencies. DOE guidance warns that unnecessarily low temperatures and narrow humidity limits increase cooling demand and, where cooling towers are used, water demand. Before adopting new technology, facilities should therefore address inefficiencies created by poor controls and air management.

For high-density AI installations, direct-to-chip liquid cooling fundamentally changes the equation. Cold plates capture heat directly from components such as GPUs and CPUs, with liquid carrying it to a coolant distribution unit and facility system. Direct-to-chip is an emerging approach for HPC thermal management and highlights liquid-assisted architectures, including rear-door heat exchangers and immersion cooling, for AI clusters.

Importantly, liquid cooling is not synonymous with water consumption. Closed loops can circulate coolant repeatedly with limited make-up. Water consumption depends largely on how heat is ultimately rejected: an evaporative cooling tower consumes water, while a suitably designed dry cooler can keep direct cooling-water consumption very low.

Temperature is the key enabler. Liquid-cooled IT capable of accepting warmer coolant increases the temperature difference between the fluid and ambient air, making dry heat rejection practical for more operating hours. High-temperature secondary cooling loops can enable direct rejection through dry coolers even in relatively warm climates.

That does not mean every Indian AI data centre can dispense with chillers. Indian summer conditions must be considered. Designers need to establish how many annual hours of dry heat rejection the system can support, how the system performs during extreme heat, and whether a hybrid architecture provides the right balance among water, energy, and IT availability.

Hybrid or adiabatic dry coolers can help achieve this balance. They operate primarily dry but use limited evaporative assistance during extreme conditions. For India, water should be enabled only when its thermal benefit justifies its use.

The hierarchy is therefore straightforward: capture heat close to its source; transport it at the highest temperature compatible with reliable IT operation; reject it without evaporation where practical; and use water strategically rather than continuously.

Rear-door heat exchangers offer a useful middle ground, particularly for retrofits. They remove much of the server exhaust heat directly behind the rack, allowing lower-density hardware to remain air-cooled while selected high-density rows receive liquid-assisted cooling. This type of segmentation between AI/HPC and lower-density computing is recommended.

Immersion cooling offers another option by submerging hardware in dielectric fluid. Its suitability depends on hardware compatibility, maintenance, fluid management, and supply chain considerations. The Government of India identifies direct-to-chip, adiabatic and immersion cooling among technologies being adopted to minimise data centre water use.

Regardless of architecture, reducing mechanical cooling hours remains valuable. Waterside economisers can reduce chiller operation when outdoor conditions permit, while maximising appropriate free-cooling opportunities and integrating controls with weather and IT load.

A US National Renewable Energy Laboratory case study illustrates the principle. Its hybrid cooling retrofit first reused available waste heat, then used dry thermosyphon cooling when conditions permitted, with a cooling tower handling the remaining load. The system saved approximately 1.16 million US gallons (4.4 megalitres) of water in its first year and 2.10 million gallons (7.9 megalitres) over two years while retaining high energy efficiency.

The lesson for India is not to replicate a U.S. facility, but to follow the sequence: reuse heat where practical, reject it dry when conditions permit, and consume water only for the residual requirement.

Heat recovery should follow the same principle. Where a reliable nearby thermal load exists, higher-temperature liquid loops can make recovery practical; without such a heat sink, additional equipment may simply add cost and complexity.

AI does not have to worsen water performance. Its thermal density is forcing a change in heat-transfer architecture that can move data centres away from cold-air delivery and continuous evaporative rejection towards warmer, more targeted and increasingly closed-loop systems.

Make Every Litre Work More Than Once

Not every data center will or should eliminate evaporative cooling. Where cooling towers remain, the next question is the quality and productivity of the water being used.

Cooling towers consume water through evaporation and blowdown. Cycles of concentration is an important operating variable: increasing cycles, where water chemistry allows, reduces blowdown and make-up water demand.

This is a water-chemistry exercise, not simply a setpoint exercise. Increasing concentration without understanding water chemistry can trade efficiency for scaling, corrosion or reliability problems. Conductivity monitoring, treatment, filtration and periodic testing should support any increase in cycles.

There are two ways to lower the freshwater burden: reduce make-up demand and replace potable water with fit-for-purpose non-potable sources.

The most promising source in many large Indian cities is treated municipal wastewater. Data centres generally do not require drinking-water quality for heat rejection. Where tertiary-treated sewage is available reliably, it can displace potable water. Project teams should engage utilities early to assess treatment capacity, pipeline routing, seasonal quality and contractual assurance. A reclaimed-water strategy introduced after mechanical design can become an expensive retrofit.

This approach is also aligned with LEED’s water-efficiency framework, which emphasises reducing potable water demand, measuring water performance, and considering alternative water sources and reuse. For data centres, this is particularly relevant where cooling represents a significant water demand: the objective is to match water quality to the intended use and reduce reliance on potable supplies wherever technically appropriate.

On-site sewage treatment provides another, smaller circular stream. While its volume is unlikely to meet the needs of a large evaporative plant, treated sewage can replace potable water for flushing, landscaping, and selected cooling applications. Cooling-tower blowdown and air-handler condensate should also be examined for reuse where practical. The principle is a cascade: use high-quality water only where necessary, then seek another productive use before discharge

Indian operating examples show how incremental changes add up. STT GDC India reports that between 2020 and 2024 it reduced WUE by 34.5 per cent to approximately 0.73 L/kWh. It attributes the improvement to liquid-cooling technologies, rainwater harvesting, treated wastewater and water monitoring. At Pune DC1, it reports reducing withdrawals through higher cycles of concentration, conductivity changes and high-TDS water reuse; at Chennai DC2, it integrated air- and water-cooled chiller piping to reduce dependence on water-based cooling. These are company-reported results, but the lessons- meter first, optimise chemistry, increase reuse and maintain cooling-mode flexibility- are broadly applicable.

Rainwater harvesting should form part of the water balance, but it must be designed realistically. India’s monsoon delivers substantial rain over short periods, while cooling demand continues year-round. Schemes should consider actual catchment, local rainfall, first-flush requirements, filtration, storage and intended end use. A nominal tank tells us little about annual potable-water displacement.

In some cases, direct reuse will be preferable; in others, groundwater recharge may have greater catchment value. India’s policy direction supports rainwater harvesting and artificial recharge, with the Central Ground Water Board including aquifer management and recharge planning in its programmes. Data centres should connect site strategy to the wider hydrological context rather than treat rainfall simply as another private supply stream.

Storage can buffer short interruptions in non-potable supply, but a tank covering only a few hours of cooling demand will not solve a multi-week drought. It must be paired with a system capable of shifting to a lower-water mode.

Treatment should be matched to purpose. High-energy purification of every non-potable stream can undermine the rationale for reuse. Filtration, disinfection, softening or membrane treatment should reflect the chemistry required by the next use, while reverse-osmosis reject must also be managed.

The same discipline applies to “water positive” operations. Catchment replenishment can be valuable when projects restore aquifers, wetlands or community water systems, but it should complement—not excuse—inefficient operational consumption. Replenishment elsewhere or in another season is not necessarily equivalent to local withdrawal during acute stress.

For an Indian data centre, the hierarchy is: avoid unnecessary water demand; reduce the heat that needs to be rejected; use non-evaporative heat rejection where technically sound; optimise remaining evaporative plant; substitute non-potable sources; reuse water internally; harvest and recharge rainfall; and only then address residual impacts through wider catchment stewardship.That hierarchy keeps engineering efficiency ahead of accounting.

Operate water like a mission-critical resource

A well-designed cooling plant can still perform poorly if it is not measured, commissioned and continually tuned.Data centre operators monitor electrical performance at extraordinary granularity. Water deserves similar attention; a utility invoice lacks the resolution required to diagnose operational inefficiencies.

WUE is an important first metric, but a single annual WUE number cannot indicate whether water came from potable water, treated sewage, rainwater, or groundwater, or whether consumption peaked when the catchment was under the greatest stress. The LBNL review also finds substantial variation in workload-level water use across climate, cooling technology, server efficiency, and other factors, reinforcing the need to look beyond a single portfolio average.

An effective dashboard should therefore pair WUE with potable-water intensity, non-potable-water share, withdrawal by source, cooling-tower make-up and blowdown, cycles of concentration, recovered-water quantities, and peak-day consumption. PUE should remain alongside these measures so operators can identify whether water savings have shifted the burden into electricity

Metering must support this dashboard. Separate meters should cover cooling make-up, blowdown, treatment systems, rainwater, reclaimed water and domestic loads. Large campuses should sub-meter individual cooling plants or phases and feed flow data into building or data-centre management systems so abnormal flows or WUE changes trigger investigation.

Commissioning must test water performance, not merely verify equipment operation. Hybrid cooling sequences should be demonstrated across dry and wet modes, economiser operation verified, conductivity controls calibrated, flow meters checked and setpoints tested under realistic IT loads.

There is also significant low-cost potential in controls. Overcooling, narrow humidity control, and poor air management drive avoidable cooling and water consumption. The recommended approach is predictive integration of economisers with weather forecasts, IT load and redundancy requirements.

This creates an interesting role for AI within the infrastructure built to run AI. Cooling controls can anticipate changes in ambient temperature and compute load, adjust water or air temperatures within safe limits and select the most efficient combination of dry cooling, refrigeration and evaporative assistance. But sophisticated optimisation does not remove the need for understandable sequences, calibrated sensors and safe fallback modes.

Water quality deserves the same operational status as thermal conditions. Facilities using reclaimed water or higher cycles of concentration require consistent monitoring because chemistry can vary with seasonal and source conditions. Reducing blowdown is not an improvement if it causes fouling, lower heat-transfer efficiency or premature equipment failure. Water efficiency and reliability must therefore be optimised together

Drought response should also be built into operating procedures before a shortage occurs. Teams should define when to shift from wet to dry operation, which IT loads can be migrated, what capacity derating occurs without evaporative assistance, how much non-potable storage is available and which supply takes priority during an emergency.

These questions are analogous to electrical contingency planning. In a water-resilient data centre, the drought sequence should be aligned with the electrical sequence of operations.

Performance requirements should also flow into procurement. Rather than assessing cooling plants only at a single full-load design condition, owners can require suppliers to demonstrate annual energy and water performance across agreed weather profiles, IT-load conditions, and a defined drought mode. Metering, commissioning and water-quality controls then become deliverables rather than optional enhancements.

The outcome is an operating culture in which every litre can be reconciled, as with every kilowatt.

A Practical Design Brief for India’s Next Generation

The debate around data centre water use is sometimes framed as a choice between digital growth and water conservation. That is the wrong framing. India will require more digital infrastructure; the relevant question is what kind of infrastructure accompanies that growth.

The first principle is design integration. Architectural, electrical and mechanical infrastructure can no longer be optimised separately because high-density computing makes their interactions too significant. The same applies to sustainability: energy, water, refrigerants, heat rejection and IT performance must be considered together from concept design onwards

For project teams, this means establishing a water budget during feasibility, not after chiller selection. It should map annual and peak-day demand against source availability, identify potable and non-potable requirements, set WUE targets and show how demand changes during heatwaves or water restrictions.

The second principle is site-specificity. India is too climatically and hydrologically diverse for a single cooling template. No single approach minimises data centre water consumption. A coastal, humid site, a plateau city, and a hot inland location may therefore require different heat-rejection systems.

This also has policy implications. Rather than applying a single prescriptive WUE threshold regardless of climate or water source, a stronger framework would require transparent reporting of WUE, potable water use, water sources, catchment conditions, and drought resilience, while allowing system-specific optimisation

The third principle is density-aware cooling. AI zones should be designed around thermal density rather than forcing every rack into a legacy air-cooling architecture. Liquid or liquid-assisted cooling is recommended for AI clusters while retaining air systems for suitable lower-density areas. This allows existing facilities to evolve without requiring an immediate conversion of the entire estate.

Where direct-to-chip cooling is used, engineers should push towards the warmest reliable fluid temperatures permitted by IT equipment. Higher temperatures increase the opportunity for dry-heat rejection and reduce refrigeration. Space, piping, CDUs, leak detection, isolation and containment should be planned from the outset for liquid distribution.

The fourth principle is dry first, wet when justified. This does not mean rejecting cooling towers; it means questioning the assumption that evaporation should be used year-round. Annual modelling should establish when dry coolers or economisers can carry the load, with hybrid systems using water only when it provides a meaningful energy or capacity benefit.

The fifth principle is fit-for-purpose water. Drinking-quality water should not be the automatic choice for cooling, flushing or landscaping. India already offers examples of this approach being translated into building operations. Infosys’ Crescent campus in Bengaluru, for example, became the first project in India to achieve LEED Platinum under LEED v5 for Operations and Maintenance. The campus captures 100% of storm-event runoff via injection wells and storage, while high-efficiency plumbing fixtures have reduced water consumption by 34%. Such examples demonstrate how site-level water management and operational efficiency can work together rather than being treated as separate sustainability efforts.

The sixth principle is design for drought, not just annual efficiency. A normal-year WUE value tells only part of the resilience story. Owners should understand how the facility performs when water is curtailed during the hottest month. A hybrid plant that shifts to dry mode with a known capacity penalty may be more resilient than a marginally more efficient plant that depends continuously on water.

The seventh principle is to measure actual outcomes. Water performance rarely comes from one technology. As explained earlier in the STT GDC India performance, a significant reduction in WUE was noted between 2020 and 2024, averaging about 0.73 L/kWh, through measures including cycles of concentration, conductivity controls, high-TDS reuse, hybridisation, rainwater systems, metering, alternative sources, and liquid cooling. The lesson is that sustained performance comes from multiple design and operational decisions working together.

The eighth principle is optimised at the workload level. Water use is not determined by cooling alone. The 2025 LBNL review identifies server efficiency as the most important determinant in its analysis, followed by the water intensity of electricity supply, server utilisation, and factors such as cooling technology and infrastructure efficiency. More useful computing per kilowatt-hour means less heat to remove for the same digital output—and potentially less cooling energy and water.

This means mechanical engineers cannot compensate indefinitely for inefficient computing, just as IT teams cannot ignore the physical consequences of hardware utilisation. Hardware, software, electrical systems and cooling need to be considered together.

There is also a broader definition of resilience. A data centre can have redundant pumps, chillers, and generators and still be vulnerable if its design assumes an unlimited water supply. True resilience means preparing for extreme summers, groundwater restrictions, variability in reclaimed-water supply, changing equipment density and a future climate that differs from historical conditions

India’s expansion creates an unusual opportunity because much of the required infrastructure is still to be built. Retrofitting a cooling-intensive estate is difficult; avoiding unnecessary water dependence at the design stage is easier. Government data show the pace of change and recognise direct-to-chip, adiabatic and immersion cooling among technologies that can reduce water requirements.

The next phase should go further. Developers can incorporate water risk into land acquisition; utilities can plan reclaimed-water infrastructure alongside data centre clusters; designers can assess warm-water liquid cooling and dry heat rejection early; operators can track potable WUE alongside PUE; and project teams can treat rainwater harvesting, recharge and wastewater reuse as part of a catchment strategy rather than isolated compliance measures.

The direction of travel is clear. The most sophisticated data centre will not necessarily have the most elaborate cooling plant. It will be the one that understands where its heat goes, where its water comes from and how both change from hour to hour and season to season.

India does not need a single “waterless” technology. It needs a disciplined hierarchy: locate responsibly; minimise heat; match cooling to rack density; raise heat-rejection temperatures where possible; use dry cooling when technically and environmentally sensible; reserve evaporation for where it adds value; replace freshwater with fit-for-purpose alternatives; reuse water; harvest rainfall; meter flows; and prepare explicitly for drought.

That is how the sector can move beyond energy-efficient data centres towards genuinely climate-responsive, water-resilient digital infrastructure.

As AI accelerates, compute capacity will continue to be measured in megawatts. The quality of India’s digital growth should also be measured in the litres those megawatts require—and increasingly, in the litres they no longer require.

About the Author

P. Gopalakrishnan As Regional Director of GBCI India, Gopalakrishnan manages business and market development of LEED and other GBCI rating systems for the Southeast Asia and Middle East regions. His expertise includes corporate strategy, international market expansion, branding, and business unit creation. An alumnus of College of Engineering, Guindy and IIM Kolkata, he has more than 20 years of corporate experience in Southeast Asia and Middle East regions.