Cooling the hysteria: Reframing the narrative around data centre water usage

Data centres are in the spotlight, frequently criticised as insatiable consumers of precious water resources. Headlines warn of “thirsty tech giants” draining local supplies, fuelling widespread anxiety about the sustainability of our digital infrastructure.

Yet, the more sensational reporting is often derived from obsolete figures and outdated cooling technologies, painting a dire picture that doesn’t reflect the industry’s ongoing evolution. This narrative, while provocative, oversimplifies reality.

Data centres are undergoing a transformative evolution. Cooling technologies such as, direct-to-chip liquid cooling, immersion cooling and refrigerant-based phase-change cooling achieve equal or better thermal management than air and water-based systems, while drastically reducing evaporative water use. These technologies exploit the high latent heat of vaporisation of refrigerants, to transfer more heat from the source to atmosphere, when compared to using water as the primary transfer medium.

This evolution is also driven by escalating thermal rejection requirements of data racks. The rapid development of high-density compute workloads powered by advanced AI chips - exemplified by NVIDIA’s Blackwell, Rubin and Feynman series - means traditional cooling systems relying on water-based evaporative processes are unable to dissipate heat within the required response time, creating operational inefficiencies and increasing failure risks. We are now designing data centres that integrate new cooling technology that can dissipate higher thermal loads, within the same physical constraints of traditional data centre design.

It is important to recognise that the water footprint of a data centre includes more than just on-site water use. To borrow from the Greenhouse Gas Protocol emissions framework, the water footprint of a data centre can be considered in three parts:

• Scope 1: Direct use - Water consumed on-site for cooling, sanitation and fire suppression.
• Scope 2: Indirect use - Indirect water consumption tied to electricity generation powering the data centre.
• Scope 3: Embodied water in the supply chain - Water embedded in the supply chain, including manufacturing and construction lifecycle activities.

The total water impact of a data centre depends heavily on the water intensity of its energy source. For instance, a data centre powered by a grid reliant on thermoelectric power plants with high water consumption will have a much larger total water footprint than an identical facility served by a grid with abundant low-water-use energy sources like geothermal, wind or solar. This regional dependency makes geographic location and local grid characteristics critical considerations for assessing and reducing a data centre's total water footprint.

Technologies that can minimise water use while maximising cooling already exist today. Some examples of these include:

Direct-to-chip liquid cooling

• What it is: Dielectric fluids or water-glycol mixtures circulate directly through cold plates attached to high-heat-load components like CPUs and GPUs, efficiently removing heat at the source.
• How it differs from traditional approaches: Unlike air cooling or evaporative systems that cool the room and require significant water loss through evaporation, direct-to-chip cooling delivers liquid coolant directly to server components like CPUs and GPUs. This targeted, closed-loop system removes heat at the source with greater efficiency and minimal water dependency.
• Water consumption improvements: Because the cooling circuit is sealed and recirculates fluid, on-site Scope 1 water use is close to zero. Any water use is limited to occasional, chemically treated blowdown for the secondary heat rejection system, which can often be air-based or hybrid.

Immersion cooling

• What it is: Servers are fully submerged in nonconductive dielectric fluids in single-phase immersion, where the heat rejection liquid does not boil or in two -phase immersion where the heat rejection liquid boils.
• How it differs from traditional approaches: Instead of relying on air movement and fans, components are submerged in a bath of non-conductive liquid that absorbs heat directly. The fluid circulates in a closed-loop system, transferring more heat per unit of liquid than air cooling - and without the need for internal server fans.
• Water consumption improvements: Evaporative water use is eliminated entirely at the rack level. The cooling fluids are recirculated, filtered and reused continuously, requiring no potable water.

Refrigerant-based phase-change cooling

• What it is: Refrigerants with high latent heat capacity (e.g., R-134a, low-global-warming-potential blends) absorb and transfer heat through evaporation and condensation in a closed, pressurised loop. These fluids can move heat over 100 times more efficiently than water by volume.
• How it differs from traditional approaches: Unlike open-loop evaporative cooling systems that rely on water evaporation to reject heat, refrigerant-based systems use sealed coils and compact heat exchangers operating at approach temperatures as low as 5°C. This enables precise, high-efficiency cooling without exposing fluids to the environment.
• Water consumption improvements: Because the system is fully closed, there are no evaporative losses. Water use is limited to infrequent maintenance or minor top-offs, resulting in a dramatic reduction in water demand compared to traditional cooling towers. However, these systems often require additional electricity for compressors, condensers, or pumps, especially in high-load or high-ambient conditions. While they eliminate water use, they can incur a moderate energy penalty - typically offset by the ability to cool high-density IT loads in water-stressed environments. Energy efficiency depends on system design, refrigerant choice and local climate.

Dry cooling and air-based systems

• What it is: This system uses air-cooled heat exchangers and high-efficiency fans to reject heat directly to the atmosphere, completely eliminating on-site water use. Often referred to as “dry cooling,” it operates without evaporating water and relies solely on ambient air to remove heat from the data centre.
• How it differs from traditional approaches: Evaporative cooling towers use continuous water evaporation to remove heat. Dry cooling systems replace water with airflow, often through finned coils or airside economisers. It typically comes with a 10-20% energy penalty due to increased fan power and reduced cooling efficiency when there are heat waves. Operators must weigh the water savings against higher cooling energy needs.
• Water consumption improvements: Zero on-site Scope 1 water use, making it ideal for arid or water-stressed regions. However, dry cooling increases electricity demand, especially during heatwaves, which may raise Scope 2 water consumption. Combine dry cooling with adiabatic assistance (adding a mist on only the hottest days) can reduce this but uses water.

Circular water management

• What it is: Implementation of water reuse and recycling practices such as closed-loop cooling tower blowdown recycling, municipal wastewater partnerships to use treated effluent, rainwater and grey water reuse to feed the water side of a cooling systems.
• How it differs from traditional approaches: Reusing treated water within cooling system reduces the need for continuous potable water makeup from mains.
• Water consumption improvements: The water savings can be substantial but are dependent on climate and site context, regulatory constraints and capital expenditure on the infrastructure needed.

Not all legacy or water-based cooling systems are obsolete. Hybrid designs - such as liquid loops paired with water-side economisers - remain effective, particularly in temperate climates or where recycled or non-potable water sources are available. These systems still play a role in balancing efficiency, reliability and local resource constraints.

Adoption of next-generation cooling is advancing unevenly. Hyperscalers like Meta, Google and Microsoft are setting the pace, but colocation providers and smaller operators often face capital constraints, retrofit limitations or operational risk concerns that slow transition. Meanwhile, technologies like immersion and two-phase cooling are progressing rapidly but still require broader standardisation, service ecosystem maturity and integration readiness.

The narrative around data centres and water is shifting. While legacy systems have shaped public concern, leading operators are already implementing advanced cooling, circular water strategies and location-aware resilience planning. The ability to decouple digital growth from water stress is no longer theoretical - it’s underway. What’s needed now is broader adoption and broader sharing of successes. Transparent reporting, climate-responsive design and efficient water management must become standard practice. 

1. Measure what matters: Benchmark total water use across Scopes 1, 2 and 3 to focus interventions where they’ll count most.
2. Plan with water in mind: Integrate water availability and intensity into site selection and resource planning - not just power and land.
3. Design for thermal efficiency: Pair direct-to-chip cooling with refrigerant-based or dry heat rejection, tailored to local climate and future weather extremes.
4. Close the loop: Implement circular water strategies - like greywater reuse, rainwater harvesting and closed-loop tower designs with on-site treatment and storage.
5. Stage water approvals: Align cooling-related water entitlements with actual ramp-up schedules to avoid over-allocation and reduce unnecessary infrastructure.
6. Automate for efficiency: Use AI and real-time sensing to optimise thermal performance dynamically, cut peak demand and reduce overall cooling loads.
7. Retrofit smart: In your next upgrade cycle, pilot liquid cooling on high-density racks to prepare your infrastructure for future thermal loads with less water.