You’ve probably read the headlines: Can the UK cope with AI’s water consumption? Will data centres drain our water supplies? Why are data centres so thirsty?

This is only half the story. Legacy hyperscale data centres with consumptive cooling systems do use excessive amounts of water. However, efficient, AI-ready data centres (such as our Bradford site) with alternative or hybrid cooling systems use only as much water as a domestic house per year.

The idea that AI will drain the UK’s water? If we design AI-ready data centres with minimal water use in mind, that’s nothing more than a myth. For organisations or enterprises deciding on an HPC colocation partner, here’s how cooling design and water efficiency make a difference to your compute’s sustainability performance and cost.

How do data centres use water?

Data centre servers—particularly those suited to high-performance workloads—get extremely hot. Data centre cooling systems use a combination of water and power to cool down the servers. The efficiency of this cooling system, and the amount of water or power used, all contribute to a data centre’s resource consumption and environmental impact. A facility’s water consumption also affects the local community’s water resources. So, while the myth is partly true—data centres can use a whole lot of water for cooling—it’s not the whole story. When we design AI-ready data centres with water efficiency in mind, our AI use doesn’t have to drain water resources dry.

Data centre cooling systems explained

Consumptive cooling

Consumptive cooling refers to any system that requires a constant input of water, and is often used by legacy data centres. Because of their high water consumption, these cooling systems are especially vulnerable to droughts, and can drain water resources from nearby communities. However, their power usage is minimal, as they rely on water for cooling.

Closed-loop liquid cooling

Closed-loop cooling circulates water directly around a data centre’s servers. By targeting the exact area that needs cooling, rather than the whole building, the system is more efficient and suitable for cooling high-density servers. The ‘closed-loop’ aspect also means the water stays within the cooling system, minimising water usage because the same water is used repeatedly. This makes it especially resilient even in water-stressed areas.

Evaporative cooling

Evaporative cooling sprays water into a facility’s airflow. The water evaporates and absorbs heat from the air, lowering the air temperature in a data centre. While more energy efficient than air cooling, for example, evaporative cooling requires a constant input of fresh water. 

Adiabatic cooling

Adiabatic cooling combines evaporative cooling with dry cooling. Based on the outdoor temperature, it switches between both methods. Dry cooling is especially effective in colder climates, where the outside air is cool enough to reduce the fluid’s temperature. Then, when outdoor temperatures increase, the system switches to evaporative cooling.

Which cooling system uses the least water?

While avoiding consumptive-only cooling is essential for keeping water usage to a minimum, the ideal cooling system depends on external factors such as the typical temperature and water resources near a data centre. This is why adiabatic cooling is often the most efficient in climates like northern Europe; it maximises the efficiency benefits of different cooling types in different temperatures.

Why does water use matter when buying HPC colocation?

Your HPC colocation partner’s water use affects your organisation or enterprise’s sustainability performance.

The more efficient the cooling system, the lower your HPC colocation partner’s PUE and water use. In turn, your organisation or enterprise’s sustainability performance will benefit, reducing your scope 2 or 3 emissions, meeting EU taxonomy standards and strengthening your ESG narrative.

Beyond cooling: how AI-ready data centres that reuse heat maximise overall efficiency

We’ve already established that all those hot data centre servers need cooling. However, discarding this heat makes a data centre inefficient, no matter its cooling system. By capturing and reusing the heat (in residential or public buildings, swimming pools or greenhouses), data centres can improve their overall efficiency and minimise cooling requirements. Jon Laban, job title, explains that an efficient data centre must consider heat holistically beyond its cooling system. “When data centre operators discuss heat transfer and reuse, it shows they are thinking of how to gain value from the heat they produce. These systems take their water usage to that of a high street family restaurant and effectively reduce energy bills by 40%.”

How our Urmston AI-ready data centre heats a swimming pool with practically zero water usage

In Urmston, our Manchester data centre keeps its water usage to practically zero with a combination of dry and closed-loop cooling. Instead, Urmston’s focus is on heating water: the data centre servers warm 400 million litres of water a year that are used by the leisure centre’s swimming pool.

Our CTO Matt Craggs explains why we chose this cooling system: “The impact on the PUE is significant,” he explains; Urmston’s PUE is at 1.17. Compared to evaporative-only cooling, the system is expensive to install. “It has a premium to be paid on day one,” Matt says. “But the total cost of ownership to deliver the cooling, and the increased efficiency, means it's far more cost-effective than air cooling in the long run.”

How our Bradford data centre adapts cooling systems based on temperature

As for our Bradford data centre, we make the most of the North of England’s cooler temperatures with an adiabatic cooling system. When the weather warms up to above 25 degrees, the facility automatically switches to adiabatic cooling.  “What this means,” Matt tells us, “is that Bradford will use less water than a domestic house per year.” We’ve predicted just 100 to 200 hours of evaporative cooling a year. “When you’re designing an AI-ready data centre, it’s always a balance between water consumption and power consumption,” Matt explains. Not only is our water consumption extremely minimal, but Bradford’s PUE will be 1.08, compared to the industry average of 1.25-1.7.

The reality: AI won’t use up water resources if we design data centres right

Let’s be honest: AI isn’t going anywhere. However, with increasing heat waves and droughts across the planet, the data centre infrastructure we build to run AI workloads must be designed to minimise water use. Want a blueprint of a water-efficient, AI-ready data centre? Check out our Urmston facility.