Is power or water best for cooling?

No matter where you are, data centres are part of your everyday life. These facilities form the foundation of the online world, a world which partly explains why the UK government recently announced that they would be deemed mission-critical infrastructure. However, these facilities are hungry. They require a lot of energy to run, and for data centre operators looking to reach sustainability targets, they must be designed effectively.

Efficient and effective cooling is a critical aspect of data centre operation. Data centres generate a large volume of heat that must be managed to maintain IT performance. Mechanical engineers select system typologies to achieve these conditions, conventionally, these are either power-based or water-based systems.

The data centre region will impact which system is the most efficient. Some areas are water-scarce, and others rely on high-carbon power grids. Many data centre operators ask when it comes to their build whether to use water or electricity for cooling.

The role of water

Over the past decade, the target of low power usage efficiency (PUE) has been the key driving factor, pushing many operators to a water-based system; these systems typically fall into direct air or indirect air for cooling. For a direct-air system, outdoor air is passed into the data hall via an evaporative cooling humidifier. Water evaporates as air passes over the humidifier, and the incoming air is cooled. A direct-air system also uses air-side economisers to recirculate warm air to increase efficiency. The indirect air system works similarly, except the data hall air stream is passed over a heat exchanger, and an external air stream rejects heat to the atmosphere, so the external air does not enter the data hall. The heat exchanger's performance is enhanced by passing water over it. There is no need to run chiller compressors in these systems unless, in an especially warm climate, the site's power consumption is minimised. This works well to reduce PUE and maximise power for IT systems.

More recently, the demand for data centre water has been under the spotlight. Utilising water that can be used for agriculture or domestic use in a data centre may not be the most sustainable proposal. Additionally, we have seen an increase in demand for liquid cooling. This is mainly in response to the rise in artificial intelligence, which needs a greater cooling density given the higher power of the data racks. Liquid cooling uses a hydronic system which passes liquid through the racks to a series of chillers; therefore, we are seeing a return of chiller-based systems. While this meets the demand of modern rack densities, it also increases the facility’s power consumption, raising PUE. However, this can also be a more suitable solution for non-AI facilities in areas where water is scarce or where clean electricity is available.

Where is the data centre located?

Where the data centre is located can often impact what is best. Each region has different conditions in which it needs to operate, and often, the existing infrastructure that is available heavily varies. For example, in California, water scarcity from drought impacts data centre operators. These companies have had to move away from water-based cooling as it is simply unavailable. The benefit they have to offset this is some companies in this area have easier access to a relatively clean and reliable power supply to support their systems with electricity. Conversely, in regions like Poland, water is easily accessible. However, their power grid is considered 'dirty', relying on high amounts of fossil fuels, increasing their carbon footprint.

This leads to the crux of the answer to the question, what is best? The most sustainable option often depends on the site's location. In water-stressed areas, air-cooled chillers running on electricity can better meet the facilities' sustainability goals. If water is more accessible, operators can rely on it for cooling, alleviating the need for a carbon-intensive power grid.

There are other innovative alternatives, too. These include waste heat recovery and using sea or river water for cooling. Each presents different opportunities to improve the sustainability and efficiency of the data centre facilities they support. Waste heat recovery involves capturing and repurposing captured heat generated by servers, redirecting it to nearby facilities like district heating networks or offices, and turning the by-product into a usable resource. We have seen success with this on a large scale with a hyperscale data centre in Denmark in a close partnership with a client and Fjernvarme Fyn. Using heat pumps, the boosted temperature of the district heating network reaches 70- 75⁰C and, when fully built out, will provide 165,000MWh of heat per year.

Similarly, sea and river water cooling systems use the naturally low temperatures of the nearby bodies of water to provide cooling without the need for energy-intensive compressors. These hydronic systems reduce reliance on power-hungry cooling methods and align with sustainability goals by using the natural environment. However, the ability to implement them is location-dependent. It requires careful consideration of the geographical and regulatory factors and necessitates an experienced and holistic engineering team.

Rethinking sustainability metrics

Sustainability is a key consideration when building a new data centre. Regulations like the Paris Agreement and local regulations like the UK Carbon Building Standard or the Corporate Sustainability Reporting Directive (CSRD) provide a pathway to reach net zero carbon.

Traditionally, the sustainability of data centres has been measured through PUE or WUE – water usage effectiveness. PUE refers to finding a ratio between the total energy consumed and the energy consumed by the computing equipment. However, this can be limiting, particularly in accounting for situations when water plays a more significant role in cooling systems. WUE looks at the amount of water consumed (evaporated or sent to drain) against the IT power. However, WUE varies greatly based on local temperature profiles and mechanical system typology, so it does not provide a good data centre-to-data centre comparison. Additionally, where water is embedded into the process of generating power, WUE fails to recognise the water involved in this process. Since different power generation processes consume varying amounts of water, the water usage volume should be incorporated into the metric.

Alternatively, some has suggested carbon usage effectiveness (CUE) as a better solution. This puts an emphasis on carbon emissions. This metric can be best suited to find the optimum option for cooling as it can measure the benefits of reducing power and water usage against the potential carbon output. However, the carbon intensity of power, water production, and network must be factored into this calculation.

Is there a right option?

No.

While companies expand their data centre operations worldwide, they often use reference design development. This template design can be used to quickly bring a facility to new regions by editing rather than designing from scratch. However, these facilities still need experienced engineers and designers. They need to account for the climatic and regulatory conditions in which they operate and find the best cooling solution for their data centre and requirements. Particularly for cooling solutions, a one-size-fits-all approach can often result in inefficient, costly, and delayed facilities.