How much water do data centers use?

Update: An updated, expanded, and peer-reviewed version of this blog post was published in the open access journal npj Clean Water in Feb 2021.

There are several components which make up environmental footprint of a data center. Energy usage is a big one – 1-1.5% of global electricity usage – but it is not the only one. Water is also a factor.

In FY18, Google used 15.79 billion litres of water (Google, 2019). Some went to offices, but the majority was consumed by its global fleet of data centers. In FY18, Microsoft used 3.5 billion litres (Microsoft, 2018), also with the majority going to its data centers.

This is important because water availability is a major concern in the context of climate change. Global water use has increased by x6 over the past 100 years, continues to grow by 1% every year (UN, 2020), and many areas suffer from water stress.

Water is essential to achieving the climate targets because climate change affects the availability, quality and quantity of water for basic needs such as industry and agriculture (UN, 2020). It is also subject to a specific Sustainable Development Goal: SDG6 – ensure availability and sustainable management of water and sanitation for all.

Projections suggest that demand for water will increase by 55% between 2000 and 2050 due to growth from manufacturing (+400%), thermal power generation (+140%) and domestic use (+130%) (OECD, 2012).

So what do data centers use water for? This post will examine where data centers use water and what that means for the environment.

Data center water use #

US data centers used 626 billion litres of water in 2014, projected to grow to 660 billion litres in 2020 (Shehabi et al, 2016). This is split across two main categories: electricity generation and cooling.

Water use in electricity was x4 greater than that used on-site for cooling: 7.6 litres of water is used for every 1 kWh of electricity generated compared to 1.8 litres per kWh of total data center site energy use (Shehabi et al, 2016). However, this is based on water consumption in US power production in 2003 so does not consider the impact of the decarbonisation of the electricity grid since then.

Water use in electricity generation #

The electricity used to power data centers requires significant volumes of water. Power plants burn fuel to heat water, generating steam to turn a turbine which then generates electricity. You often see the results from the huge cooling towers next to power plants.

The same happens in nuclear power plants where the heat is generated through nuclear fission. Hydroelectricity also relies on water (although it does not “consume” the water because it can be passed on to downstream users).

Only solar and wind do not involve water in the generation – instead the manufacturing process contributes the majority of the water footprint.

Although the IT industry reports x6 more compute instances, x10 more network traffic and x25 more storage capacity in 2018 compared to 2010, data centre energy consumption has only grown by 6% (Masanet et al, 2020).

This is helped by the move to the cloud. 53% of servers are expected to be in hyperscale facilities by 2021 (Cisco, 2018) and cloud vendors are some of the largest purchasers of renewable electricity (IEA, 2019). If the electricity is generated from wind and solar – which is not yet possible 100% of the time – then the water footprint is significantly reduced.

Water is a binary resource: it is either used or not used. This is different from non-renewable electricity generation which can at least be offset. Switching to renewable sources is the only way to reduce the impact of this water consumption in this category.

Water use in data center cooling #

Think how often you hear the fans on your computer spin up. IT equipment generates a lot of heat and must be kept cool to operate efficiently. Drawing cool air over the hot metal components transfers the heat energy to the air, which is then pushed out of the computer. This works because the computer temperature is usually higher than the surrounding air.

The same happens in data centers, just at a much larger scale. Chillers exchange the internal heat with the cooler external environment, or air is drawn in and then cooled down. A 1MW data center using traditional cooling can use around 25.55 million litres of water per year (Heslin, 2016).

Cooling towers cool warm condenser water from the chillers by pulling ambient air in from the sides, which passes over a wet media, causing the water to evaporate. The cooling tower then rejects the heat by blowing hot, wet air out of the top. The cooled condenser water then returns back to the chiller to again accept heat to be rejected.

Heslin, 2016.

Adiabatic economizers typically spray water onto heat exchanger surfaces or in some cases directly into the air stream. The water serves to cool the air as it enters the data center. In the process, however, some of the water evaporates and is lost. This means you need a reliable, continuous source of water for these systems to be effective.

Frizziero, 2018.

The typical data center has tended to operate at temperatures of 20-22C, with a lower bound of 12C (Miller, 2011). As temperatures increase, equipment failure rates also increase. However, experiments have shown that the failure rate does not grow linearly with temperature (Miller, 2008).

Raising the chiller temperature from the usual 7-10C to 18-20C can reduce operational expenses by 40%, due to the reduction in energy required for cooling. When designing new data centers, using smaller chillers can reduce capex by up to 30% (Frizziero, 2016).

Some Google and Microsoft data centers have been operating with cooling just from at ambient air temperatures since 2009 but this is not possible in hot environments.

Measuring data center water usage #

Power Usage Effectiveness is the metric used to report data center energy efficiency. This is the ratio between power drawn by the infrastructure components and power delivered to the servers, disks and networking equipment (GHG Protocol, 2017).

A PUE ratio of 1.0 would mean that 100% of energy going into the data centre went to the IT equipment. This is impossible because some energy is needed for cooling, lighting and power distribution. Use of equipment like uninterruptible power supples involves losses because nothing is 100% efficient.

Cloud vendors are pushing these ratios very low – Google reports a Q1 2020 trailing twelve month PUE of 1.10 for the 15 facilities it owns (Google, 2020).

Water Usage Effectiveness (WUE) is a similar metric for water. This is defined as the annual water usage divided by the IT equipment energy, with units of litres/kilowatt hour (L/kWh). It covers the operations, not the full life-cycle from construction to decommissioning.

However, less than a third of data center operators track any water metrics and water conservation is ranked as a low priority (Heslin, 2016). Facebook is one of the few companies to report their WUE ratio. They are very transparent about its overall environmental footprint and it even has real time dashboards for its Lulea, Forest City and Prineville data centers. None of the big cloud vendors publish water efficiency metrics.

Google and Microsoft water usage #

Although Google and Microsoft do not report WUE figures, they do report overall water consumption. These are total consumption figures which include offices and other facilities but consist primarily of water from data center operations. However, they only represent direct water usage, not water used in electricity generation.

Both companies have good track records when it comes to working towards reducing their carbon footprint and increasing use of renewable energy.

Google has been carbon neutral globally since 2007 and has matched its energy usage with renewables since 2017 (Google, 2020). Microsoft has been carbon neutral globally since 2012 and will match its energy usage by 2025 (Smith, 2020). However, they do not publish as much about their water consumption as they do with their energy and carbon footprints.

Microsoft submits an annual report to CDP, a sustainability reporting charity which includes detailed metrics and information about their water footprint (as well as other categories).

Google submits reports for climate change but not for water. Perhaps this is because there is significant controversy around Google’s data center water consumption:

Google considers its water use a proprietary trade secret and bars even public officials from disclosing the company’s consumption. But information has leaked out, sometimes through legal battles with local utilities and conservation groups…The conservation league called out the DHEC for giving Google so much water while asking a local public utility, Mount Pleasant Waterworks, to reduce its withdrawal from the aquifer by 57% over the next four years. The utility exceeded its previous peak use demand by 25% in May 2019, one of the driest months last year in Berkeley County, according to Clay Duffie, general manager of Mount Pleasant Waterworks

Sattiraju, 2020.

Amazon’s water usage #

Amazon is the worst environmental performer of the big three. It plans to be carbon neutral by 2040 and only matched 50% of its energy with purchases of renewables in 2018, aiming for 100% by 2030 (Amazon, 2019).

Unlike Google and Microsoft, it has a huge logistics business which it includes in vague carbon footprint figures. However, it is not transparent submitting no CDP reports and publishing no AWS cloud specific figures for its environmental footprint. Amazon does have a short page about water usage but it only describes in generalities and publishes no metrics.

Do data centers use a lot of water? #

Billions of litres sounds like a lot of water, but it depends what you compare it to.

For example, total water usage in the US in 2015 was 446 billion litres per day for irrigation, 88.2 billion gallons per day for domestic usage, and 15 billion gallons per day for mining (Dieter et al, 2015). These are massive daily quantities for quite large sectors. Compared to the 626 billion litres of water used by US data centers in the whole of 2014 (Shehabi et al, 2016), this is a very small proportion of overall water consumption.

Comparing these figures to other buildings allows them to be put into perspective.

Alternative water sources #

Despite Amazon’s poor environmental efforts in comparison to Google and Microsoft, it does make use of recycled water:

AWS is expanding its use of non-potable water for cooling purposes to help conserve local drinking water sources. In Northern Virginia, AWS was the first data center operator to be approved to use recycled water with direct evaporative cooling technology. We partnered with Loudoun Water to demonstrate the benefits of recycled water for industrial cooling applications, and shared our operational best practices for utilizing recycled water in our data centers. In the AWS U.S. West (Oregon) Region, we have partnered with a local utility to use non-potable water for multiple data centers, and we are retrofitting AWS data centers in Northern California to use recycled water.

Amazon, 2020.

This is important to avoid depleting sources better used for other purposes. Data centers are able to use water from any source but most come from municipal sources such as reservoirs (Heslin, 2016). The main reason for using these municipal sources is the reliability.

Other sources of water include rainfall, gray water, and surface water. Very few data centers use these sources for a variety of reasons. Because rainfall can be unpredictable, for instance, it is mostly collected and used as a secondary or supplemental water supply. Similarly only a handful of data centers around the world are sited near lakes, rivers, or the ocean, but those data center operators could pump water from these sources through a heat exchanger. Data centers also sometimes use a body of water for an emergency water source for cooling towers or evaporative cooling systems. Finally, gray water, which is partially treated wastewater, can be utilized as a non-potable water source for irrigation or cooling tower use. These water sources are interdependent and may be short in supply during a sustained regional drought

Heslin, 2016.

Taking water from the sea is a potential option, but the salt-water can cause problems with equipment. One idea is to co-locate data centers and desalination facilities, both energy intensity operations.

Deep Water Desal proposes to mitigate the power consumption of desalination in a very creative way. Rather than reduce the power required to desalinate water, they proposed to co-locate up to 150MW of data center facilities on site and reduce the power required to cool the data center. Essentially the desalination plant and data centers would be symbiotic and the overall power consumption of the combination of the two plants together would be lower.

Hamilton, 2014.

Other companies are working with local utilities to make use of their wastewater.

At its 18 Bay Area data centers, Digital Realty Trust Inc., which runs several dozen centers around the world, has set an internal goal to eventually cut its water use by a quarter. The company struck deals with local utilities to use recycled wastewater where available, but that “gray water” isn’t always available. In Los Angeles, for instance, a new recycled water pipeline could take years to reach most of the region’s downtown data centers

FitzGerald, 2015.

Reducing water usage increases energy usage? #

A full lifecycle analysis must be considered when evaluating alternatives. Reducing absolute water consumption may increase the environmental footprint if more energy is required. Water treatment involves energy intensive filtration equipment and chemicals.

The WUEsource metric takes these external factors into account. It is defined as (annual source energy water usage + annual site water usage) divided by IT equipment energy (The Green Grid, 2011).

Edge locations #

Choosing the optimum data center location involves many tradeoffs. Hyperscale data centers are often in regions with abundant access to renewable energy, such as Google’s Finland data center (Google, 2018). However, these locations tend to be away from population centers which means higher network response times as data must travel further to the end-user. As urbanisation increases, the need for low latency will require data centers to be sited closer to the user (Kass and Ravagni, 2019) but these locations may be less suitable for access to renewable sources of electricity or natural water sources for cooling.

Google has taken its DeepMind AI expertise to reduce cooling costs by up to 40% (Evans and Gao, 2016; Gao, 2017). Using data to extract efficiency improvements from existing infrastructure shows there are still gains to be made without significant capital expenditure.

Microsoft have experimented with underwater data centers through Project Natick, which avoids the need for cooling due to the sea temperature.

Project Natick is focused on a cloud future that can help better serve customers in areas which are near large bodies of water (where nearly 50% of society resides). The vision of operating containerized datacenters offshore near major population centers anticipates a highly interactive future requiring data resources located close to users. Deepwater deployment offers ready access to cooling and a controlled environment, and has the potential to be powered by co-located renewable power sources.

Where submerging servers isn’t possible, alternative cooling methods may be viable. For example, in Jan 2020, Mark Russinovich, Azure CTO, also presented some tests Microsoft has been conducting around liquid cooling for its servers.

Conclusions #

Not enough attention is paid to water usage. In the developed world, we are too used to turning the tap and having instant access to clean water for minimal cost. Yet, there are still over 785 million people without access to simple drinking water (UN, 2019). With the added pressure of risk to water due to climate change, even small savings in water usage can have a major impact.

A saving of 1% per year by better energy use or efficiency could provide water for 219 million people based on 50 L/day, depending on location and other factors.

UN, 2020.

In this context, we pay too little attention to the efficiency of data center water consumption. Cooling is not the largest consumer of data center power, and is therefore only a small % of total cost, especially in the hyperscale cloud environments (Bullard, 2019).

As usual with the data center industry, there is insufficient transparency. Microsoft sets the standard but even they are only reporting absolute numbers. There is no way to understand how efficient their operations are. The largest cloud provider, Amazon, doesn’t publish anything. Facebook is a leading example but they do not sell their infrastructure capacity so 3rd parties are unable to take advantage of their efficiency.

That said, it is clear that moving to the cloud helps significantly. Cloud data centers are more efficient, the big three providers procure the most renewable energy, offer the most environmentally aware products, and they all continue to invest in improving their environmental sustainability. They operate at such a scale which allows them to justify experimental projects like submersible data centers. All their customers benefit from the efficiency improvements.

Just as customers are starting to ask cloud vendors for data about their carbon footprint, they should also include water consumption in their evaluation criteria. Cloud providers should talk as much about how they are reducing water usage through monitoring their WUEsource ratio as they do about reducing carbon emissions and monitoring their PUE ratio. However, until there is pressure from the market to select based on a full range of environmental standards, things are unlikely to change.