How well does your cloud hold water?

All datacentres consume water, but some are more thirsty than others

The third in our series of articles based on our cloud sustainability research takes an in depth look at water use.

Key takeaways

• All three hyperscalers share a target of being net water positive by 20230
• Amazon does not publish a figure for the water consumption of AWS datacentres
• Microsoft Azure datacentres increased water consumption by one third in the FY22
• GCP datacentrates consumed triple the amount of water that Microsoft did, which indicates lack of innovation re cooling
• WUE (total water consumption divided by IT Energy Efficiency) is not published by Google or Microsoft
• Amazon published WUE is 0.19L/kwh - that's pretty good
• Overall, Amazon and Microsoft tie in these categories, and Google does badly

Deep dive

Since April, Computing has posted a series of articles based on our cloud sustainability research. The first article was a summary of the main findings of the research and the second focused on the indirect carbon emissions which constitute the majority of emissions overall.

Our attention now turns to water – a resource consumed in vast quantities by cloud datacentres.

Data centres are now the tenth largest industry for water consumption according to a recent Bank of America study, and a report earlier this year suggested that China's datacentres consumed approximately 1.3 trillion litres of water a year which is the equivalent to the residential water use of about 26 million people in China.

Closer to home, it was reported last year that Thames Water is considering physically restricting the water flow to datacentres in its area (which given that it covers the M4 Heathrow – Slough corridor is most of them) during peak periods.

There are several criteria in the Sustainability Matrix that relate directly to water use. The first is net water withdrawal. The second is Water Usage Efficiency (WUE) which is calculated by dividing the overall consumption in litres over a given time period by the IT Equipment Energy (IEEE) which is measured in kilowatt hours over the same time. This is an important metric, because it measures not just how much water is consumed but how efficiently it is used.

Heavily influencing both of these metrics are the methods of cooling used, and the extent of innovation. The types of cooling used, and the physical location of the datacentre (lots are in areas of water scarcity) will also have an impact. The location will also influence the type of cooling – free air cooling is more effective in temperate and cooler climes and uses next to no water.

Before diving in, it's worth pausing to note that all three hyperscalers have the same target with regard to water, and that is to be water positive (replenishing more water than consumed) by 2030.

It's also worth noting that the metrics provided on water by AWS, Microsoft and Google concern their direct water consumption. Water is also consumed in the generation of electricity consumed by datacentres but this is out of reporting scope for now.

Water consumption and efficiency

This is a revealing metric. It's impossible to hide behind favourable ratio calculations or cherry-picked readings because net consumption is an absolute number. It's unforgiving. Amazon does not provide this figure for AWS at all. Instead, Amazon provides a WUE of 0.19L/kwh which is an improvement on the 0.25 L/kwh in 2021. It's impressively low and probably indicative of the innovation in the cooling space that AWS is engaged in.

The company subsequently scores one out of a possible three points due to a simple lack of data,

Microsoft scores zero because both water withdrawal and consumption have increased significantly year-on-year. Microsoft reports water consumption in megalitres (ML) - each ML is equivalent to 1 million litres. Even if we overlook the increase in water withdrawn (from 8068 ML last year to 10,706ML) consumption increased from 4773ML to 6399ML. An increase in water consumption of 34% feels at odds with the water positivity target - 2030 really isn't very far away.

A footnote adds that 1,310ML of total water consumption (20 per cent) was from water stressed areas.

No WUE is provided despite, incredibly, Microsoft stating in their sustainability report how important a metric it is. However, a source told Computing that Microsoft datacentres have a global average WUE of 0.49L/kwh which makes them less water efficient than their AWS competitors.

There is good and bad news on Google datacentres. The good news is that Google manages to provide water data for datacentres only, and even breaks down consumption by datacentre. The total amount of water consumed by Google's datacentres stood at 5.2 billion gallons. At first this doesn't look too bad - until you check the unit of measurement. Each gallon is equivalent to approximately 3.78 litres. 5.2 billion gallons is equivalent to roughly 19,732 ML.

Google was invited to comment about what looks like extraordinary water use when compared with Microsoft and the company provided the following statement:

"We constantly look for more efficient ways to use water, using data through our water risk framework (released Dec. 2023) to understand local hydrology and engaging with local experts to find the best solution for each community. We also invest heavily in our data center systems and operations, which we're continually reviewing and improving, to maximize efficiency with the water that we do use. And wherever we use water, we are committed to doing so responsibly. This includes using alternatives to freshwater whenever possible, like wastewater, industrial water, or even seawater. We utilize reclaimed or non-potable water at more than 25% of our data center campuses."

Cooling off

Amazons WUE indicates both a high degree of innovation and that AWS datacentres use a high proportion of recycled water for cooling. The company states a preference for evaporative cooling which uses outdoor air to cool datacentres and when outdoor air becomes too warm, cools air via evaporation.

As of 2022, 20 AWS datacentres used recycled wastewater in cooling - although there is no way of knowing what proportion of overall datacentres that accounts for. No further detail is provided so based on WUE, AWS get two out of three points available here. If more detail were provided, AWS would have received all three points.

Microsoft scores full points by virtue of providing greater detail and showcasing considerable innovation. Like AWS, Microsoft Azure use direct outside air most of the year to cool servers. Otherwise, the company uses direct evaporation which is less water intensive than alternatives like water-cooled chillers.

Microsoft also gets credit for being the first cloud provider to run two-phased liquid immersion cooling in a production environment. Further Micrsoft says:

"At the component level, we're testing new biodegradable materials to empower greater circularity of our servers and exploring microfluidics to cool chips more efficiently. As Microsoft and the industry continues to advance immersion cooling technology, we're also looking at ways to optimize the performance of chips beyond their pre-defined voltage, thermal, and power design limits, a concept called overclocking." Microsoft has also invested in air to water generation.

Having said that Microsoft has recently shelved its groundbreaking underwater datacentre project.

Google provides minimal information. The main sustainability report refers you back here, should you require any further detail. Incidentally, this is the same page that Google referenced when Computing reached out to them for comment on the same topic last year. The company uses water cooling in most datacentres and uses wastewater, sea water and canal water in some cases. 18% of freshwater withdrawals are from regions in high water stress.

One thing Google does show is the trade-off between water and carbon when it comes to making decisions about cooling. According to Google:

"In many places, water is the most efficient means of cooling. When used responsibly, water cooling can play an important role in reducing emissions and mitigating climate change. Water-cooled data centers use about 10% less energy and thus emit roughly 10% less carbon emissions than many air-cooled data centers. In 2021, water cooling helped us reduce the energy-related carbon footprint of our data center portfolio by roughly 300,000 tons of CO2."

Which cloud wins?

There is no clear-cut winner. AWS and Azure tie for the most points but for slightly different reasons. If AWS had provided the water consumption figures it would have won here.

Perhaps more importantly, water is the only area of the whole sustainability matrix where GCP does badly. It has fewer datacentres than AWS or Azure but seems to consume approximately three times as much water as Microsoft.

Google states that it's datacentres are primarily water cooled and these must require a constant flow of water. According to Google's data, out of 24 datacentre locations, only 4 use a mix of potable, non-potable and reclaimed wastewater.

You may also like

/news/4338523/tatas-uk-gigafactory-project-takes-major-step-forward

Components

Tata's UK gigafactory project takes major step forward

Sir Robert McAlpine to build multi-billion-pound factory

/opinion/4332800/peter-cochrane-energy-resources-free

Green

Peter Cochrane: Energy and resources are no longer free

We need new thinking

/news/4327230/microsoft-quietly-shelved-underwater-datacentre-project

Cloud and Infrastructure

Microsoft quietly shelved underwater datacentre project

Lessons have been learned from Project Natick, says Microsoft’s Noelle Walsh