Cloud Atlas: Sustainability In The Age Of Water-Guzzling AI Data Centres

How are these two connected? Data centres need water; lots of it. Chile’s environmental court determined that the new Google project would need at least 7.6 million litres of groundwater per day to cool the servers.

To put that in perspective, some Chileans only receive 15-20 litres per day, despite a 2021 Supreme Court ruling that mandates a minimum of 100 litres per person daily, in line with World Health Organization (WHO) recommendations.

With no national data centre policy to regulate the industry, local disdain and pressure mounted and last month, the court stopped Google’s plans short till it comes up with a new sustainable strategy.

While encouraged by investments of transnational tech giants, Big Tech data centres are mushrooming in developing countries, they are coming at a cost. Local communities and experts in the Global South warn of “data colonialism”. 

Much like traditional colonialism, where foreign powers controlled local resources, data colonialism is perceived as a form of exploitation where global tech companies benefit disproportionately at the cost of local people, natural resources, and environments. The most direct example is water-intensive data centres in drought-stricken Chile. 

The people want clouds, just not the kind the government is raking in investments for.

Data Centres Running Hot

Water is essential to life, and surprisingly to our virtual lives as well. It’s estimated that by 2030, each person in developed countries would have a “virtual doppelganger” that consumes as much water as their physical bodies. Here’s how:

The process of computing uses water in three primary ways — on-site water usage for cooling the system; off-site water usage at the time of electricity generation; and water usage during the manufacture of the equipment used.

With a need for high computing power, artificial intelligence (AI) models are particularly “thirsty”. Does asking ChatGPT to write a bad poem about your cat actually have a negative impact? Let’s take the case of GPT-3.

Shaolei Ren, one of the authors of the seminal research paper Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models, talked to The Secretariat about the numbers for India.

“Mechanical cooling is very energy-intensive and indirectly water-intensive,” said the Associate Professor of Electrical and Computer Engineering at the University of California.

“On-site water usage efficiency (WUE) for India is zero, only because Microsoft plans to use purely mechanical cooling, without water evaporation. You can view this type of cooling system as a bigger-sized household AC,” Ren says, adding, “This approach, however, is very energy intensive. You can see that the power usage effectiveness in India is 1.42, much higher than elsewhere for Microsoft's data centres. As a result, there'll be more off-site water consumption, energy consumption and carbon emissions.”

IF GPT-3 was used to incorporate “method-writing”, about 500 millilitres would go into structuring this article and ideating its headline. Multiply such usage by millions of users, and the water footprint is enormous.

Policy Push

The case of Chile is not far from home. According to a NITI Aayog report, the data hubs of Mumbai, Bengaluru, Delhi and Chennai are all likely to face major water crises. 

In particular, Bengaluru — India’s very own Silicon Valley — is suffering from an acute water shortage, with a daily shortage of 500 million litres.

This underscores the urgent need for legislation and transparency. In the European Union, organisations operating data centres will now need to file reports on their water and energy consumptions, as well as detailing the steps they are taking to reduce it. 

In India, the draft National Data Centre Policy, 2020 encourages the use of “solar or wind” energy for data centres, but it doesn’t mandate the use of these renewables, nor does it set a minimum threshold for green energy. 

The Way Forward 

AI may hold the potential to solve environmental challenges, but if its growth comes at the expense of already scarce natural resources like water, the very technologies meant to help could worsen the problem.

David Gyulnazaryan, a specialist in decarbonisation and heat-reuse of data centres advocates for a synergy of industries as a way forward. “We know data centres produce heat as a by-product. Heat is also energy. If it can be used for heating, or desalination of water, it would bring down the impact and cost. And this isn’t only for cold countries.” 

A holistic approach involving both industry and regulators is essential to ensure that AI contributes to a sustainable future, rather than deepening existing environmental crises.

This is Part II of a two-part series on the environmental footprint of data centres in India. Read Part I, which focuses on the surging power needs of data centres and why they need so much electricity (Hint: it has something to do with AI).