AI data centres | Tech promises should not eclipse the climate change crisis
In March, the Government of India approved a Rs 103 billion ($1.25 billion) investment in the IndiaAI mission, for strategic AI innovation, including the development of computing infrastructure and the creation of large language models. According to recent reports, India has emerged as one of the leading nations in AI adoption and innovation globally.
AI is often viewed in abstract terms, overshadowing the crucial physical infrastructure it relies on. AI training and operations demand vast amounts of data, which necessitates powerful servers running on electricity and requiring water to cool down. According to some estimates, ChatGPT uses up to 500 millilitres of water every time you ask it a series of 5 to 50 questions. As AI usage grows, so does the need for data processing centres to handle the increasing data load, and complex computations.
As of 2023, India ranked 14th globally in the number of data centres, with capacity projected to reach 1.4 GW by 2025, implying abilities to manage more data and process it faster. In response to growing needs, the expansion of data centres beyond cities like Mumbai and Chennai to tier-two and tier-three cities is anticipated. Several state governments have already adopted data centre policies, offering incentives like waiver of electricity duty and electricity tax.
Environmental implications of data centres
Energy and water consumption constitute a substantial part of data centre operational costs, prompting companies to invest in ‘green’ data centres using energy-efficient technologies. However, despite such efforts, energy and water consumption rates remain unsustainable. For example, in Uruguay, a Google data centre envisaged using 7.6 million litres of water per day, roughly equivalent to the daily water needs of 55,000 people in a city. According to a new report, big tech companies have only increased water consumption to cool down their data centres in the last few years.
Even transitioning to alternatives like air cooling would demand electricity, with cooling accounting for 30-50% of total energy consumption. By 2030, the world’s data centres are expected to use more electricity than a country like India. They are also responsible for nearly 1% of global energy-related GHG emissions. Since 2020, Microsoft’s emissions have seen a 30% increase as it builds more data centres for its AI and computing systems.
Shifting to renewable energy sources also has environmental costs — the manufacturing of hardware systems for clean energy leads to carbon emissions. Renewable infrastructure development is expected to produce 185 billion tonnes of carbon dioxide by 2100. With the negative environmental implications of rare earth minerals essential for clean technologies, it is imperative to assess the overall impacts of renewable energy production used by data centres.
Additionally, establishing data centres can have negative outcomes for people’s relationship with land and property. Physical data centres need land — as of 2022, over 500 acres were acquired in India for data centre development, a number set to increase with the data centre boom. In many countries, local communities have protested land acquisition for data centres on account of concerns over land rights, displacement, and environmental degradation. In North Virginia, the largest data centre market in the United States, locals filed a lawsuit, claiming zoning permissions for an Amazon data centre were not obtained through due process.
A need for guard rails
As data centres expand in India, there is a critical need to establish institutional safeguards to mitigate the risks they pose.
First and foremost, at a societal level, it is necessary to introspect on the high data-consumption trajectory driven by current imaginations of technological innovation. The fascination with larger models trained on a billion parameters is fuelling a digital economy that runs on big data. Companies are storing and collecting large tracts of data, holding out for a future promise of innovation breakthroughs in AI. These data storage and processing needs will keep up the demand for data centres.
While innovations in greener data centres are commendable and necessary, the current trajectory of AI growth is not environmentally sustainable. Pivoting to miniaturised AI models, built on principles of data minimisation and purpose-driven data collection are fundamental to curb prevailing data-hungry approaches.
Second, for existing data centres, the establishment of measurement metrics for water and energy consumption is essential for accurately assessing their impact on natural resources. Reliable measurement standards would enable oversight and allow operators to monitor and manage water and energy usage effectively, facilitating informed decision-making and resource allocation in data centre operations.
Third, the development of a comprehensive impact assessment framework for setting up new data centres is crucial to address the potential environmental impacts of operations. Existing frameworks often prioritise factors such as cost and infrastructure availability without adequately considering environmental effects. Site selection for data centres should be guided by parameters like projected electricity and water consumption, estimated emissions and impact on communities in and around the proposed location. Assessment of environmental impacts should include calculation of emissions across renewable energy supply chains.
Additionally, designating ecologically sensitive areas and residential areas as no-data centre zones can preserve biodiversity, protect fragile ecosystems, prevent environmental degradation, and mitigate pollution and resource crunch in densely populated spaces.
In a world facing unprecedented climate change, where lives and livelihoods are at stake, dependence on the promises of technology should not eclipse the planetary crisis at hand.
(Urvashi Aneja is Founder and Director, and Dona Mathew is Research Associate, Digital Futures Lab)
Disclaimer: The views expressed above are the authors' own. They do not necessarily reflect the view of DH.