KARACHI: A policy paper published by the Islamabad Policy Research Institute last month claims that Pakistan’s move to allocate 2,000 MW for AI data centres may significantly strain the country’s already stressed water resources.
According to the paper, ‘AI Data Centres vs Water Sustainability in Pakistan’, AI data centres pose critical environmental and infrastructural challenges. These centres consume water directly as a coolant while also contributing to indirect water usage through their reliance on Pakistan’s electricity grid.
These concerns are shared by several experts in the tech and AI research fields. In his comments to The News, chairman Wireless and Internet Service Providers Association of Pakistan (WISPAP), Shahzad Arshad said that “Widespread deployment of such centres without sustainable planning could intensify competition for water with agriculture and households, raise operational costs and provoke public and regulatory resistance”.
A single data centre can consume up to five million gallons of drinking water a day, enough to supply thousands of households or agricultural farms. While the policy paper also mentions AI data centres that use minimal water, according to Dr Irum Khan, a researcher specialising in AI policy and one of the co-authors of the policy paper, there are very few such centres and their new technology makes them more expensive.
An AI data centre houses the IT infrastructure needed to train, deploy and deliver AI services and applications and have computing, network and storage capabilities and the energy and cooling capabilities to handle high AI workloads.
“Water cooling is less expensive as it is most convenient and less costly…also it is the conventional method for any hyper-scale data centre”, says Dr Irum. However, this is just the direct water consumption and indirect water consumption through the electricity grid is harder to trace.
Due to the country’s heavy reliance on thermoelectric power plants, it is vulnerable to getting a high water footprint as a result of these plants’ cooling requirements. Arshad argues that “unless Pakistan prioritises water-efficient cooling, renewable energy and responsible site selection, the growth of AI infrastructure could significantly aggravate the country’s water problems”.
The policy paper warns that even using just 10 per cent of the 2000MW allocated for AI data centres could cause water use to spike by 5.2 billion litres per year. As per the International Organisation for Migration, Pakistan is already predicted to become water scarce by 2035.
To mitigate the water problem, Dr Irum says “data centres should be located at sites where there is no clash between local communities and data centre water use. Water should be enough all year round because data centres do not take heed of seasonal water shortages”.
It should be noted that the country’s key economic and industrial zones fall within ‘high’ to ‘extremely-high’ baseline water stress zones, as affirmed by the policy paper’s analysis. While countries like South Africa have reportedly limited data centres use of water, experts say that this is an unlikely path for Pakistan.
Peshawar-based researcher Furqan Ali points to a “governance vacuum” and thinks “it is unlikely any serious restrictions or pricing mechanisms for AI-related water use will emerge soon”. This is a view backed by Shahzad Arshad who says “If we see [the] past practices of [the] cement industry. We (the government) were not able to limit them for use of fresh water”.
When asked about the potential use of desalinated water to mitigate the problem, Furqan Ali claims that reports of a potential Chinese-backed desalination plant at Port Qasim “could be a game-changer for Pakistan and a critical intervention in the looming water-AI conflict. But again, we need long-term thinking, without rushing ahead, but also without delaying the necessary reflection and planning”.
