AS plans for large data centres take shape in Pakistan, the debate has centred on jobs and investment, while a more basic concern has received little attention: water. Why would a data centre compete for water? Computing generates heat, and evaporating water is one of the most cost-effective ways to remove it. By 2030, the United Nations University projects that data centres globally will use nearly triple the combined electricity usage of Pakistan, Bangladesh and Nigeria, with an associated water footprint equal to the basic domestic needs of 1.3 billion people. In Pakistan, the per capita water availability had plummeted from 5,600 cubic metres in 1947 to roughly 930 by 2023.
Some weeks ago, we argued in these pages for bringing compute to where Pakistan’s surplus electricity sits, thereby monetising capacity that consumers already pay for. Water is integral to this equation, and a deeper analysis is warranted, which we aim to provide here.
How data centres keep cool: Data centres are usually cooled through open evaporative systems, which results in most of the water being lost to the atmosphere. However, these systems are comparatively cost-effective and relatively energy-efficient. Other options include closed-loop systems that recirculate water or a coolant through sealed pipes, which can cut freshwater use by up to 70 per cent; and direct-to-chip and immersion cooling, which bring liquid straight to the processor. Additionally, dry cooling uses almost no water but, by Google’s own estimate, about 10pc more energy. So there is a trade-off in cost and energy, but for a country like ours that already pays for idle generation, it is a strategic investment choice.
Running the numbers: A 2016 US Department of Energy study puts average on-site data centre water use at 1.8 litres per kWh. At that rate, a 1 GW campus running continuously would consume about 16 million cubic metres a year, roughly 0.01pc of the 175 billion cubic metres the Indus supplies annually. Sky47’s 8.5MW Karakoram-01 would, based on these estimates, use around 134,000 cubic metres a year. Even accounting for Pakistan’s hotter temperatures, these figures remain marginal. Hence, nationally, data centres are not Pakistan’s water problem.
Thirsty infrastructure is workable when siting follows the water.
However, locally they can be. A 1 GW campus would draw about 10 million gallons a day from a single source, the same volume that Karachi’s new Rs14bn Dumlottee pipeline will add for the Defence and Clifton areas, in a city receiving 550-650 million gallons daily against a demand of around 1,200 million gallons. The burden falls on the nearest communities.
Pakistan has faced this question of siting before with its nuclear power plants. Reactors with cooling towers consume about 2.5 litres per kWh, according to a 2012 review by the US National Renewable Energy Laboratory. At 6,755 GWh generated by Karachi’s K-3 unit in 2024, by the Pakistan Atomic Energy Commission’s count, that’s about 17 million cubic metres a year, or about 12 million gallons a day, more than a 1 GW data centre campus. Pakistan avoided that burden through siting: K-2 and K-3 sit on the coast and cool with seawater, while Chashma draws on the Chashma-Jhelum link canal. Thirsty infrastructure is workable when siting follows the water.
The problem at the foundation: Pakistan’s water crisis lies elsewhere. More than 93pc of its water goes to agriculture, and over 60pc of that is lost in conveyance and application. Four crops (rice, wheat, sugarcane, and cotton) consume over 80pc of water resources yet contribute about 5pc of GDP; Pakistan generates just $1.4 of GDP per cubic metre withdrawn, the eighth-lowest in the world. The parallel with power is striking: consumers pay for electricity that never reaches them, and the country withdraws water that never reaches a crop.
The way forward: Three interventions follow:
1) Immediate: make water a licensing condition, with energy-compute zones mandating closed-loop or dry cooling, public reporting of water use, and siting away from stressed aquifers. Dubai has promoted recycled water in district cooling since 2008, and Qatar has long incentivised non-potable cooling.
2) Short-term: cool with wastewater. Pakistan treats barely 1pc of its collected wastewater. Meanwhile, Chennai’s tertiary treatment plants are designed to recycle a fifth of the city’s sewage for industry, and Amazon, which air-cools its Indian data centres, will use treated sewage in upcoming facilities in Mumbai and Hyderabad. Operators could co-finance similar plants in Karachi and Lahore.
3) Long-term: reform agricultural water use. Punjab’s World Bank-backed irrigation programme improved over 15,000 watercourses while promoting drip irrigation and laser-levelling. Scaling it nationally and shifting acreage from rice and sugarcane towards pulses, vegetables, and sunflower would save water and cut edible oil imports. Saving just 1pc of the Indus water used for irrigation would supply a 1 GW campus for a century.
Weighing both sides: Both sides of this debate hold part of the truth. Critics are right that a poorly sited data centre can strain a community; advocates are right that the sector will not exhaust a country. Pakistan has missed too many economic waves to let a misframed debate cost it this one. The power and water are there to support the data centre industry. What remains is the discipline to manage matters both wisely and efficiently.
The writers are Fulbright scholars.
shahzaibsaqibwarraich@gmail.com
muhammadhassandajana@gmail.com
Published in Dawn, October 19th, 2026
