Sahar Iqbal
Akhund Forbes, Pakistan
sahar.iqbal@akhundforbes.com

Introduction

The rapid ascent of artificial intelligence (AI) is fundamentally reshaping global industry, and Pakistan is no longer a bystander to this revolution. Through its ‘Digital Pakistan’ plan, the state aims to weave cutting-edge computation into the national fabric, promising a future of optimised governance, automated fiscal management and a thriving tech ecosystem that can compete on the global stage. From AI-driven crop monitoring in the heartlands of Punjab to sophisticated Fintech solutions in Karachi, the promise of digital efficiency is undeniable. Yet, as we race towards this silicon-clad future, we are largely ignoring a fundamental biological and physical reality: intelligence, whether carbon based or silicon based, requires a staggering amount of water. While AI promises to solve our most complex societal problems, it introduces a ‘hidden’ environmental cost that may challenge infrastructure and legal frameworks.

The timing of this digital push could not be more critical or more paradoxical. Pakistan remains a permanent fixture on the list of the world’s most climate-vulnerable nations, a reality made painfully clear by the devastating floods of 2022 and the record-breaking heatwaves that followed. We are a country where the memory of water-driven catastrophe is fresh, yet we are simultaneously hurtling towards ‘absolute water scarcity’. Our per capita water availability has plummeted from 5,000 cubic meters in the 1950s to a precarious 1,000 cubic meters today. Within this context, the massive water requirements of AI, used primarily to cool the high-density servers that process complex algorithms, represent a systemic risk to the security of our national resources. This article argues that unless we align our national AI policy with our environmental and water laws, we risk building a digital economy that we cannot afford to sustain.

The status of water consumption by AI: from global to local

The environmental footprint of AI is often invisible to the end user because we interact with it through polished, weightless digital interfaces. However, the physical reality beneath the ‘cloud’ is massive and heavy. AI models, particularly large language models (LLMs), like ChatGPT or Google Gemini, require immense computational power housed in specialised data centres.^[1] These facilities generate heat so intense that they require constant, high-volume cooling to prevent hardware failure. The most common method used globally is ‘evaporative cooling’, a process where fresh water is evaporated to chill the air surrounding the servers.

To put this into perspective, peer-reviewed global research now indicates that training a single high-end AI model can consume as much as 700,000 litres of fresh water, enough to sustain a small village for an entire season. On a more granular, individual level, every set of 20 to 50 prompts sent to an AI service effectively ‘drinks’ half a litre of water. In Pakistan, we have avoided the brunt of this impact so far because our AI usage is effectively ‘outsourced’. When a developer in Lahore or an entrepreneur in Islamabad uses a cloud-based AI service, the actual water consumption is happening in data centres located in Singapore, the United Arab Emirates or the United States. We are, in effect, consuming the water of other nations to power our digital growth.

However, this ‘geographic shield’ is rapidly dissolving. As part of the Ministry of IT and Telecommunication’s (MoITT) push for ‘local data sovereignty’ and the ‘cloud first policy’, there is a concerted effort to build hyperscale data centres within Pakistan’s own borders. The goal is to reduce latency, improve national security and lower the costs for local startups. But when these massive facilities arrive, they will meet a harsh geographic reality: Pakistan’s ambient temperatures frequently exceed 45°C in the summer months. In cities like Multan, Jacobabad or Sukkur, the cooling requirements for a data centre will be exponentially higher than for a similar facility located in a temperate climate like Ireland. We are essentially preparing to host the most water-intensive industry on earth in a geography that is already running dry. Without strict regulations, these digital hubs will soon find themselves in direct competition with local citizens for the same dwindling groundwater.

The energy–water nexus: scientific realities and local blind spots

The scientific impact of AI on water in Pakistan is multifaceted, tied together by what researchers call the ‘energy–water nexus’. The impact is not just about the water used on-site at a data centre; it is also about the water used at the source of power generation. Pakistan’s energy grid relies heavily on a mix of thermal and hydro power. Thermal plants, which still form a large part of our base load, require millions of gallons of water for turbine cooling. Hydropower, while cleaner in terms of carbon, is inherently tied to reservoir levels and river flows. Therefore, every kilowatt hour used to train an AI model in a Pakistani data centre has a specific ‘water price’ paid at the Tarbela dam or at a thermal plant in Jamshoro.

Currently, Pakistan suffers from a significant scientific ‘blind spot’ regarding this digital footprint. While our national discourse is dominated by agricultural water use, which accounts for over 90 per cent of our consumption, we have almost no peer-reviewed, localised research on industrial water-use efficiency (WUE) within the burgeoning IT sector. There is a total lack of transparency regarding how much water our existing private data centres are pulling from urban aquifers. In cities like Lahore, where the water table is dropping by nearly a meter every year due to unregulated pumping, the addition of unmonitored, water-intensive digital infrastructure could accelerate the arrival of a ‘day zero’ scenario.

Furthermore, the thermodynamics of cooling in Pakistan presents a unique challenge. In high-humidity coastal cities, like Karachi, evaporative cooling is less efficient, leading to higher water waste. In dry, high-heat inland cities, the evaporation rate is so aggressive that it can place a sudden, massive strain on local municipal supplies. Scientists and hydrologists in Pakistan have yet to map these specific risks, leaving policymakers to fly blind as they approve new digital infrastructure projects. We are prioritising ‘compute’ power without calculating the ‘hydration’ cost required to keep that power running.

The regulatory vacuum: analysing the ‘applicable law’

Despite these clear and present risks, Pakistan’s legal and regulatory landscape remains remarkably silent on the environmental costs of the digital age. When we look at the ‘respective applicable laws’, we find a fragmented system that fails to recognise AI as an environmental actor. The draft National AI Policy 2023^[2] is an ambitious document focused on fostering innovation, building a ‘compute’ infrastructure and attracting venture capital. However, a close reading of the draft reveals a total absence of sustainability mandates. There are no clauses requiring data centre operators to report their WUE or to implement water-recycling technologies.

Similarly, the National Water Policy 2018,^[3] while a landmark achievement for the country, is trapped in a 20th century mindset. It rightly identifies that agriculture is the primary consumer of water, but it treats the ‘industrial’ and ‘corporate’ sectors as a vague monolith. It does not recognise ‘digital infrastructure’ as a specific industrial category with unique, high-intensity water consumption patterns. This policy gap means that data centres are currently regulated under general industrial water permits, which do not account for the specific evaporative losses inherent to server cooling.

However, we can find a foothold for regulation in the Pakistan Environmental Protection Act 1997 (PEPA).^[4] Section 11 of the PEPA prohibits the discharge of ‘effluents’ or ‘pollutants’ that exceed the National Environmental Quality Standards (NEQSs). Traditionally, this has been used to stop chemical dumping into rivers. However, legal experts argue that ‘thermal discharge’, the release of heated water back into local systems, is also a form of pollution that disrupts aquatic life and lowers water quality. Under the 18th Amendment, provincial Environmental Protection Agencies (EPAs) have the authority to mandate rigorous environmental impact assessments (EIAs). We must move towards a legal standard where no hyperscale data centre is granted a ‘no objection certificate’ (NOC) without a comprehensive ‘water scarcity impact report’.

Furthermore, the Securities and Exchange Commission of Pakistan (SECP) has a role to play. While the SECP has introduced environmental, social and governance (ESG) guidelines for listed companies,^[5] they remain largely voluntary and focused on carbon emissions rather than digital water footprints. By making water-use disclosure mandatory for the tech sector, the SECP could drive a culture of corporate accountability that aligns digital growth with the country’s environmental survival.

Targeting funding for research to save water^[6]

The objective of this critique is not to stifle AI development in Pakistan, but to ensure that it is ‘water neutral’ from the outset.^[7] This requires a radical shift in how we fund research and technology. Currently, funding from bodies like Ignite (the national tech fund) is geared towards software development and market entry. There is an urgent need for Ignite to pivot and launch a ‘sustainable digital infrastructure challenge’. This would provide seed funding for hardware startups working on ‘dry cooling’ technologies or ‘liquid immersion cooling’, a method where servers are submerged in specialised, non-conductive fluids that eliminate the need for water evaporation entirely.^[8]

The Higher Education Commission (HEC) also has a pivotal role to play. In most Pakistani universities, the computer science departments and the environmental science departments exist in silos. We need HEC-funded interdisciplinary labs that can calculate the precise ‘water price’ of different AI architectures in the Pakistani climate. We should be training a new generation of ‘green AI’ engineers who understand that the efficiency of an algorithm is measured not just in milliseconds, but in millilitres.

Beyond domestic funding, Pakistan has a strong moral and legal claim to international climate finance. Under the ‘loss and damage’ mechanisms secured at recent Conference of the Parties (COP) summits organised by the United Nations Framework Convention on Climate Change, Pakistan can argue that transitioning to water-neutral digital infrastructure is a form of climate adaptation.^[9] We should seek grants from the Green Climate Fund (GCF) to retrofit our national data centres with circular water systems.^[10] These systems treat and reuse greywater for local irrigation or urban green spaces, ensuring that the water used for our ‘intelligence’ is not taken away from our ‘subsistence’.

Finally, the government can use fiscal law as a tool for change. The Ministry of Finance and the MoITT should offer significant tax credits to any tech company that achieves a certified ‘water-positive’ status. This would incentivise companies to invest in rainwater harvesting and wastewater treatment plants at their data centre sites, returning more water to the local aquifer than they extract.

Conclusion

As Pakistan stands on the threshold of a new digital era, we must decide what kind of future we are building. If our AI models are trained at the cost of our citizens’ drinking water, the ‘Digital Pakistan’ vision will be built on a foundation of sand. The ‘hidden thirst’^[11] of AI is a scientific and thermodynamic reality that our current laws, from the 1997 PEPA to the 2018 National Water Policy, are not yet equipped to handle.

The path forward requires a ‘water-first’ digital strategy. This means creating a formal, legally mandated bridge between the MoITT and the Ministry of Climate Change. It means updating our NEQSs to include digital resource consumption. And it means ensuring that our push for ‘local data sovereignty’ does not lead to ‘local water bankruptcy’.

Pakistan has the opportunity to lead the Global South in regard to sustainable technology by building infrastructure that is resilient, transparent and respectful of our natural heritage. AI has the potential to solve many of our most pressing problems, from predicting floods to optimising energy grids, but it must not be allowed to consume the very resource that sustains the people it is meant to serve. The machines of the future must be designed to thrive in our climate without depriving our children of the water of today. Only then can we truly call our progress ‘intelligent’.