Smart, thirsty machines: how to govern AI without draining the planet

Smart, thirsty machines: how to govern AI without draining the planet

By Fatima Ali, Consultant – Board Services – PICG

The conversation around AI often amazes us with futuristic possibilities, but what is less dazzling is the environmental bill that comes with every interaction. AI is, in effect, a resource intensive exercise: data centers need water to cool servers, electricity to keep them running, and minerals to manufacture the underlying hardware. If AI is the brainchild of our century, then its environmental footprint is the messy home that nobody wants to clean up.

We cannot sensibly slow down technological progress, nor can we allow data-center growth to drain our local clean water supplies. The question is not whether we choose AI or sustainability, but how we ensure both can coexist albeit sustainably, equitably, and transparently.

Consider water, for instance. A 2023 study from the University of California, Riverside, and the University of Texas, Arlington, found that training GPT-3 consumed an estimated 700,000 liters of freshwater equivalent to producing 370 BMW cars or 320 Tesla vehicles. Even a single chatbot conversation is not free; Microsoft’s water consumption rose by 34% between 2021 and 2022, much of it attributed to AI workloads. Google reported similar jumps, with its Iowa data center alone drawing nearly 1.5 billion gallons of water in 2022 to support AI training. If the irony isn’t sharp enough, think of this; while some regions are in the grip of water scarcity, our AI companions are, quite literally, being kept cool by sipping from the same dwindling supply.

Good thing is that innovation is emerging. Microsoft and other providers are rolling out new chip-level cooling systems and “zero-water” designs for AI workloads. What lags is transparency. Policymakers are beginning to demand mandatory reporting of water and energy use at data centers so that citizens are not left guessing who is drawing from their clean drinking water. This matters from governance POV because water is inherently local, even if AI feels global. A model trained in Karachi or California may serve the same chatbot, but its impact on surrounding communities is not the same. Boards tasked with ESG oversight cannot treat this as a distant technical detail; they must see it as a direct resource risk with social and reputational consequences. Opacity here is a governance failure. If companies refuse to disclose water use by site or workload, regulators and citizens cannot plan. On the bright side the problem is solvable, but only if boards, policymakers, and technology leaders act together.

In Pakistan, the challenge takes on a sharper edge. The World Bank has already warned that the country could become “water scarce” by 2025. As the state champions digital transformation and cloud adoption, governance cannot afford to lag. Regulators such as SECP and SBP could extend ESG reporting requirements to include water use in digital infrastructure. Boards of listed companies and state-owned enterprises can add oversight questions on the environmental footprint of their digital expansion. Public–private partnerships could ensure that any new data-center development also invests in local water resilience, such as recycling, leak repairs, or replenish groundwater. Just as Pakistan has imported international best practices in financial reporting, it can adopt global benchmarks for data-center disclosure and water stewardship.

The trick, of course, is to avoid narrow metrics. Some cooling methods save electricity but consume more water, and vice versa. Optimizing for one number, such as power efficiency, while ignoring water security risks starving a community. Policy nudges can help require public reporting of water and energy use by site, tie tax incentives to verifiable savings and local benefits, and create a “digital passport” for large AI models that discloses their energy and water footprint alongside where they were trained. Think of it as a nutrition label for AI – listing the hidden ingredients of energy and water.

AI is too valuable to stumble upon and too resource-intensive to ignore. The technology side is moving fast, with zero water cooling already rolling out, but without oversight, the hidden costs will land on the most vulnerable communities like ours. Pakistan, facing both water scarcity and a digital transformation, has a chance to lead by embedding water conscious AI governance into its corporate and regulatory frameworks. That way, the country keeps the promise of AI without draining its rivers dry and boards everywhere learn that smart machines can thrive without making us thirstier.

Sources

• Ahmed, S., et al. (2023). Making AI Less Thirsty: Uncovering and Addressing the Secret Water Footprint of AI Models. University of California, Riverside & University of Texas, Arlington. [2304.03271] Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models
• Strubell, E., Ganesh, A., & McCallum, A. (2019). Energy and Policy Considerations for Deep Learning in NLP. MIT & UMass Amherst. [1906.02243] Energy and Policy Considerations for Deep Learning in NLP
• World Bank (2019). Pakistan: Getting More from Water. World Bank Document