The algorithmic thirst — AI’s growth demands a unified approach to water and energy

As the artificial intelligence (AI) revolution accelerates, we must confront a largely overlooked reality: AI is not only a digital force, but also a physical one, with an immense appetite for energy and water.

Every chatbot query, language model or automated service we deploy is powered by servers that run hot, draw electricity and demand cooling. Water, energy and computing are no longer separate systems in this new era. They are a tightly bound nexus: interdependent, dynamic and increasingly vulnerable.

We must move beyond siloed thinking and embrace a systems-level approach to govern this complex reality. One powerful lens is coupled dynamic systems, frameworks that study how interconnected elements influence one another over time.

This lens reveals the hidden feedback loops, tipping points and emergent risks that shape the water, energy and computing nexus in the AI age.

Consider the feedback loop at the heart of this system: surging AI demand increases energy consumption; this energy, often generated in thermoelectric plants, requires vast volumes of water for cooling.

Conversely, water scarcity, whether due to climate shocks or overuse, can disrupt energy production and, in turn, computing operations. A data centre without reliable cooling is a liability, not an asset. Thus, efforts to improve AI energy efficiency could backfire if they intensify water stress, especially if new hardware designs require more water-intensive materials or manufacturing processes.

Equally urgent is the risk of non-linear disruption. In coupled systems, slow changes can yield sudden breakdowns. A steady rise in AI infrastructure in a water-stressed region could abruptly trigger local ecological collapse or energy shortages.

Cooling failures could cascade into server crashes, service outages or grid instability. These are not theoretical risks, but real threats unless governance becomes anticipatory, informed by predictive analytics and real-time monitoring.

Furthermore, optimising one component in isolation, for example, reducing AI’s power draw, will not suffice. Proper system-wide optimisation requires aligning infrastructure design, energy sourcing and water use.

That means colocating data centres with low-water renewables like wind or solar, deploying cutting-edge cooling technologies and designing chips that minimise both energy and thermal output.

It also means planning across the entire AI lifecycle, from training to disposal, with water and energy impacts in mind.

Emergent behaviours also present a challenge. A large AI hub might stress a local grid, disrupting water pumps or treatment systems reliant on electricity. Such interdependencies demand interdisciplinary and adaptive governance models. Engineers, ecologists, AI developers and policymakers must collaborate, not in parallel, but in concert, to mitigate systemic risks and build intelligent resilience.

Resilience is paramount. Climate-induced shocks, like heatwaves or droughts, can simultaneously spike cooling needs and shrink water supplies.

To withstand environmental variability, resilient infrastructure must be distributed, modular, and flexible, drawing on diverse energy sources and low-water designs.

Crucially, we must distinguish between two related but distinct governance challenges. First, there is AI’s water-energy-computing footprint, the resource cost of AI itself. Addressing this requires environmentally responsible infrastructure planning and technology choices.

Second, there is AI for the water-energy nexus, using AI to optimise water and energy systems. This includes smart grids, drought prediction and real-time leak detection. One challenge is about reducing harm; the other is about increasing value. We must pursue both together.

This dual perspective has been central to the work of the United Nations University in championing the application of systems thinking to AI, drawing from the combination of control theory and artificial intelligence to understand and model complex, interdependent networks.

This body of work highlights how AI must be governed, not just for performance, but for its broader ecological and social impact, especially in the Global South.

Nowhere is this more relevant than in Africa. Across the continent, rapid urbanisation, growing populations and fragile infrastructure amplify the stakes. Many countries are beginning to harness AI for agriculture, public health and governance, but these applications often rely on precarious energy and water systems.

Without integrated planning, digital expansion could exacerbate environmental strain and deepen inequality.

Yet Africa also has an opportunity to leapfrog to a model of inclusive, sustainable AI governance. Policymakers can prioritise off-grid renewables, support equitable access to computing power and deploy AI to modernise water and energy systems.

By applying the principles of coupled systems, feedback awareness, anticipatory governance, resilience and holistic optimisation, Africa can build an AI future that is both innovative and ecologically responsible.

As the algorithmic age deepens, so must our ecological wisdom. AI’s promise will not be fulfilled through speed alone, it must also be guided by sustainability.

Let us ensure that the lifelines of water and energy, now inextricably linked with the future of computing, are protected, balanced and governed with foresight.

Only then can we say that intelligence, artificial or otherwise, is truly aligned with planetary wellbeing. DM