When AI drinks our water

In the Global South, data centres and large language models can strain fragile water and power systems while the real benefits flow offshore.

Article link: https://open.substack.com/pub/johanosteyn/p/when-ai-drinks-our-water?r=73gqa&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Audio summary: https://youtu.be/NLgzHrqVtKc

Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/

I write about various issues of interest to me that I want to bring to the reader’s attention. While my main work is in Artificial Intelligence and technology, I also cover areas around politics, education, and the future of our children.

Most people think of AI as something abstract that lives “in the cloud”. We picture friendly chatbots and clever tools that help us draft emails or summarise documents. We rarely picture dams, cooling towers, power lines and substations. Yet that is where AI actually lives: in physical data centres that consume electricity and water at an industrial scale.

Recent analyses from MIT and the United Nations Environment Programme make it clear that generative AI is already adding a measurable load to global energy demand and water use. It is still early days, but the pattern is familiar: infrastructure and risk are local, while ownership and profit are often remote. For countries in the Global South, including South Africa, the question is not whether we should embrace AI, but whether we understand who pays the environmental bill.

CONTEXT AND BACKGROUND

In 2025, MIT researchers explained that a single AI request can use many times more electricity than a conventional web search, and that the biggest models require vast amounts of power and cooling water during training and everyday use. At the same time, a report by the Environmental and Energy Study Institute in the United States noted that a large hyperscale data centre can consume millions of litres of water per day for cooling, comparable to the needs of a medium-sized town. These are not abstract numbers. They translate into real pressure on rivers, aquifers and power grids.

The “one prompt equals one glass of water” claim that has circulated online is an exaggeration. Newer studies suggest that a short interaction with a language model is closer to a few millilitres – teaspoons, not a full glass. The problem is scale. Billions of prompts, continuous training of ever-larger models, and a global race to build new facilities turn those teaspoons into lakes and reservoirs. A joint report by UNESCO and University College London in 2025 showed that relatively small design changes can cut the energy footprint of large language models dramatically, which tells us something important: none of this is inevitable. It is the result of design choices, business incentives and policy decisions.

INSIGHT AND ANALYSIS

In the Global South, these decisions often happen far from the communities that live with the consequences. Data centres are sited where land is cheap, regulations are loose, or governments are desperate for foreign investment. Local residents are promised jobs and “innovation”, but the highly skilled work tends to stay with foreign vendors, while the long-term environmental risks are quietly localised. In regions already grappling with unreliable electricity and water scarcity, plugging an energy-hungry, water-thirsty facility into the system can make existing vulnerabilities worse.

We should also be honest about who really benefits. The language models running in those data centres mostly belong to firms headquartered in the Global North. The data that trains them often comes from users all over the world. Yet the economic value – from subscription fees to the productivity gains of wealthy clients – typically flows back to a small number of technology giants and their shareholders.

This is not simply a story about carbon and water; it is a story about power and ownership. We risk replaying old extractive patterns: our minerals for chips, our land for data centres, our water for cooling, and our data for training – all in service of someone else’s digital empire.

IMPLICATIONS

For policymakers and business leaders in South Africa and across Africa, responsible AI cannot only mean avoiding bias or protecting privacy. It must also mean asking hard questions about infrastructure. Who decides where data centres are built? What guarantees exist for local water security, energy stability and environmental monitoring? Are communities consulted meaningfully, or simply informed after deals are signed? Environmental impact assessments, public participation and transparent reporting must keep pace with the speed of AI investment.

Companies that adopt AI also have a role to play. Boards should be asking their technology providers about the energy mix of the data centres they use, the cooling technologies employed and the environmental standards they meet. They should be favouring more efficient models and being honest about when a heavyweight AI system is truly needed and when a simpler tool will do. Governments, in turn, should align AI strategies with climate and water strategies. There is little sense in celebrating digital transformation if it silently undermines our resilience to drought and loadshedding.

CLOSING TAKEAWAY

The promise of AI for the Global South is real: better health systems, smarter agriculture, and more efficient public services. But these benefits will not come for free, and they will not necessarily flow to those who bear the environmental cost. When AI drinks our water, we should at least know how much, from where, and in whose interest. The task before us is not to reject AI, but to insist that its physical footprint is visible, measured and governed.

That means demanding transparency from technology providers, integrating environmental concerns into AI policy, and giving local communities a voice in decisions about infrastructure. If we fail to do that, we may discover too late that in chasing digital progress, we have quietly traded away the most basic resources our children will need to survive.

Author Bio: Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net