A UN report found that in 2025, the world’s data centres had a combined water footprint of 4.5 trillion litres — enough to satisfy Chennai’s water needs for about 8.8 years. By 2030, that number is expected to hit 9.3 trillion litres — roughly the annual basic water needs of 1.3 billion people in Sub-Saharan Africa.

And this is just the water story. There’s also electricity. By 2030, data centres are expected to consume 945 terawatt-hours of electricity annually — nearly triple the combined yearly power use of Pakistan, Bangladesh, and Nigeria, three countries that together have over 650 million people.

This is the invisible bill of the AI revolution.

The Men Building This World

To understand why data centres are multiplying so fast, follow the money — and the ambition.

In 2025, the four biggest tech companies — Google, Meta, Amazon, and Microsoft — spent $364 billion on data centre infrastructure. That’s nearly India’s entire annual central government expenditure. The race isn’t just about profit. It’s about who controls the infrastructure of the future.

The United States: AI Superpower, Grid Under Stress

America runs more AI than any other country. Its data centres currently consume about 4.4% of all US electricity, and that share is expected to jump to somewhere between 6.7% and 12% by 2028. By 2030, energy demand from AI infrastructure alone could be 130% higher than today.

The problem? The US power grid wasn’t built for this. Getting permission to build a new power plant or transmission line can take over a decade. In Northern Virginia, which hosts the highest concentration of data centres on Earth, some projects are facing grid connection queues of 7 to 13 years. The world’s AI ambition is outrunning America’s ability to actually power it.

Tech giants are already looking for unusual solutions. Microsoft is restarting the Three Mile Island nuclear plant — yes, the one famous for America’s worst nuclear accident in 1979. Nuclear power is gaining traction fast, because solar and wind alone can’t keep up with demand that is growing 30% annually.

India: Digital Ambition Meets Water Bankruptcy

India tells a particularly sharp story — because the stakes here are uniquely high.
The country extracts more groundwater than any other nation on earth, accounting for roughly 25% of all global groundwater withdrawals. It has 18% of the world’s population, but only 4% of its freshwater. Already, 600 million Indians face high to extreme water scarcity.
Into this already stressed system, India is now adding a massive data centre boom. The country’s capacity has grown from about 375 MW in 2020 to 1,500 MW in 2025, with projections pointing to 13,560 MW by 2031-32. Every megawatt of data centre capacity requires approximately 2.5 crore litres of water per year. At 1,500 MW, India’s data centres are already consuming roughly 37.5 billion litres annually. Morgan Stanley estimates that global AI data centre water demand could rise elevenfold by 2028.

The geography of this expansion makes it worse. The biggest hubs are in Maharashtra, Karnataka, Tamil Nadu, and Telangana — all states that already experience severe water stress. Mumbai, where over 71% of residents receive water for less than 4 hours a day, is set to host 45% of India’s planned data centre capacity. In Bengaluru, water already has to be pumped across significant elevations from the Cauvery River — a river at the centre of a decade-long interstate dispute.

In Gautam Buddha Nagar, Uttar Pradesh — which hosts at least 17 data centres — groundwater extraction is already at 104.79% of recharge capacity. The district is drawing down its water table every year. And yet, the state’s 2026 data centre policy promises another 2 GW of capacity worth ₹6 lakh crore.

Protests have erupted in Visakhapatnam over the Google-AdaniConneX data centre, and in Thane against Amazon’s hyperscale project. But protests alone won’t rewrite policy.

Europe: Regulation Awakens, But Slowly

Europe is home to some of the world’s most aggressive climate legislation, but it is only beginning to seriously address the resource costs of AI infrastructure. In Ireland, data centres consumed 21% of all national electricity in 2023 — more than all urban homes combined.

The Irish grid operator paused approvals for new data centres until 2028. In Frankfurt, data centre demand has reached 42% of local electricity load. In Dublin, it’s nearly 80%. The Netherlands imposed a nine-month moratorium on hyperscale permits.

In Spain’s Aragón region, Amazon Web Services data centres use up to 53.9 million litres of water annually — and when audited, Amazon admitted it had been underestimating its usage and sought a 48% increase in water permits. The EU AI Act now requires energy consumption reporting for large AI models, but water disclosure remains largely voluntary.

Southeast Asia and Africa: The New Frontiers — and the New Risks

As land, power, and political pressure mount in the US and Europe, tech companies are flooding into Southeast Asia and Africa — often to regions with weaker regulations and less organised public resistance. In Southeast Asia, data centre investment reached $13.71 billion in 2024 and is projected to hit $30.47 billion by 2030.

Bangladesh tells the starkest story: it is receiving $290 million in data centre investments while over 40% of its population lacks safe drinking water. Dhaka’s water table is falling by more than 3 metres per year in some areas, and farmers who once struck water at 150 feet now have to drill to 300 feet. Despite Bangladesh’s own Water Act prioritising drinking water over industrial use, permits are routinely granted in violation of that hierarchy.

In Brazil, Scala is building an “AI City” with 4.75 GW of capacity in Eldorado do Sul — classified as high water stress risk. Indigenous communities, Landless Workers’ Movement settlements, and organic farmers are reporting compacted, drought-cracked soil and disappearing water sources. Tribal councils have unanimously rejected related dam proposals. Construction continues regardless.

Analysts have begun calling this pattern “digital colonialism” — Western corporations extract resources from Global South nations while local communities absorb the environmental costs.

Who Bears the Cost? The Most Vulnerable, Always.

Across every country, every region, every continent, the pattern is the same: the benefits of AI go to the educated, the connected, the wealthy — and the costs are borne by those least responsible for creating them.

The Energy Emergency Nobody Is Talking About

By 2030, data centres will need 945 TWh of electricity — triple the combined consumption of Pakistan, Bangladesh, and Nigeria. There’s a painful irony hiding in the numbers too: Google reduced the energy used per prompt by 33 times through efficiency improvements.

But its total energy consumption grew by 27% year-on-year. This is known as Jevons Paradox — when you make something more efficient, people use so much more of it that total consumption rises. Efficiency gains are being overwhelmed by demand growth of 30% annually.

Meanwhile, only 32 countries on earth host AI-specialised data centres. Over 150 countries — mostly in Africa, South Asia, and the Pacific — have no sovereign AI compute infrastructure at all. They are permanently dependent on technology built, owned, and controlled by others. Their data flows through someone else’s servers. Their digital futures are held in someone else’s buildings.

The One Fix Everyone Agrees On, But Nobody Is Doing

Experts across the UN, IEA, academic institutions, and civil society agree on one urgent step: mandatory disclosure.

Right now, most data centres don’t have to tell anyone how much water they use, where they source it from, or whether they’re drawing down a stressed aquifer.

No public dashboard. No annual report. No accountability. Shaolei Ren of UC Riverside, the researcher who first quantified ChatGPT’s water footprint, says what’s needed is simple: force data centres to publicly disclose monthly and peak water use, the source — potable, reclaimed, or groundwater — and whether the site is in a water-stressed zone.

The EU is moving in this direction. But globally, most frameworks remain voluntary. And voluntary means nothing gets disclosed unless it looks good.

Until that changes, the AI revolution will keep writing invisible bills — and handing them to people who never asked for the service.