The data centre boom driven by AI could place major pressure on power grids, water supplies and land use. 

That’s the warning of a new report from the United Nations University Institute for Water, Environment and Health. It warns that data centres powering AI could consume 945 TWh of electricity by 2030 globally, which is nearly triple the combined annual electricity use of Pakistan, Bangladesh and Nigeria – countries collectively home to more than 650 million people.

The associated water footprint is projected to reach 9.3 trillion litres, equivalent to the basic annual domestic water needs of 1.3 billion people in Sub-Saharan Africa, while the land footprint could exceed 14,500km².

While many will have heard these warnings before, the report goes further to put a spotlight on the data centre industry. It noted that many in the industry may be measuring the wrong thing, with many sustainability assessments still focused too narrowly on carbon emissions.

Additionally, the report warned of potential greenwashing from the data centre industry. It noted that switching from coal to bioenergy, something that has been heralded by the industry as a way to cut carbon emissions, may not be as environmentally sound as it may seem. That’s because while it’s certainly ‘low carbon’, that doesn’t mean it’s ‘low water’ or ‘low land’ 

That is a problem for an industry already under pressure to demonstrate its sustainability credentials, particularly as AI drives higher-density facilities, greater power demand and more urgent grid connection requirements.

“This report is not a case against artificial intelligence, a technological transformation that is improving the lives of billions of people around the world,” said Professor Kaveh Madani, Director of UNU-INWEH who led the investigation team.

“It is a call for using it responsibly and addressing its unintended impacts proactively to make it sustainable and equitable. We have a narrow window to ensure that the backbone of the technological revolution of our era develops within planetary limits, and that the communities who provide the critical minerals for advancing AI and the ones that host its infrastructure and e-waste are also among those who benefit from it.”

Inference is the real problem

Much of the public debate around AI’s environmental impact has focused on the energy required to train large models. But the report argues that this is no longer where the main burden sits.

Once an AI model is deployed, the day-to-day process of running it – known as inference – becomes the dominant source of energy use. The report estimates that inference accounts for 80% to 90% of total AI energy demand.

That matters because inference is driven by user volume. The report estimates that ChatGPT alone processes around 2.5 billion prompts a day, translating to roughly 383 GWh of electricity a year for a single product.

The type of AI workload also makes a significant difference. A typical conversational query is estimated to be around 200 times more energy-intensive than basic text classification, while a single AI-generated image can require around 1,450 times that baseline. A short AI-generated video can consume as much electricity as 200,000 spam classifications.

For data centre operators, that means the environmental impact of AI will not just be determined by PUE, cooling efficiency or renewable energy procurement. It will also be shaped by model choice, product defaults, prompt length, output resolution and whether users are being routed to unnecessarily heavy AI systems for relatively simple tasks.

That could become one of the defining tensions in AI infrastructure. Better chips and more efficient models may reduce the footprint of individual tasks, but cheaper and more accessible AI could simply drive much higher overall usage.

The siting challenge

The report also highlights the local impact of global AI demand. Data centres may serve users across borders, but the pressure on electricity grids, water supplies and land is felt in specific communities.

Ireland is cited as one of the clearest examples of what can happen when data centre growth outpaces infrastructure planning. Data centres accounted for 21% of total metered electricity use in the country in 2023, exceeding all urban households, while the national grid operator has paused new approvals around Dublin until 2028.

The report also points to Querétaro in Mexico, where expanding compute infrastructure is drawing on water supplies amid prolonged drought, and Uruguay, where plans for a water-intensive data centre coincided with a drought that depleted Montevideo’s freshwater reserves.

The report calls for Governments to integrate AI infrastructure into energy planning, water governance and land-use permitting. It also recommends standardised reporting across carbon, water and land footprints, rather than relying on carbon metrics alone.

For operators, the message is fairly clear. Siting, energy procurement, cooling strategy and workload management are no longer just technical or commercial decisions. They are environmental footprint decisions.

The data centre industry has long argued that it is essential infrastructure for the modern economy, and the rise of AI will only strengthen that case. But this report is a reminder that essential infrastructure still has to be planned, powered and cooled somewhere.

That matters because while operators have made repeated commitments to improve their sustainability credentials, public scepticism has not gone away. If the sector cannot show meaningful progress — or at least improve its PR — opposition to new developments is likely to grow. That could become a very real barrier to expansion, particularly as more local authorities consider restrictions on data centre development.