Feature
A sharp opinion on a topical issue. This edition: Xin Sun.

The rise of artificial intelligence (AI) has been nothing short of revolutionary, transforming industries, redefining productivity, and reshaping how societies function. Data centers are the invisible backbone of the AI economy, but their environmental impact is anything but subtle. In 2025, global data centers consumed 448 TWh of electricity, more than the entire annual residential electricity needs of Sub-Saharan Africa’s 1.3 billion people. By 2030, this figure is projected to nearly double to 945 TWh, accounting for almost 3% of global electricity use. AI workloads alone, which already account for 20% of data center electricity use, could reach 378 TWh by 2030. Meanwhile, AI inference, the continuous generation of responses for billions of daily interactions, now accounts for 80–90% of total AI energy use. In Ireland, data centers already account for 21% of national electricity demand, prompting grid operators to pause new connections due to capacity constraints.

Water and land footprints are also alarming. In 2025, data centers’ electricity consumption carried a water footprint of 4.5 trillion liters, which enough to meet the annual basic domestic needs of over 620 million people in Sub-Saharan Africa. The land footprint was 6,900 km², nearly 4.5 times the size of Greater London. Training GPT-4 alone required 600 million liters of water and 0.9 km² of land, equivalent to 237 Olympic pools or 126 football fields. These resources could have met the annual needs of tens of thousands of people, yet were consumed in a matter of months for a single AI model. In water-scarce regions like Querétaro, Mexico, expanding AI data centers are exacerbating droughts, prioritizing industrial demand over human consumption.

Voluntary efficiency improvements are insufficient. The Jevons Paradox where increased efficiency leads to greater overall consumption is already in play. A single AI-generated video can consume as much electricity as 200,000 spam classifications. Moreover, renewable energy alone is not a panacea. Hydropower-dominated grids can have water footprints 30 times greater than coal, while bioenergy may lower carbon emissions but increases land use by 100 times.

Stricter regulations are critical. Governments must enforce mandatory disclosure of energy, water, and land footprints for data centers, with standardized reporting, real-time monitoring, and third-party verification to prevent greenwashing. Siting restrictions should not only prevent construction in water-scarce or high-carbon regions but also in areas facing grid instability or ecological vulnerability, while granting local communities veto power over projects that threaten their resources. Resource budgets such as annual caps on GPU-hours, electricity, and water consumption must be tied to regional carrying capacities, with penalties for exceedances and incentives for efficiency. These measures should be paired with efficiency mandates, requiring data centers to meet minimum performance standards for power usage and cooling, alongside comprehensive e-waste management plans to address hardware lifecycle impacts.

The window for action is closing. Without binding regulations, the environmental footprint of AI will continue to grow, deepening global inequities and pushing planetary boundaries to the brink.

Xin Sun is a senior lecturer at the Energy and Sustainability Research Institute Groningen (ESRIG) at the University of Groningen.