Why AI electricity consumption could soon rival entire nations

The tech industry frequently defends the resource demands of modern data centers by arguing that artificial intelligence will naturally require less power as models become highly optimized. But a stark new United Nations report exposes the flaw in this logic, warning that AI electricity consumption is poised to double by 2030. Instead of reducing the strain on our power grids, technological improvements are actually expected to drive a massive surge in adoption, pushing the environmental cost of these systems to unprecedented heights.

The Paradox Fueling AI Electricity Consumption

Economists call it the “Jevons paradox”—a phenomenon first observed in 19th-century coal usage where increasing efficiency actually leads to greater total resource demand.

As AI models become cheaper and more capable, they are integrated into far more applications and used at much higher volumes. This explosive growth effectively erases any energy savings gained from hardware or software efficiency.

Emissions, Water, Land

By 2030, the UN estimates that the energy required to power these models could account for up to 3% of the world’s total electricity. The physical footprint is equally alarming:

Offsetting the resulting carbon footprint would require growing 6.7 billion trees over ten years. Cooling these sprawling data centers could devour 9.3 trillion liters of water—more than the annual drinking water requirements of the entire human population.

Future data centers could occupy a landmass nearly ten times the size of Mexico City.

The Widening Environmental Divide

Beyond the staggering resource drain, the AI boom is deeply inequitable. Currently, just 32 nations host specialized AI cloud infrastructure, with a massive 90% of that capacity concentrated in the United States and China.

This creates a stark divide: a handful of nations build, control, and profit from the technology, while developing countries often shoulder the environmental burdens of the raw mineral extraction and eventual e-waste disposal required to sustain it.