AI to Consume 3% of Global Electricity, Generate Massive E-Waste by 2030, UN Warns

The rapid expansion of artificial intelligence is on track to consume nearly 3% of the world's electricity by 2030 while generating millions of tons of electronic waste, according to a United Nations report released Wednesday.

The assessment from the United Nations University Institute for Water, Environment and Health measures the physical toll AI infrastructure is extracting from global resources.

Moving beyond standard carbon emission metrics, the research quantifies the massive amounts of land, water, and raw materials required to keep global data centers running.

Data centers currently draw an estimated 448 terawatt-hours of electricity a year, comparable to the total power consumption of France.

As artificial intelligence becomes embedded in search engines, software platforms, and media generation tools, that demand is projected to more than double to 945 terawatt-hours by the end of the decade.

The hardware required to process these workloads is creating an immense disposal problem.

By 2030, the regular turnover of obsolete servers, specialized microchips, and cooling equipment will generate up to 2.5 million metric tons of electronic waste annually.

The UN equates this volume to discarding 250 Eiffel Towers every year.

Water and land requirements are escalating alongside power demands.

The liquid cooling systems designed to prevent server farms from overheating are projected to consume 9.3 trillion liters of water in 2030.

The report notes this volume is enough to meet the basic domestic water needs of 1.3 billion people in sub-Saharan Africa.

To generate the power required for these facilities, energy infrastructure will claim roughly 14,000 square kilometers of land, an area roughly the size of Northern Ireland.

The bulk of this consumption stems from daily public use rather than initial software development.

While training large-language models requires massive energy spikes, the routine operation of deployed models (known as inference) now accounts for 80% to 90% of AI's total energy footprint.

The energy cost of individual tasks varies widely depending on the prompt.

A conventional internet query uses about 0.3 watt-hours of electricity, but an AI-enhanced generative search requires up to 3 watt-hours.

Producing a single high-resolution AI video clip consumes as much power as processing 200,000 spam emails.

The UN findings identify a rebound effect driven by hardware improvements.

As tech companies build faster, more energy-efficient processors, the cost of computing drops.

This creates higher volumes of daily use, driving total energy and water consumption up and entirely erasing the resource savings gained from hardware efficiency.

A steep geographic divide separates where AI is deployed from where its environmental footprint lands.

Roughly 90% of specialized AI computing power is concentrated in the United States and China.

However, the downstream effects, such as the mining of critical minerals, localized water stress, and the processing of hazardous e-waste, are frequently pushed to lower-income nations equipped with fewer environmental safeguards.

The United Nations is urging governments to factor data centers into national water and energy planning and to mandate standardized footprint reporting for tech companies before the resource strain overwhelms local utility grids.