Data Center Energy Use Rivals Nations, Set to Double by 2030

Global data center footprint reaches national scale

Data centers worldwide consumed 448 trillion watt-hours of electricity last year—more than all but 10 countries—according to a United Nations University report released this week. That power consumption generated approximately 208 million tons of carbon dioxide, matching Argentina's annual emissions, while requiring 1.2 trillion gallons of water for energy production.

The environmental impact is projected to double by 2030, when data centers will account for nearly 3% of global electricity demand at 935 trillion watt-hours. If treated as a single entity, data center power consumption would rank sixth-highest worldwide, producing 440 million tons of carbon dioxide annually.

Why it matters

This marks the first comprehensive UN assessment of AI's environmental costs, providing business leaders with concrete metrics to evaluate infrastructure decisions. As enterprises accelerate AI adoption, understanding the resource implications becomes critical for sustainability planning, regulatory compliance, and stakeholder accountability. The scale revealed—comparable to entire nations—suggests AI infrastructure demands will increasingly compete with other societal needs for electricity and water.

AI drives accelerating demand

Artificial intelligence currently accounts for 20% of data center energy consumption, but researchers project that share will reach 40% by 2030. The growth stems primarily from operational queries rather than model training, which represents only 10% of total power use.

Kaveh Madani, study co-author and director of the UN University Institute for Water, Environment and Health, emphasized the physical reality behind seemingly virtual operations. "AI is not just a virtual thing. We're talking about something that has physics, something that has real impacts," he said.

The energy intensity varies dramatically by task. A typical ChatGPT-style query consumes 200 times more power than basic email spam filtering. AI-generated images and video require substantially more energy still. Training newer models amplifies consumption further—GPT-4 used 50 to 70 billion watt-hours for training compared to GPT-3's 1.3 billion.

Efficiency paradox and transparency gaps

While individual AI operations become more efficient, total energy consumption continues rising as usage expands—a pattern researchers call the efficiency paradox. When renewable energy powers data centers, it depletes clean electricity supply, potentially forcing other users toward dirtier sources.

The report identified significant transparency problems, with many companies declining to disclose data center locations, sizes, or consumption figures. "We cannot manage what companies do not disclose," said Fengqi You, a Cornell University energy engineering professor.

Industry representatives acknowledged environmental concerns. Josh Levi, president of the Data Center Coalition, stated the sector "remains committed to working with policymakers, local communities, and industry partners to ensure that as data centers grow, they do so responsibly."

Users can reduce AI's energy appetite through more concise queries. The report found that cutting word count by 30% reduces energy consumption by 25%—equivalent to annual electricity use for 700,000 people in Africa.

Jean Su, director of the Energy Justice Program at the Center for Biological Diversity, called the assessment "the first U.N., or even global, report that shines a light on the environmental harms of AI."

These findings were first reported by the Associated Press based on the United Nations University report.