The United Nations University published a report on June 3 that is the first UN study to quantify artificial intelligence's water and land footprints alongside its carbon footprint. The headline numbers: data centers consumed 448 terawatt-hours of electricity in 2025 — the 11th-largest "country" if counted separately — and will consume 945 TWh by 2030, triple the combined use of Pakistan, Bangladesh, and Nigeria. Water consumption is projected to reach 9.3 trillion liters annually by 2030, equal to the basic domestic water needs of 1.3 billion people in Sub-Saharan Africa. [1]
The report, compiled by the UNU Institute for Water, Environment and Health, found that day-to-day AI usage — not training — accounts for 80-90% of energy demand. Training a large language model is resource-intensive; running billions of queries each day is more so. The AI market is expected to grow 25-fold in the coming decade, from $189 billion in 2023 to nearly $5 trillion by 2033. [1]
"This report is not a case against artificial intelligence," Professor Kaveh Madani, the report's lead author, said. "It is a call for using it responsibly." [2]
The land footprint is 14,500 square kilometers by 2030 — twice Jakarta's metropolitan area. Different data-center hubs have dramatically different water-to-carbon ratios, making location a governance variable. A data center in a water-stressed region consumes a different resource than one in a water-abundant region, even if their electricity consumption is identical. [1]
What carbon-only ESG reporting misses: a company can purchase renewable energy credits to offset its electricity consumption while drawing millions of liters of water from aquifers that communities depend on for drinking water [3]. The UNU report makes the water footprint visible as a separate metric — and the comparison to Sub-Saharan Africa's domestic needs is the framing that forces the accountability question.
The paper's ai-state-power thread has tracked the factory, financing, and control-plane records. This report adds the resource-consumption layer — the physical infrastructure that AI depends on, and the communities that compete with data centers for water. The 9.3 trillion-liter projection is not an abstraction. It is a volume of water that 1.3 billion people need to drink.
The report finds that different data-center hubs have dramatically different water-to-carbon ratios. This means the governance variable is not just how much water AI consumes, but where it consumes it. A data center in Arizona draws from a different aquifer than one in Norway. The environmental cost is local even when the AI service is global.
The gap between carbon-only reporting and water-inclusive accounting is the gap the paper tracks. ESG frameworks measure what is easy to measure. The UNU report measures what matters.
-- KENJI NAKAMURA, Tokyo