UNU-INWEH: By 2030, global data-centre electricity use could nearly double to 945 TWh, with water impacts comparable to the basic annual domestic needs of 1.3 billion people in Sub-Saharan Africa.

2025-11-28 · Source: Pascal’s Substack · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Environmental Science & Earth Systems · Depth: Intermediate, long

Summary

The UNU-INWEH report highlights that AI is a physical industrial system with significant environmental costs beyond carbon. By 2030, global data-centre electricity use could nearly double to 945 TWh, up from an estimated 448 TWh in 2025, potentially consuming almost 3% of global electricity. This expansion could lead to a water footprint of 9.3 trillion litres, equivalent to the basic annual domestic needs of 1.3 billion people in Sub-Saharan Africa, and a land footprint exceeding 14,500 km². The report emphasizes that inference, not just training, drives 80-90% of AI energy use, with tasks like video generation being far more energy-intensive than text classification. It also warns against the rebound effect, where efficiency gains are offset by increased total consumption, and notes growing public resistance to data centers.

Key takeaway

For Policy Makers addressing AI's environmental impact, you must move beyond carbon-only accounting and mandate comprehensive disclosure for electricity, water, land, and e-waste footprints. Implement permitting tied to cumulative environmental impact, especially in water-stressed regions, and adopt "small-model-first" procurement. Your regulations should also require user-facing environmental design and address rebound effects to ensure sustainable AI growth.

Key insights

AI's environmental footprint extends beyond carbon, encompassing vast electricity, water, land, and e-waste, driven largely by inference.

Principles

• AI sustainability requires a multi-footprint framework.
• User-interface design is now environmental policy.
• Rebound effects can negate efficiency gains.

Method

The report proposes a governance framework including transparency, efficiency by design, equity, lifecycle responsibility, global cooperation, and sustainable use.

In practice

• Require task-level inference reporting by model and region.
• Implement user-facing environmental design defaults.

Topics

• AI Environmental Impact
• Data Center Energy Consumption
• Water Footprint
• E-waste
• AI Governance
• Environmental Justice

Best for: Executive, Investor, CTO, Policy Maker, AI Ethicist, Consultant

Related on AIssential

• AI laws overlook environmental damage – here’s what needs to change — AI's substantial environmental footprint is largely unaddressed by current global regulatory efforts.
• Can you explain to me the hatred of AI and its water usage? — AI's water usage controversy centers on data center cooling demands in water-stressed regions, impacting local communities.
• UN report warns AI could soon use 3% of world’s electricity and more water than we need to drink — AI's increasing efficiency risks higher overall resource consumption due to the Jevons paradox, demanding urgent environmental governance.
• Sustainable AI: A Beginners Guide — Sustainable AI minimizes environmental harm by optimizing energy efficiency, renewable energy sourcing, and transparency.
• AAAI presidential panel – AI and sustainability — Balancing AI's transformative potential with its environmental impact is a critical challenge for sustainable development.
• Climate — Unchecked, large-scale AI development is unsustainable, bolstering the fossil fuel industry and creating a "three-legged stool" with the state.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Pascal’s Substack.