A common claim to ease worries over data center energy and resource needs is that future AI models will require less as they grow more efficient. Yet a new United Nations report shows this view is misleading and details the environmental toll of AI. The study projects that by 2030 AI energy demand could double and reach 3 percent of global electricity, with emissions matching those of the United Kingdom and cooling water use surpassing yearly drinking water needs worldwide. The report also expects AI growth to follow the Jevons paradox, where efficiency gains lead to higher total resource consumption rather than lower use. Named after economist William Stanley Jevons, the pattern was first seen with coal in 19th-century England, as lower costs drove expanded demand. As AI becomes cheaper, new applications and greater volumes of use are likely to offset efficiency savings. To address this, the report offers a framework for responsible AI based on transparency, built-in efficiency, fairness, lifecycle accountability, international cooperation and sustainable practices. Last year data centers already used as much electricity as Saudi Arabia. If demand doubles by 2030, the resulting carbon output would need 6.7 billion trees grown for ten years to offset. Data centers would also need 9.3 trillion liters of water and land nearly ten times the size of Mexico City. The report notes structural imbalance, with only 32 countries hosting AI cloud infrastructure and 90 percent of capacity in the United States and China. It warns of a widening gap between nations that control AI systems and those that mainly use them, often facing heavier environmental costs from mining and waste. Two main factors shape AI impact: the amount of use and the type of tasks performed. Different tasks and model choices carry varying energy and environmental costs. Responsible AI calls for oversight across the full supply chain, from mineral extraction to disposal. Environmental reporting should become standard in AI development and be included in climate planning. In New Zealand the government has introduced a national AI strategy and public service framework, yet it lacks mandatory environmental disclosures. Australia follows a similar light-touch approach focused on public service improvements, such as transcription tools, but risks missing rising environmental costs that efficiency alone cannot solve. The natural environment underpins economy, culture and well-being and must guide AI planning toward a sustainable future.
