The environmental costs of AI
In a recent report, the UNDP (United Nations Environment Programme) states that, despite its considerable promise, AI also poses environmental problems related to resource consumption, social inequalities in access and use, and technological dependencies that can exacerbate rather than mitigate climate challenges.
AI systems require enormous amounts of electricity, with data centers consuming 10 to 50 times more energy per square meter than conventional commercial buildings.
In 2024, data centers in the Asia-Pacific region, particularly those in China, Japan, and Australia, consumed between 105 and 180 terawatt hours (TWh). In Singapore alone, they accounted for 9% of national electricity consumption.
In addition, cooling servers consumes enormous amounts of water. By 2027, data centers in the Asia-Pacific region could require up to 6.6 billion cubic meters per year, half of the UK's total annual water withdrawal. In Malaysia, less than 18% of data center water use applications have been approved, reflecting concerns about diverting water from households and ecosystems.
The hardware used for AI relies on critical minerals such as cobalt, lithium, and rare earths, the extraction of which generates CO2 and electronic waste that is harmful to the environment.
“Mining often leads to deforestation, water contamination, and habitat destruction,” the study notes.
With the kind participation of our partner, PRETAPARLER – www.pretaparler.ch
