The real-world cost of AI (www.rnz.co.nz)

A recent report from a UN think-tank has highlighted the environmental impact of artificial intelligence (AI), revealing that global AI use consumes approximately 93 terawatt-hours (TWh) of electricity annually—more than double New Zealand's total electricity generation. AI currently accounts for 20% of data center electricity consumption, projected to double by 2030. This shift toward AI's energy use primarily stems from daily interactions, known as the 'inference' phase, which consumes significantly more energy than the initial model training. For example, while a single interaction with AI like ChatGPT uses around 0.42 watt-hours, the collective energy use from billions of queries contributes to an estimated 80-90% of AI's total energy consumption. The findings raise critical concerns about the sustainability of expanding AI infrastructure, particularly as New Zealand contemplates building more data centers, including Datagrid's planned 'AI factory' which is set to be the country's second-largest electricity user. Experts warn that the environmental costs extend beyond electricity usage, highlighting demands for land and water resources as well. As AI becomes increasingly integrated into everyday services, addressing these implications requires a shift toward efficiency and sustainability in AI development, emphasizing the need for responsible resource management and careful consideration of the infrastructure that supports AI in various environments.