I’m starting my third day at the IAMCR 2025 conference in Singapore with a panel on the political economy of AI, which starts with Benedetta Brevini. Attention to artificial intelligence has increased substantially in recent years, of course, and so has concern about the political economy of AI – with a growing focus also on the environmental impact of AI technologies and services. The massive environmental impact of artificial intelligence has now been recognised much more clearly.
This is a conversation that can no longer be avoided; it has produced substantial coverage in media, reports, and other documentation, and there is a growing realisation that the power needs of AI are unsustainable. AI companies and CEOs are strongly pushing for the rapid development of greater power generation plants: the data centres they are now building rely on power supply that simply isn’t available at this point.
This places the blame on the public and politicians for not doing enough to service this massive expansion of power needs, rather than on companies for planning data centres that rely on power resources which are not currently available. The hegemonic discourse is overlooking the materiality of data centres and AI; this connects economic power issues with social, political, and environmental issues; what is emerging here is an eco-political economy of AI.
We must borrow from other disciplines to better understand what is going on here: geography, anthropology, Indigenous studies, environmental economics, wellbeing economics are some of those fields. We need to move beyond reductionist approaches that simplify the environmental impact of AI, and instead consider the entire production chain of artificial intelligence.
This starts with a consideration of the rare minerals and resources that are required for building data centres – cobalt, for instance, mined under inhumane labour conditions for poverty-level wages. It continues with what is currently the best-understood aspect of the value chain: the excessive carbon footprint of training AI models, in comparison to the pollution caused by other human activities. It ends with the e-waste of the devices that run and use AI, as they are replaced by newer generations of technology.
We need a policy toolkit to better understand this.
