When deconstructing AI’s infrastructures, a multiplicity of different components is made apparent. In this chapter, we focus on the relevance of minerals for generative AI infrastructures, considering minerals as the cradle of AI infrastructures. Using a socio-technological concept of infrastructures, which underlines their relational and (re)shaping role, we argue that besides materials and technologies, actors and their practices are intrinsic to the formation of AI infrastructures. A sustainability perspective, which we define in terms of more-than-human justice, reveals the severe impacts caused by extraction processes of minerals in different locations—mainly, although not solely, in the “Global South”. Focusing on the Amazon region in Brazil, which is one of the key locations worldwide for vast deposits of different minerals relevant to AI systems, we foreground the harmful effects that result from the extraction of minerals for AI. Our analysis strengthens the argument for a re-conceptualisation of what is commonly understood as sustainability, including green growth, towards a version of sustainability that focuses on more-than-human justice, also stressing the perspectives of indigenous actors. In conclusion, we argue that the extraction of minerals for AI infrastructures can never be sustainable in this sense, as mining will always be a means of exploitation.