Equipment used to train and run generative AI models could produce up to 5 million tons of e-waste by 2030, a relatively small but significant fraction of the global total.