Many tech firms and global companies believe that artificial intelligence (AI) could reduce greenhouse-gas emissions. A recent report by the Boston Consulting Group finds that the technology could mitigate as much as 10% of global emissions. But such optimistic projections are tempered by other concerns including its high energy consumption could undermine its environmental benefits. Training a single large AI model consumes as much energy as 126 homes in a year. Additionally, AI’s water consumption is becoming a concern as 10 to 50 searches can use as much water as a 500ml bottle, with global search volumes pointing to worrying impacts on water resources. While AI offers transformative potential, how can its benefits be harnessed without exacerbating environmental damage?