Mapping AI Users and Potential Inequality Pattern: A Spatial Downscaling Study in London

With the rapid development of artificial intelligence (AI), energy consumption and carbon emissions from high-demand computing power have gradually attracted widespread attention in the environmental field. Existing research largely focuses on data centers, which are the infrastructure that directly generates AI-related carbon emissions. However, the users who truly drive computing power demand have long been neglected. A major reason is the difficulty in tracking and accurately locating users, so that AI-related carbon emissions from users’ perspective have lacked systematic identification and discussion so far. It is worth noting that with the wide application of AI, the primary source of carbon emissions has shifted from model training to large-scale and multi-domain usage. This means that understanding the spatial distribution pattern of AI users is crucial to explore demand-side emission reduction in AI, especially during periods of bottlenecks in production-side emission reduction, such as the slow green transformation of the electricity energy mix. Demand-side management can, to some extent, contribute to mitigating AI-related carbon emissions. In this study, we aim to display the spatial distribution of AI users within the city and assess whether variations in usage across different wards may lead to potential spatial inequalities in AI-related carbon emissions. Taking London as a case study, we utilize regional AI penetration rates and AI user profiles to spatially decompose urban AI users at a more granular scale, quantifying the corresponding AI-related carbon emissions and comparing the proportion of AI-related carbon emissions in residents' carbon footprints and potential inequalities. We expect to find spatial clustering of AI-related carbon emissions and a positive correlation with the distribution of educational resources and wealth. Our study may provide an empirical basis for understanding the new environmental inequalities brought about by AI development and offers key references for future green digital governance on the demand side.