Researchers have used large language models (LLMs) to map the literature on water and energy use in AI infrastructure, from mining to electronic waste. The conclusion: we mainly measure the electricity used by data centres, but the rest is poorly measured.
The rise of AI is based on a chain of resource-intensive activities spread across the globe: mineral extraction, semiconductor manufacturing, data center operation, and e-waste processing. One team proposes to analyze them together through a framework called the water-energy-data (WED) nexus, which tracks both the lifecycle of the infrastructure and pressure transfers from one region to another.
To draw up the state of knowledge, the authors synthesized 12,825 studies on water or energy in these four activities using large language models. The resulting map shows a clear discrepancy: the work focuses on electricity consumption during the operation phase, much more than on water or indirect needs, and rarely studies interactions between sites and between stages of the life cycle.
Who pays the bill?
These gaps make it difficult to know where resource needs are moving, which populations are bearing the consequences, and whether the benefits of the data actually outweigh their costs. The WED framework and the evidence map are presented as a basis for prioritizing research and management priorities in the face of the AI boom.
Read the full article on the original source
Go to the source →