AI's climate footprint extends beyond data center energy consumption, overlooking rebound effects in environmental debate

Environmental analysis of artificial intelligence typically focuses on data center energy consumption and associated emissions. Yet a critical mechanism frequently escapes attention: rebound effects, where technological efficiency improvements trigger paradoxically greater overall consumption. This dynamic has historical precedent. Nineteenth-century economist William Jevons demonstrated that improved coal efficiency did not decrease total coal consumption—it increased it. The logic is straightforward: when technology becomes more efficient and consequently cheaper, demand for its use expands substantially, often negating initial efficiency gains. Applied to AI systems, this historical pattern suggests that advancements in model efficiency may not reduce aggregate consumption. Rather, more efficient AI systems could accelerate large-scale deployment and training initiatives, potentially offsetting or reversing the environmental benefits of efficiency improvements.