A NEW LENS ON THE SUSTAINABILITY OF THE AI REVOLUTION

Analyzing cross-country EPE data over the last three decades, the study identifies distinct trends across different economic clusters. Advanced economies show a consistent, almost linear growth in EPE over the past 40 years, signifying improved energy efficiency and value generation. This group accounts for a significant portion of global GDP and energy use, making their EPE trajectory crucial for overall world EPE, which has also grown monotonically. Developing economies exhibit persistent EPE growth primarily after 2005. Underdeveloped economies, surprisingly, have the highest EPE values in absolute terms, though their mean EPE shows a nearly constant trend. When accounting for human metabolic energy consumption (estimated at 2.9 kWh/day), the EPE of underdeveloped and developing economies decreases, suggesting that human labor is less energy-intensive than industrial processes in generating GDP. These diverse trends underscore the heterogeneity of EPE dynamics globally and the need for nuanced interpretation.

The article posits that the Artificial Intelligence revolution shares characteristics with the first industrial revolution, classifying AI as a 'pre-scientific engine.' This means AI's large-scale deployment and monetization are preceding a consolidated theoretical understanding of its systemic efficiencies, limits, and energy externalities. The analogy is drawn with early steam power: high profitability drives adoption without full regard for energy costs, potentially leading to a temporary decline in EPE as output rises faster than energy efficiency improves. The text outlines two scenarios: either EPE declines initially then recovers as AI technology matures and integrates efficiently, or sustained pressure for sustainability delays adoption until energetic costs are manageable, maintaining or strengthening current EPE. AI's impact will be heterogeneous across sectors and geographies, with advanced economies expected to drive near-term AI-driven GDP growth and influence global EPE trends. The technology presents both risks (grid stress, water constraints, carbon-intensive deployments) and opportunities (grid optimization, smart controls) for EPE.

The authors advocate for EPE as a critical sustainability indicator, arguing it is more relevant than energy intensity in a resource-constrained world, as it frames efficiency as productivity. They propose a comprehensive EPE monitoring agenda: establish EPE as an official indicator, publish quarterly dashboards with sectoral and AI-related energy use details, and condition AI infrastructure incentives on demonstrated EPE neutrality or improvement. Key policy recommendations include mandating standardized disclosure of AI training and inference energy (including cooling and water), funding cross-disciplinary research on the thermodynamics of computation and energy productivity, and promoting research connecting physics of computation with energy economics. This framework aims to align technological innovation with energy productivity, detect early signals of efficiency losses, and guide regulation and investment towards sustainable AI-driven growth.