Empirical Lifecycle Assessment of Generative AI Inference Carbon Emissions: Global Trends and the Malaysian 'Carbon Arbitrage' Risk (2023–2030)

Abstract

The integration of Generative Artificial Intelligence (GenAI) into global digital infrastructure has initiated a profound structural shift, transitioning the primary environmental impact of machine learning from discrete model training events to continuous, distributed inference workloads. This research addresses the critical problem of unquantified and geographically externalised carbon emissions resulting from the exponential growth of GenAI queries, specifically focusing on the vulnerability of emerging markets in the Global South to digital carbon arbitrage. Utilising an Attribution-Based Life Cycle Assessment (A-LCA) framework compliant with ISO 14044 standards, this study models global GenAI inference emissions from 2023 to 2025 and employs logistic regression forecasting alongside Monte Carlo simulations to project demand through 2030. The primary results demonstrate that while specialised hardware and algorithmic optimisations successfully reduced the energy intensity of median text queries to 0.34 Wh per interaction, the accompanying surge to one billion daily interactions completely negated these efficiency dividends. This dynamic resulted in a calculated rebound elasticity of 1.08, providing strict statistical validation of the Jevons paradox within the artificial intelligence sector. Furthermore, a comparative location penalty analysis reveals that processing identical computational workloads in Malaysia’s fossil-heavy grid, compounded by tropical thermodynamic cooling constraints, yields a carbon footprint approximately 42 times higher than identical deployments in temperate, low-carbon regions. The study concludes that the unmanaged influx of hyperscale inference data centres poses a systemic risk to Malaysia’s Nationally Determined Contribution (NDC) targets, necessitating an urgent regulatory paradigm shift from traditional Power Usage Effectiveness (PUE) metrics to comprehensive Carbon Usage Effectiveness (CUE) standards.