The hidden environmental cost of artificial intelligence: AI expansion and electricity-system vulnerability in Italy

Artificial intelligence (AI) is seen more and more as a key factor for productivity, innovation, and digital change. Still, as AI applications and digital infrastructure grow, they could raise electricity use, which may affect the environment depending on how the electricity system works. This study looks at how AI adoption, electricity demand, and environmental pressure from the power sector are connected in Italy, using monthly data from January 2021 to April 2026. We use a Google Trends-based AI diffusion index along with data on electricity demand, prices, natural gas prices, carbon intensity, renewable energy, and the broader economy. Our approach includes short-term regressions, an Autoregressive Distributed Lag model, interaction models, and a simple Artificial Neural Network. In the short term, we find no clear link between changes in AI adoption and either electricity demand or emissions from the power sector. However, the ARDL model shows that over the long term, a one-standard-deviation rise in the AI diffusion index is linked to a 1.7% increase in electricity demand. The interaction models show that the link between AI adoption and power-sector emissions becomes stronger and more important when wholesale electricity prices are high. Machine learning results also show that carbon intensity and electricity demand are the main factors driving emissions. Overall, our findings suggest that the environmental impact of AI growth depends more on the state of the electricity system than on a direct effect from AI itself.