Artificial intelligence enters higher education: does research drive carbon reduction or accelerate emissions?

Purpose With the deep integration of artificial intelligence (AI) into the higher education sector, research activities have expanded rapidly and their associated environmental impacts have attracted growing attention. This study aims to systematically examine the interrelationships among AI, research output (RO) and carbon emissions (CE), uncovering the environmental effects of AI-driven research systems and providing empirical evidence and policy implications for promoting sustainable research development. Design/methodology/approach This study uses monthly data from September 2019 to May 2025 and applies a time-varying parameter stochastic volatility vector autoregression (TVP–VAR–SV) model to systematically analyze the time-varying effects among RO, AI and CE. Findings AI suppresses RO in the short and medium term but significantly promotes it in the long term, reflecting the transition of the research system from adaptation to expansion. Meanwhile, driven by widespread AI adoption, the impact of RO on CE varies across phases. In the short term, the positive effect is strong, as early AI-driven research remains resource-intensive and operationally demanding. In the medium term, the effect of RO on CE shifts from negative to positive, indicating that emission reductions associated with AI-driven improvements in RO are not sustained. Originality/value First, this study innovatively incorporates AI, RO and CE into a unified analytical framework, revealing the dynamic environmental effects of AI-driven research expansion. Second, the TVP–VAR–SV model is used to identify nonlinear and time-varying characteristics, addressing the theoretical and methodological gaps in existing studies. Finally, the findings provide new empirical evidence and theoretical insights for understanding the sustainability of research activities within the higher education system and for informing policy formulation.