<p>Amid intense global academic competition, Generative AI (GenAI) has become a vital tool for STEM postgraduates, yet its widespread adoption has also given rise to concerns about GenAI dependency in research. Drawing on the I-PACE model, this mixed-methods study investigated how a competitive climate fosters GenAI dependency in research. A survey (<i>n</i> = 865) and semi-structured interviews (<i>n</i> = 17) revealed a key pathway: competitive climate enhanced GenAI proficiency in research, which then increased GenAI perceived usefulness in research, culminating in GenAI dependency in research. Our qualitative analysis illuminated the complex mechanisms behind this phenomenon, explaining the ‘nuclear balance’ logic that drives its formation and its two roles as a ‘cognitive crutch’ and an ‘emotional safety net,’ which led to consequences of tension, cognitive regression, and academic isolation. This study highlighted the interplay between macro-level environment and micro-level individual factors in shaping GenAI dependency in research. It underscored the responsibility of higher education institutions and supervisors to guide postgraduates toward critical, rather than dependent, engagement with GenAI, particularly within a high-pressure academic climate.</p>