This study explores the relationship between academic writing self-efficacy, writing emotions, and dependency on artificial intelligence (AI) tools among English as a Foreign Language (EFL) learners in China. While self-efficacy is widely recognized as a key factor in reducing writing anxiety and enhancing motivation, its connection to writing enjoyment remains less examined. At the same time, the growing presence of AI applications, such as ChatGPT, raises questions about whether reliance on these tools enhances or undermines students' learning experiences. A quantitative survey was conducted with 245 students from four English-medium instruction universities in China. Participants completed self-reported measures of writing self-efficacy, writing anxiety, enjoyment, and AI dependency. Correlational analyses revealed that self-efficacy was strongly negatively correlated with anxiety (r = -.50, p < .01) and positively correlated with enjoyment (r = .67, p < .01), suggesting that students with higher confidence in their writing abilities experience less anxiety and greater enjoyment. Regression analysis further confirmed that self-efficacy significantly predicted both reduced anxiety (β = -.50, p < .01) and increased enjoyment (β = .67, p < .001). Importantly, moderation analysis showed that AI dependency did not significantly affect the relationship between self-efficacy and anxiety, but it did weaken the positive link between self-efficacy and enjoyment (B = -.85, p < .05). These findings highlight the dual role of AI in academic writing. While self-efficacy remains a critical predictor of positive emotional outcomes, heavy reliance on AI may dampen the benefits of confidence by reducing students' intrinsic enjoyment of writing. The results suggest that educators should support both self-efficacy development and responsible AI use to foster sustainable learning outcomes in EFL academic writing.