As social media and artificial intelligence (AI) become increasingly embedded in university students’ learning and daily lives, fatigue arising from established digital platforms and anxiety surrounding emerging AI may constitute interrelated forms of technology-related distress. However, how these experiences are jointly associated with depressive symptoms, and whether these associations vary by subjective socioeconomic status (SSES), remains insufficiently understood. Drawing on conservation of resources theory and research on status maintenance, this cross-sectional study examined AI anxiety as a statistical mediator of the association between social media fatigue and depressive symptoms and further tested whether SSES moderated the association between AI anxiety and depressive symptoms. Survey data from 1108 Chinese university students were analyzed using regression models with heteroscedasticity-consistent type 3 (HC3) robust standard errors and percentile-bootstrap confidence intervals. Social media fatigue was positively associated with AI anxiety, which in turn was positively associated with depressive symptoms, yielding a significant statistical indirect association. As SSES increased, the association between AI anxiety and depressive symptoms became stronger, and the corresponding conditional indirect association also increased. Overall, these findings suggest that understanding university students’ digital mental health requires attention to both distress arising across different technological environments and the subjective social positions from which these experiences are evaluated, offering a more context-sensitive perspective on how established and emerging forms of digital stress may be jointly associated with psychological well-being.

Using a sample of Chinese junior high school students (N = 1133), this study examined the latent heterogeneity and structural characteristics of mobile phone addiction risk. Latent profile analysis was conducted to identify subgroups with varying levels of risk. Network analysis was then used to model the relationships among smartphone addiction, materialism, and childhood neglect and compare structural differences across subgroups. Finally, five machine learning models were applied to model smartphone addiction scores and compare model performance across different combinations of variables. Results revealed three distinct risk groups of smartphone addiction. Network analysis indicated that loneliness-related nodes exhibited the highest expected influence in the overall network. Across latent profiles, childhood neglect-related nodes consistently occupied central positions, whereas materialism-related nodes showed relatively stable centrality. Network comparison tests further demonstrated significant structural differences across risk groups. In addition, incorporating latent profile information and centrality indices improved model performance, suggesting that these features capture individual differences in smartphone addiction. These findings provide structural evidence for the heterogeneity of mobile phone addiction risk and offer implications for subgroup-specific intervention strategies.