Natural Language Processing (NLP) interfaces are becoming more common in higher education, but adoption by faculty is inconsistent in private universities with limited resources. Drawing on a study of a private university in Shanghai, we investigate institutional and individual determinants of how instructors adopt NLP tools (such as ChatGPT, Doubao and Kimi) for feedback, assessment, summarization, question generation, and instructional drafting. We employed an exploratory sequential mixed-methods design involving six interviews, as well as a survey (n = 195; ≈28% of faculty). Building on a TAM model enriched by Perceived Behavioral Control (PBC), University Facilitating Conditions (UFC), and Social Attitudes (SA), we tested a partial least squares structural equation model (PLS-SEM). Findings indicate that UFC is by far the most powerful predictor of adoption intentions and perceived control, while SA exerts a weaker but statistically significant influence which partially operates through PBC. PBC not only directly influences intention but also moderates the effects of UFC, indicating a central role of self-efficacy in instructor-focused interventions. These pathways are fleshed out by qualitative evidence that points to enabling policies, training, and aligned use cases as levers, and challenges around digital literacy, policy misalignment, and uneven infrastructure. We suggest professional development, incentives linked to pedagogical outputs, and ongoing resourcing for the integration of text-focused AI. The results extend TAM to the context of a private university setting by introducing institutional support and perceived control, which may be applied to similar institutions that are experiencing digital transformation.