The deep integration of digital transformation and artificial intelligence (AI) is driving profound changes in mathematics education. International forums like the 15th International Congress on Mathematical Education (ICME-15) have emphasized “reconstructing teaching paradigms through AI” as a central theme. However, current research often remains limited to a tool-oriented approach involving specific technologies, creating a disconnect between learning and teaching processes. This gap hinders a fundamental solution to the core challenge of balancing standardized education with personalized cultivation. To address this, this study proposes the “AI Dual-Loop Empowerment” model. This data-driven framework establishes a dynamic closed-loop system. Within the “student self-learning loop,” activities such as “preview” and “instant diagnosis” generate “learning data.” These data, in turn, drive the “teacher teaching loop,” where educators perform “learning analytics” and “implement interventions.” The outcomes of these interventions feed back into the students’ subsequent learning, creating a virtuous cycle in which “learning informs teaching and teaching promotes learning” and enabling continuous “data-driven decision.”