A fine-tuned Spanish-language AI model called MHC-Coach-ES demonstrated superior performance compared to human expert messages in promoting physical activity among Hispanic/Latinx populations. The AI was trained on 2.21 million tokens of Spanish health content and 3,268 expert messages using the Transtheoretical Model of behavior change. In testing with 77 Spanish-speaking participants, the AI coach achieved 81% preference over translated human expert messages, while clinical experts rated it higher for behavioral science alignment (4.83 vs 4.38 out of 5). This finding challenges conventional assumptions about human expertise in health coaching and suggests AI may excel at operationalizing behavioral science frameworks across languages. The implications are significant for addressing health disparities among the 16 million Hispanic/Latinx adults with limited English proficiency who face higher chronic disease rates linked to physical inactivity. However, this preprint awaits peer review, and the study's relatively small sample size and focus on message preference rather than actual behavior change outcomes limit immediate conclusions about real-world effectiveness. The research represents an incremental but meaningful advance in culturally adaptive digital health interventions.