The rapid advancement of digital technology has made it imperative to adopt innovative instructional approaches in elementary-level IPAS (Natural and Social Science) education in Indonesia. Baseline data from Cluster 2, Pule District revealed that only 40.19% of Grade VI students met the minimum mastery criterion (KKM) in 2024/2025. This study examines the effects of AI-based IPAS learning through deep learning (X1), digital literacy (X2), and learning independence (X3) on student learning outcomes (Y) using a quantitative correlational non-experimental design. A proportional random sample of 129 students was drawn from 190 Grade VI students across 12 elementary schools. Data were analyzed through Partial Least Squares Structural Equation Modeling (PLS-SEM) using SmartPLS 4.0. The measurement model confirmed satisfactory convergent validity (all AVE > 0.50), discriminant validity (Fornell–Larcker criterion), and composite reliability (all CR > 0.90). The structural model yielded R² = 0.784, indicating that the three variables jointly explain 78.4% of the variance in learning outcomes (GoF = 0.752, very high). All four hypotheses were accepted: AI-Deep Learning (β = 0.324, t = 4.872, p < 0.05), Digital Literacy (β = 0.287, t = 4.215, p < 0.05), Learning Independence (β = 0.412, t = 6.134, p < 0.05), and their simultaneous effect (F = 124.567, p < 0.05). Learning independence emerged as the dominant predictor (f² = 0.892). These findings advocate for an integrated digital-autonomous learning ecosystem as a systemic response to persistent IPAS underachievement in Indonesian elementary schools.