Background: Premature ovarian failure (POF) is a clinical condition characterized by the cessation of ovarian function, leading to infertility. The underlying molecular mechanisms remain unclear, and no predictable biomarkers have been identified. This study aimed to investigate the protein and metabolite contents of serum exosomes to identify underlying molecular mechanisms and explore potential biomarkers. Methods: This study was conducted on a cohort consisting of 14 POF patients and 16 healthy controls. The exosomes extracted from the serum of each group were subjected to label-free proteomic and unbiased metabolomic analysis. Differentially expressed proteins and metabolites were annotated. Pathway network clustering was conducted with further correlation analysis. The biomarkers were confirmed by ROC analysis and random forest machine learning. Results: The proteomic and metabolomic profiles of POF patients and healthy controls were compared. Two subgroups of POF patients, Pre-POF and Pro-POF, were identified based on the proteomic profile, while all patients displayed a distinguishable metabolomic profile. Signaling pathway clustering revealed progression of dysfunctional energy metabolism during the development of POF. The differentially expressed proteins and metabolites were highly correlated, with six of them selected as potential biomarkers. ROC curve analysis and random forest machine learning suggested that AFM combined with GE was a diagnostic biomarker for POF. Conclusion: Omic analysis revealed that inflammation and oxidative stress are factors that damage ovarian tissue, but progressive dysfunction of energy metabolism might be the critical pathogenetic pathway contributing to the development of POF. AFM combined with GE together serves as the best biomarker for clinical POF diagnosis.