Sepsis is one of the most important problems to be addressed in pediatrics, characterized by insidious onset, rapid progression, and high rates of severe infection and even mortality. Biomarkers with high sensitivity and robustness are urgently required for the early diagnosis of infant sepsis. Serum metabolomic approaches based on liquid chromatography-mass spectrometry were used to analyze the samples from 30 infants with sepsis at an early stage and 30 infants with noninfectious diseases. Multivariate statistical analysis was used to screen for differential metabolites and ROC curves were generated to find potential biomarkers. Six metabolites, including phosphatidic acid (PA (8:0/14:0)), phosphatidyl ethanolamine (PE (16:0/18:2(9Z,12Z))), cytidine 5'-diphosphocholine (CDP-CHO), sphingomyelin (SM (d18:0/16:1(9Z))), prolylhydroxyproline and phosphorylcholine (P-CHO), were identified between the two groups. ROC curve analysis showed that prolylhydroxyproline (AUC = 0.832) had potential diagnostic values for infant sepsis. The AUC value was 0.859 (CI: 0.764, 0.954) in the combined model. Prolylhydroxyproline were found to be correlated with CRP and PCT levels, while PE and CDP-CHO associated with PCT levels. Pathway analysis indicated that glycerophospholipid metabolism, aminoacyl-tRNA biosynthesis and necroptosis pathways played important roles in infant sepsis. Network analysis showed that the differential metabolites were linked to ERK/ MAPK, NF-κB, AMPK, mTOR, and other classical inflammatory and metabolic signaling pathways. This study identified serum metabolite profiles and three metabolites as potential biomarkers in infants with sepsis. The findings will help improve the early diagnosis of sepsis in infants.