Objective: Early risk prediction for Gestational Diabetes Mellitus (GDM) may lead to early intervention, which can improve the adverse pregnancy outcomes. To find novel predictive biomarkers of GDM, we analyzed maternal plasma across the trimester during pregnancy using a metabolomics approach. Methods: Plasma samples were collected at each trimester (visit 1: 10-12 gestational weeks, visit 2: 24~26 weeks, visit 3: 35~37 weeks) from singleton pregnant women who had enrolled in longitudinal study of pregnancy from 2013 to 2016. A total 258 women matched by age and BMI (GDM n = 91; control n = 167) were included for analysis. 1H proton 700 MHz nuclear magnetic resonance (NMR) spectroscopy was performed for metabolite identification. Multiple logistic regression analysis was used to assess the association between GDM and metabolites. Results: Among 17 metabolites identified by NMR measurements, 7 metabolites (glucose, histidine, glutamine, isobutyrate, isoleucine, threonine, and acetoin) showed significantly different levels between GDM and controls (p < 0.05). After adjusting for age, BMI, prepregnancy BMI, family history of DM, parity, gravidity, and status of smoking and alcohol drinking before pregnancy, the first trimester (odds ratio [OR] 1.44; 95% CI, 1.10-1.90) and second trimester (OR 2.17; 95% CI, 1.56-3.02) isobutyrate were significantly associated with subsequent GDM diagnosis. Interestingly, women with high isobutyrate level at both first and second trimester had a higher risk of GDM compared to those who retained a low isobutyrate level (OR 5.87; 95% CI, 2.62-13.15; p < 0.0001). Compared to a model with traditional GDM risk factors, addition of first and second trimester isobutyrate levels significantly improved the discrimination (area under the ROC curve, 0.587 to 0.706; p < 0.0001) and reclassification ability (net reclassification improvement, 53.3%; p < 0.0001). Conclusion: Plasma isobutyrate can be a potential metabolic biomarker to predict subsequent developing GDM.