The association of antiretroviral therapy (ART) with diabetes is inconsistent and varies widely across primary epidemiological studies. A comprehensive and more precise estimate of this association is fundamental to establishing a plausible causal link between ART and diabetes. We identified epidemiological studies that compared mean fasting plasma glucose (FPG) concentrations and proportions of diabetes and metabolic syndrome between HIV-infected patients naïve and exposed to ART. Mean difference in FPG concentrations and odds ratios of diabetes and metabolic syndrome were pooled using random-effects meta-analyses. Data on 20 178 participants from 41 observational studies were included in the meta-analyses. Mean FPG concentrations (Pooled mean difference: 4.66 mg/dL; 95% confidence interval [CI], 2.52 to 6.80; 24 studies) and the odds of diabetes (Pooled odds ratios: 3.85; 95% CI, 2.93 to 5.07; 10 studies) and metabolic syndrome (Pooled odds ratios: 1.45; 95% CI, 1.03 to 2.03; 18 studies) were significantly higher among ART-exposed patients, compared to their naïve counterparts. ART was also associated with significant increases in FPG levels in studies with mean ART duration ≥18 months (Pooled mean difference: 4.97 mg/dL; 95% CI, 3.10 to 6.84; 14 studies), but not in studies with mean ART duration <18 months (Pooled mean difference: 4.40 mg/dL, 95% CI, -0.59 to 9.38; 7 studies). ART may potentially be the single most consistent determinant of diabetes in people living with HIV worldwide. However, given the preponderance of cross-sectional studies in the meta-analysis, the association between ART and diabetes cannot be interpreted as cause and effect.