The relationship of systolic blood pressure variability (SBPV) with Alzheimer's disease (AD) remains controversial. We aimed to explore the roles of SBPV in predicting AD incidence and to test the pathways that mediated the relationship of SBPV with cognitive functions. Longitudinal data across 96 months (T0 to T4) were derived from the Alzheimer's disease Neuroimaging Initiative cohort. SBPV for each participant was calculated based on the four measurements of SBP across 24 months (T0 to T3). At T3, logistic regression models were used to test the SBPV difference between 86 new-onset AD and 743 controls. Linear regression models were used to test the associations of SBPV with cognition and AD imaging endophenotypes for 743 non-demented participants (median age = 77.0, female = 42%). Causal mediation analyses were conducted to explore the effects of imaging endophenotypes in mediating the relationships of SBPV with cognitive function. Finally, Cox proportional hazard model was utilized to explore the association of SBPV with incident risk of AD (T3 to T4, mean follow-up = 3.5 years). Participants with new-onset AD at T3 had significantly higher SBPV compared to their controls (p = 0.018). Higher SBPV was associated with lower scores of cognitive function (p = 0.005 for general cognition, p = 0.029 for memory, and p = 0.016 for executive function), higher cerebral burden of amyloid deposition by AV45 PET (p = 0.044), lower brain metabolism by FDG PET (p = 0.052), and higher burden of white matter hyperintensities (WMH) (p = 0.012). Amyloid pathology, brain metabolism, and WMH partially (ranging from 17.44% to 36.10%) mediated the associations of SBPV with cognition. Higher SBPV was significantly associated with elevated risk of developing AD (hazard ratio = 1.29, 95% confidence interval = 1.07 to 1.57, p = 0.008). These findings supported that maintaining stable SBP in late life helped lower the risk of AD, partially by modulating amyloid pathology, cerebral metabolism, and cerebrovascular health.