Prostate cancer is characterized by profound heterogeneity in its clinical trajectory. While genomic heterogeneity has been well-characterized, epigenomic heterogeneity remains less understood. To fill this gap, we compiled 2,149 multi-ancestric prostate methylomes spanning normal tissue through localized disease of varying grades to poly-metastatic disease, most with multi-omic DNA and/or RNA characterization. We identify four subtypes that varied by stage, grade and mutational subtype. We identify extensive interplay between DNA ploidy and DNA methylation, with transcriptional consequences that vary across driver-genes. Each major prostate cancer driver gene mutation triggers a specific epigenetic dysregulation, and we define a set of 14 reusable models that accurately predict clinico-molecular features of a prostate cancer from DNA methylation data. Specific epigenetic features predict disease aggression, including metastasis, with epigenomic and genomic features synergizing to optimize predictions. These results define a complex interplay between tumour genetics and epigenetics that converges to modify gene-expression programs, and subsequent clinical presentation.