Activated B-cell (ABC)-diffuse large B-cell lymphomas (DLBCLs) are clinically aggressive and phenotypically complex malignancies, whose transformation mechanisms remain unclear. Partially differentiated antigen-secreting cells (plasmablasts) have long been regarded as cells-of-origin for these tumors, despite lack of definitive experimental evidence. Recent DLBCL reclassification based on mutational landscapes identified MCD/C5 tumors as specific ABC-DLBCLs with unfavorable clinical outcome, activating mutations in the signaling adaptors MYD88 and CD79B, and immune evasion through mutation of antigen-presenting genes. MCD/C5s manifest prominent extranodal dissemination and similarities with primary extranodal lymphomas (PENLs). In this regard, recent studies on TBL1XR1, a gene recurrently mutated in MCD/C5s and PENLs, suggest that aberrant memory B cells (MBs), and not plasmablasts, are the true cells-of-origin for these tumors. Moreover, transcriptional and phenotypic profiling suggests that MCD/C5s, as a class, represent bona fide MB tumors. Based on emerging findings we propose herein a generalized stepwise model for MCD/C5 and PENLs pathogenesis, whereby acquisition of founder mutations in activated B cells favors the development of aberrant MBs prone to avoid plasmacytic differentiation on recall and undergo systemic dissemination. Cyclic reactivation of these MBs through persistent antigen exposure favors their clonal expansion and accumulation of mutations, which further facilitate their activation. As a result, MB-like clonal precursors become trapped in an oscillatory state of semipermanent activation and phenotypic sway that facilitates ulterior transformation and accounts for the extranodal clinical presentation and biology of these tumors. In addition, we discuss diagnostic and therapeutic implications of a MB cell-of-origin for these lymphomas.