Cancer is a major cause of global mortality underpinned by genomic and epigenomic derangements. Here, we highlight the importance of multimodal data integration in understanding the molecular evolution of malignant cell states across the cancer life cycle. The widespread presence of driver mutations and epigenetic alterations in normal-appearing tissues is prompting a reassessment of how cancer initiation is defined. In later-stage cancers, studying the roles of clonal selection, epigenomic adaptation, and persister cells in metastasis and therapy resistance is an emerging field. Finally, the importance of tumor ecosystems in driving cancer development is being unraveled by single-cell and spatial technologies at unprecedented resolution. Improving cancer risk assessment and accelerating therapeutic discovery for patients will require robust, comprehensive, and integrated temporal, spatial, and multilevel tumor atlases across the cancer life cycle.