Over the last decade, healthcare has undergone a transformative digitalization process, which in turn has generated a new wealth of biomedical data. This availability of massive amounts of clinical data, combined with the latest advancements in Deep Learning, might open new promising scenarios in complex and multifaceted diseases, such as cancer. Specifically, the field of oncology has already benefited from Artificial Intelligence-based diagnostic and prognostic systems, yet mostly built on single, unimodal, data types. However, the development of new integrated models combining multiple clinical data modalities (e.g. clinical records, -omics data, histology, radiology, endoscopy) might capture synergistic patterns revealing new multimodal/multiscale biomarkers with potentially large diagnostic or prognostic value. While still in its dawn, by integrating information from multiple perspectives and scales, multimodal fusion of clinical data might offer new opportunities in representing the complex pathophysiology of cancer, ultimately advancing cancer care.