While advances in treatment of breast cancer (BC) have improved outcomes for many, not all BCs are biologically or clinically alike (1). Rare subtypes remain clinically challenging, collectively comprise around 15% of cases and disproportionately contribute to global BC mortality. They are often highly metastatic, chemo-resistant, and have limited representation in large-scale studies (2). Leveraging high-resolution profiling of archival FFPE tissue, a critical step for rare cancers where fresh samples are scarce, we have generated a comprehensive single-cell transcriptome and spatial dataset spanning multiple rare six histological subtypes and two classical subtypes. In total, we have profiled ~1 million high-quality single cells from 61 FFPE tumours using snRNA-seq, followed by large-scale integrative analysis to resolve the cellular landscape of each subtype. Striking diversity amongst tumor cell states across histotypes was found, as well as profound differences in the tumor microenvironment. Complementing this, spatial profiling of ~8 million cells from 47 matched samples with Xenium uncovered universal niches and cell–cell communication patterns with remarkable histotype specificity. For example, in Micropapillary tumours, tumour cell crosstalk occurs predominantly with other tumor cells, fibroblasts, and endothelial cells, with minimal immune interactions, features consistent with their distinctive morphology. Together, these findings highlight subtype-specific ecological rules, encompassing both tumour-intrinsic signalling programs and distinct tumour–immune or tumour–fibroblast interactions. This work has the potential to open new avenues for identifying novel biomarkers and therapeutic targets through key cell–cell interactions and immune regulatory mechanisms that drive disease progression and treatment resistance.