Breast cancer is a clinically heterogeneous disease, where the diverse tumour and microenvironmental landscape determines cancer progression and prognosis. Although single-cell sequencing has revealed cellular diversity, these approaches lack spatial resolution, limiting insights into how stromal subsets interact and support malignant cells. To address this gap, we applied the Human 6K Discovery Panel on the CosMx Spatial Molecular Imager to our cohort of 28 breast cancer patients diagnosed with major molecular subtypes, 9 matched adjacent normal tissues and lymph node metastases, along with 4 healthy women. We spatially profiled 2.28 million cells and identified 15 broad cell types across the tumour and stromal compartments. Cell type analysis revealed high abundance of cancer-associated fibroblasts (CAFs) in both the primary and metastatic lesions. In particular, myofibroblastic CAF (myCAF)-like s4-CAFs are enriched for Notch1, which is correlated with poor survival amongst breast cancer patients. We also noted that CAF phenotypes and spatial distributions vary across tissue locations and molecular subtypes, shaping unique niche architecture that enables distinct patterns of CAF interactions with the tumour microenvironment. Similarly, we observed heterogeneity within the immune compartments across disease states. Investigation into the spatial arrangement of the diverse cell types also revealed unique cellular communications between epithelial cells, CAFs and endothelial cells at various tissue locations. By analysing the diverse molecular signatures and cellular architecture, our in-depth spatial atlas will help gain deeper insights into the intercellular communication that characterises major breast cancer subtypes and contributes to disease progression.