Immunotherapies provide long-lasting responses across a wide range of cancers; however, only a fraction of patients responds to them. Multiple reasons contribute to the failure, including the existence of myeloid-derived cells (MDCs) within tumors. Due to their high plasticity, these cells display numerous cell states and, as a result, distinct pro-tumorigenic behaviors, making them interesting targets. However, the creation of cutting-edge anticancer therapies is hampered by the lack of MDC-specific markers. To fully define the MDC landscape in solid tumors, we combined single-cell RNA-Seq from 13 public datasets, including samples from seven different cancers and normal samples, yielding the largest collection of MDC subpopulations within the tumor microenvironment. We identified five major lineages subdivided into one mast cell cluster, three neutrophils, eight dendritic cells, six monocytes, and eleven macrophage states. Transcriptional profiles coupled with deconvolution estimates of cell populations in large cohorts revealed five MDC subpopulations as independent prognostic markers in different cancer types, including resident tissue interstitial macrophages and FCGR3A+ monocytes associated with an unfavorable clinical outcome in ovarian and breast cancer patients, respectively. Our work reveals that TREM2+ macrophages can be distinguished in different populations and associated with distinct prognoses. By analyzing a Brazilian cohort of ovarian cancer, we found that TREM2+ macrophages are associated with a better prognosis, indicating that their role might be dependent on the tumor niche and co-expression of immunosuppressive markers. Collectively, this atlas reveals in high-resolution the heterogeneous MDC identity as well as new avenues for understanding and manipulating their fate in cancer.