BACKGROUND: Characterizing the variety of cell types in the tumor microenvironment (TME) and their organization into cellular communities is vital for elucidating the biological diversity of cancer and informing therapeutic strategies. METHODS: Here, we employed a machine learning-based algorithm framework, EcoTyper, to analyze single-cell transcriptomes from 139 patients with head and neck squamous cell carcinoma (HNSC)and gene expression profiles from 983 additional HNSC patients, aiming to delineate the fundamental cell states and ecosystems integral to HNSC. RESULTS: A diverse landscape of 66 cell states and 9 ecosystems within the HNSC microenvironment was identified, revealing classical cell types while also expanding upon previous immune classifications. Survival analysis revealed that specific cell states and ecotypes (ecosystems) are associated with patient prognosis, underscoring their potential as indicators of clinical outcomes. Moreover, distinct cell states and ecotypes exhibited varying responses to immunotherapy and chemotherapy, with several showing promise as predictive biomarkers for treatment efficacy. CONCLUSION: Our large-scale integrative transcriptome analysis provides high-resolution insights into the cellular states and ecosystems of HNSC, facilitating the discovery of novel biomarkers and supporting the development of precision therapies.