Carcinoma-associated fibroblasts (CAFs) exhibit significant heterogeneity and are closely associated with progression, resistance to anticancer therapies, and poor prognosis in head and neck squamous cell carcinoma (HNSCC). However, the specific functional role of CAFs in HNSCC has been inadequately explored. In this study, we utilized a single-cell RNA sequencing dataset from HNSCC (GSE103322) to recluster CAFs via the Seurat pipeline. On the basis of the reported markers, we identified two CAF subtypes, LOX-myCAFs and LOX + iCAFs, and generated signature markers for each. Through unsupervised consensus clustering, we identified and characterized two molecular subtypes of HNSCC-TCGA, each exhibiting distinct dysregulated cancer hallmarks, immunological tumor microenvironments, and stemness characteristics. The robustness of the LOX + iCAF-related signature clustering, particularly in terms of prognosis and prediction of immunotherapeutic response, was validated in an ANOVA cohort via a GEO dataset (GSE159067) consisting of 102 HNSCC patients. A positive correlation was validated between the expression of LOX and that of CD86, a marker of M1 macrophage polarization. Further experiments involving the coculture of conditioned medium derived from LOX-silenced CAFs with CAL-27 and UM-SCC-1 cell lines revealed that LOX silencing led to decreased proliferation and migration of these cancer cells, which was mediated by epithelial-mesenchymal transition (EMT) through IL-34- induced CSF1R/Akt signaling. In summary, our single-cell and bulk RNA sequencing analyses revealed a LOX + iCAF-related signature that can predict the prognosis and response to immunotherapy in HNSCC patients. Additionally, the LOX gene was identified as a promising therapeutic target for HNSCC treatment.