We aimed to identify an effective metabolic subtype and risk score to predict survival and immunotherapy response in head and neck squamous cell carcinoma (HNSCC). Data were obtained from an online database. We screened significant prognostic metabolism-related genes between the normal and tumor groups using a series of bioinformatics methods. Based on the selected prognostic genes, we conducted a subtype analysis to identify significantly different subtypes in HNSCC. We then investigated survival, immune features, and hallmark differences among different subtypes. LASSO was utilized to identify optimal genes for the risk score model construction. Finally, distribution of the risk score samples was analyzed for different subtypes. A total of 32 significantly prognostic metabolism-related genes were screened, and all samples were grouped into two subtypes: cluster 1 and cluster 2. Cluster 1 had worse survival. Different immune cell infiltration (CD8 T cells, macrophages, and regulatory T cells) and immune checkpoint gene expression (PD-1 and CLAT-4) were observed between the two clusters. Twelve optimal genes were involved in risk score model, and high-risk group had poorer survival. Cluster 1 contained more high-risk samples (60%). Finally, four genes CAV1, GGT6, PYGL, and HS3ST1 were identified as significantly related to immune cells, and these genes were differentially expressed in the normal oral epithelial cells and HNSCC cells. The subtypes and risk score model in the study provide a promising biomarker for prognosis and immunotherapy response.