Understanding the details of metabolic reprogramming in hepatocellular carcinoma (HCC) is critical to improve stratification for therapy. Both multiomics analysis and cross-cohort validation were performed to investigate the metabolic dysregulation of 562 HCC patients from 4 cohorts. On the basis of the identified dynamic network biomarkers, 227 substantial metabolic genes were identified and a total of 343 HCC patients were classified into 4 heterogeneous metabolic clusters with distinct metabolic characteristics: cluster 1, the pyruvate subtype, associated with upregulated pyruvate metabolism; cluster 2, the amino acid subtype, with dysregulated amino acid metabolism as the reference; cluster 3, the mixed subtype, in which lipid metabolism, amino acid metabolism, and glycan metabolism are dysregulated; and cluster 4, the glycolytic subtype, associated with the dysregulated carbohydrate metabolism. These 4 clusters showed distinct prognoses, clinical characteristics and immune cell infiltrations, which was further validated by genomic alterations, transcriptomics, metabolomics, and immune cell profiles in the other 3 independent cohorts. Besides, the sensitivity of different clusters to metabolic inhibitors varied depending on their metabolic features. Importantly, cluster 2 is rich in immune cells in tumor tissues, especially programmed cell death protein 1 (PD-1)-expressing cells, which may be due to the tryptophan metabolism disorders, and potentially benefiting more from PD-1 treatment. In conclusion, our results suggest the metabolic heterogeneity of HCC and make it possible to treat HCC patients precisely and effectively on specific metabolic characteristics.