Aberrant metabolism is a hallmark of cancer, but whole metabolomic flux measurements remain scarce. To bridge this gap, we developed a novel metabolic phenotypic analysis (MPA) method that infers metabolic phenotypes based on the integration of transcriptomics or proteomics data within a human genome-scale metabolic model. MPA was applied to conduct the first genome-scale study of breast cancer metabolism based on the gene expression of a large cohort of clinical samples. The modeling correctly predicted cell lines' growth rates, tumor lipid levels, and amino acid biomarkers, outperforming extant metabolic modeling methods. Experimental validation was obtained in vitro. The analysis revealed a subtype-independent "go or grow" dichotomy in breast cancer, where proliferation rates decrease as tumors evolve metastatic capability. MPA also identified a stoichiometric tradeoff that links the observed reduction in proliferation rates to the growing need to detoxify reactive oxygen species. Finally, a fundamental stoichiometric tradeoff between serine and glutamine metabolism was found, presenting a novel hallmark of estrogen receptor (ER)(+) versus ER(-) tumor metabolism. Together, our findings greatly extend insights into core metabolic aberrations and their impact in breast cancer.