<p><i>Staphylococcus aureus</i> is increasingly resistant to β-lactam antibiotics, making non-β-lactam cell-wall-targeting drugs crucial alternatives. Growing resistance to these agents highlights the need to identify genomic factors influencing susceptibility. Machine learning can integrate genomic and phenotypic data to predict minimum inhibitory concentrations (MICs) and uncover resistance mechanisms across time and regions. We obtained 112,360 <i>S. aureus</i> genomes from NCBI GenBank (March 2024), applying quality filters and standardizing metadata. Resistance genes and mutations were identified using AMRFinderPlus and CARD, focusing on glycopeptide, lipopeptide, bacitracin, and fosfomycin resistance. MICs for five antibiotics were compiled, standardized, and log₂-transformed for analysis. Allelic profiles for seven housekeeping genes were assigned using PubMLST’s BIGSdb and MLST CLI v2.19.0. Temporal and geographic resistance trends were modeled using logistic regression and statistical tests. Machine learning models (Random Forest, XGBoost, Elastic Net, Partial Least Squares (PLS)) predicted MICs from genomic features, with performance assessed via cross-validation. Statistical analyses and visualizations were performed in R, with all data and scripts provided for reproducibility. We analyzed 111,350 <i>S. aureus</i> genomes from 137 countries, with 78% from clinical sources, 10% from environmental, veterinary, or food-related origins, and some from animals. Glycopeptide MICs were low across all sources: vancomycin (0.96&#xa0;µg/mL) and teicoplanin (0.52&#xa0;µg/mL), while daptomycin showed more variability (0.44&#xa0;µg/mL). Fosfomycin resistance genes, particularly fosB, were detected in 65.3% of genomes overall, with significantly higher prevalence in clinical isolates (32.5%) compared to environmental (2.1%), food (4.0%), and animal sources (7.5%). Bacitracin resistance genes (bcrAB) were detected in 6.2% of clinical isolates versus 1.3% environmental and 2.8% animal sources. However, phenotypic MIC data were severely limited (fosfomycin <i>n</i> = 1, bacitracin <i>n</i> = 1), precluding validation of genotype-phenotype correlations and limiting epidemiological interpretation to genetic prevalence alone. Resistance to glycopeptides and lipopeptides remained rare (&lt; 0.1%). Fosfomycin resistance protein B (f<i>osB</i>) resistance increased by 0.20% annually, especially in clinical and animal sources, while other mutations like <i>glpT_V213I</i> and <i>murA_D278E</i> declined. Geographic trends showed fosB resistance exceeded 50% in North America, Europe, and South America, with <i>MurA_G257D</i> most prevalent in the Middle East. Machine learning models showed moderate predictive performance for daptomycin MICs (R² = 0.49), with <i>mprF</i> mutations as key predictors, but demonstrated poor accuracy for glycopeptides (vancomycin R² = 0.05; teicoplanin R² = -0.13) due to extremely limited MIC variability in the dataset. Fosfomycin and bacitracin models could not be trained due to insufficient phenotypic data (<i>n</i> = 1 each). This study highlights the growing challenge of <i>S. aureus</i> resistance, especially to non-β-lactam antibiotics such as fosfomycin and bacitracin, with clinical isolates showing the highest resistance. Geographic and temporal trends indicate an increasing prevalence of <i>fosB</i> resistance, particularly in clinical and animal sources. Machine learning models failed to predict glycopeptide MICs (vancomycin R² = 0.05; teicoplanin R² = -0.13) due to limited phenotypic variability, achieved moderate success for daptomycin (R² = 0.49), and could not be trained for fosfomycin and bacitracin (<i>n</i> = 1 MIC measurement each) despite high genetic prevalence of resistance determinants. These findings emphasize the critical need for integrated genomic and phenotypic surveillance, and highlight fundamental data requirements for ML-based resistance prediction.</p>