Background <p>Although several occupational metals are established or probable carcinogens, their attributable cancer burdens have rarely been compared within a unified global framework. Existing evidence is often restricted to single metals, selected cancer sites, or individual countries, leaving uncertainty about how arsenic, cadmium, chromium, nickel, and beryllium differ in long-term trajectories and Socio-demographic Index (SDI)-linked inequalities.</p> Objective <p>We quantified the cancer burden attributable to five occupational metal exposures across global, regional, and national settings from 1990 to 2021 and examined whether these burdens converged or diverged across SDI strata. Interpretable extreme gradient boosting (XGBoost) models were then used to project metal-specific trajectories to 2030.</p> Methods <p>Data were obtained from the Global Burden of Disease Study 2021 and covered 206 countries and territories. The main outcomes were disability-adjusted life years (DALYs) and age-standardized DALY rates per 100,000 population. Temporal trends were assessed using estimated annual percentage changes. SDI quintiles and concentration indices were used to evaluate development-related inequalities. Metal-specific XGBoost models were developed to project age-standardized DALY rates from 2022 to 2030, and Shapley additive explanations (SHAP) were used to interpret predictor contributions.</p> Results <p>In 2021, arsenic and nickel accounted for the largest attributable cancer burdens. From 1990 to 2021, global age-standardized DALY rates declined for most metals, but improvements were heterogeneous. Countries with higher baseline burdens generally showed larger reductions, whereas several low-middle- and middle-SDI settings had stable or increasing trajectories. Arsenic and nickel showed generally positive concentration indices, indicating greater concentration in higher-SDI countries, while cadmium, chromium, and beryllium shifted from negative or near-zero values toward positive values over time. In temporal validation, XGBoost models showed moderate performance in predicting annual log-rate changes, with R² values of 0.238–0.466 and root mean square errors of 0.006–0.024. Recursive projections of age-standardized DALY rates were consistent with observed values. Projections to 2030 suggested persistent divergence across SDI groups, with the largest relative increases mainly in low-middle- and middle-SDI countries. SHAP analysis indicated that lagged burden was the dominant predictor.</p> Conclusions <p>The cancer burden related to occupational metal exposure has declined in some settings, but inequalities between countries and across development levels remain. Countries undergoing socioeconomic and industrial transition may face increasing relative burdens in the future and should be prioritized for occupational exposure monitoring, engineering controls, and long-term cancer prevention.</p>