The study is devoted to a quantitative and qualitative assessment of the impact of integrating machine learning technologies on the operational efficiency and environmental sustainability of industrial production in the context of Industry 4.0 principles. The purpose of the work is to identify and measure the relationship between the level of ML implementation maturity and key production and environmental metrics, as well as to determine managerial mechanisms for achieving the effect. The empirical base includes 45 enterprises in metallurgy, chemical industry, and heavy engineering for 2018–2023, stratified by scale and ML maturity level (0–5). Regression and correlation analyses were applied with control for modernization investments, asset age, industry specifics, and clustering. The target indicators used were OEE, specific energy consumption (ECPU), waste generation rate (WGR), and carbon footprint (CF). Statistically significant effects were obtained: an increase in OEE by 14.33 p.p. due to improved availability, performance, and quality; a reduction in ECPU by 18.53% along with decreased energy consumption variability; a reduction in WGR by 22.31% due to defect reduction and process optimization; and a decrease in CF by 25.09% as a synergistic result of energy saving and resource efficiency. Regression estimates show that each level of ML maturity is associated with a 4.2% reduction in ECPU and a 5.1% reduction in WGR; however, the dependence is nonlinear: the greatest marginal effect occurs when moving from initial to intermediate levels, while further integration requires organizational and technological restructuring and yields diminishing returns. The discussion reveals the mechanism of the effect through predictive maintenance, real-time parameter optimization, and computer vision in quality control. Practical conclusions include a phased implementation strategy with quick wins, prioritization of energy-intensive units, and development of an end-to-end data architecture. Limitations concern industry representativeness and the share of secondary data; future directions include longitudinal panels, analysis of contributions from specific algorithmic classes, and assessment of the socio-economic consequences for employment and skills.