Large-scale commercial facilities depend on mechanical, electrical, and plumbing (MEP) assets whose failures can disrupt tenant services, compromise safety, increase energy use, and generate emergency work. This systematic literature review examines how artificial intelligence supports predictive maintenance of MEP assets by transforming operational data into early warnings, fault diagnoses, health classifications, and maintenance decisions. A structured review protocol, aligned with PRISMA 2020 reporting principles, synthesized 24 studies published between 2020 and 2026 together with the reporting guideline. Evidence was organized around maintenance strategies, artificial intelligence techniques, asset failure modes, data requirements, implementation settings, benefits, and barriers. The review shows that supervised learning, deep neural networks, autoencoders, transfer learning, hybrid rule-based models, and digital twins are increasingly used for chillers, air-handling units, sensors, actuators, pumps, and building energy systems. However, most evidence remains concentrated on HVAC applications, while electrical distribution, plumbing, fire protection, and vertical transportation receive limited attention. Recurring implementation constraints include scarce labeled fault data, inconsistent sensor quality, weak interoperability between building automation and maintenance systems, limited model explainability, cybersecurity exposure, and insufficient feedback from technicians. The paper proposes a framework linking asset criticality, data governance, model validation, work-order integration, and continuous learning for reliable deployment in commercial facilities.