This special session focuses on the rapid advancement of Multimodal Large Language Models (MLLMs) and other foundational AI models and how industries leverage data for decision-making and process optimization. In smart manufacturing, these models can learn from heterogeneous data sources ranging from sensor streams and machine logs to visual inspection data and textual process documentation while demonstrating the ability to adapt across diverse industrial sectors. Their potential spans predictive maintenance, quality control, supply chain optimization, and real-time decision support, ultimately driving efficiency, sustainability, and innovation. Despite this promise, the deployment of such models in manufacturing settings introduces significant challenges, including data privacy, domain generalization, scalability, explainability, and integration with legacy systems. Addressing these issues requires coordinated efforts across academia, industry, and technology providers to establish best practices, create benchmarks, and explore novel architectures that meet the reliability and safety demands of industrial environments. The special session seeks to convene experts from AI, manufacturing, and Industry 4.0 to examine these opportunities and challenges, fostering dialogue on methods to advance cross-domain generalization and transfer learning for industrial applications. It further aims to encourage the sharing of open datasets, standardized benchmarks, and reproducible research frameworks to accelerate technological progress. By stimulating academic–industrial collaboration and outlining actionable research directions, the special session aspires to set the stage for the development of trustworthy, sustainable, and high-performance AI systems capable of transforming manufacturing into a more intelligent, autonomous, and resilient domain.