The increasing complexity and volatility of global supply chains, driven by expanding market demands, technological advancements, and geopolitical uncertainties, challenge traditional forecasting methods. Conventional approaches often fail to capture the intricate dynamics of modern supply networks, leading to inefficiencies in operational costs and supply chain resilience. This paper evaluates the performance of Transformer-based models—specifically Transformer, Informer, Reformer, and Autoformer—for demand and supply forecasting in volatile environments. Contrary to expectations, empirical analysis reveals that the original Transformer outperforms its specialized counterparts in handling short-term fluctuations and sudden demand spikes. While models like Informer and Autoformer succeed in capturing long-term trends, they often miss short-term variations. These findings underscore the critical importance of model selection based on data characteristics and highlight opportunities for further exploration into balancing computational efficiency with forecasting accuracy of advanced machine learning models in dynamic supply chain scenarios.