Metal Additive Manufacturing (MAM) has reshaped the manufacturing industry, offering benefits like intricate design, minimal waste, rapid prototyping, material versatility, and customized solutions. However, its full industry adoption faces hurdles, particularly in achieving consistent product quality. A crucial aspect for MAM's success is understanding the relationship between process parameters and melt pool characteristics. Integrating Artificial Intelligence (AI) into MAM is essential. Traditional machine learning (ML) methods, while effective, depend on large datasets to capture complex relationships, a significant challenge in MAM due to the extensive time and resources required for dataset creation. Our study introduces a novel surprise-guided sequential learning framework, SurpriseAF-BO, signaling a significant shift in MAM. This framework uses an iterative, adaptive learning process, modeling the dynamics between process parameters and melt pool characteristics with limited data, a key benefit in MAM's cyber manufacturing context. Compared to traditional ML models, our sequential learning method shows enhanced predictive accuracy for melt pool dimensions. Further improving our approach, we integrated a Conditional Tabular Generative Adversarial Network (CTGAN) into our framework, forming the CT-SurpriseAF-BO. This produces synthetic data resembling real experimental data, improving learning effectiveness. This enhancement boosts predictive precision without requiring additional physical experiments. Our study demonstrates the power of advanced data-driven techniques in cyber manufacturing and the substantial impact of sequential AI and ML, particularly in overcoming MAM's traditional challenges.