NVIDIA’s fintech and payments lead Georgios explains how AI in payments is evolving from traditional machine learning to payment foundation models.

For years, financial institutions relied on isolated models for fraud detection, underwriting, and personalization. But a new architecture is emerging. Companies are accelerating traditional ML with GPU computing, applying graph neural networks to analyze transaction relationships in real time, and now moving toward tabular transformer-based foundation models trained on payments data.

These models allow organizations to reuse the same underlying intelligence across multiple functions instead of building separate models for every problem.

The shift also changes the objective of AI deployment. What started as cost optimization through fraud reduction is now evolving into a broader strategy focused on data platforms, insights generation, and new product creation.

Once payments data is transformed into embeddings and insights, companies can power downstream applications such as personalization, transaction intelligence, and even agentic commerce, where AI systems anticipate customer behavior and initiate transactions.