Financial Institutions Embrace Transaction Foundation Models for Intelligence

NVIDIA and Revolut Transform Financial AI with Transaction Foundation Models

Financial institutions are on the brink of a significant transformation in artificial intelligence (AI) applications, as NVIDIA and Revolut unveil new transaction foundation models. This innovative approach aims to unify disparate AI systems, providing a comprehensive understanding of consumer financial behavior. The collaboration was announced recently, highlighting an urgent need for financial firms to enhance their AI capabilities amidst growing data complexities.

The Challenge of Siloed Systems

For years, financial institutions have developed various AI models tailored for specific tasks, such as fraud detection and credit scoring. However, these siloed systems limit the ability of organizations to gain a holistic view of consumer behavior. As datasets expand, the gap between available data and actionable insights widens, presenting an opportunity for the industry to leverage proprietary data more effectively.

NVIDIA’s 2026 State of AI in Financial Services report indicates that 65% of institutions currently utilize AI technologies, with nearly 90% either deploying or evaluating their use. Despite this widespread adoption, the complexity associated with scaling these technologies often leads to fragmented model architectures that hinder performance.

To address these challenges, leading firms are rethinking their architectural approaches. Instead of relying on traditional statistical and machine learning algorithms designed for specific business lines, they are now turning to transformer-based transaction foundation models. These models enable organizations to learn a unified representation of consumer behavior by training on proprietary data sets.

Understanding Transaction Foundation Models

Transaction foundation models represent a paradigm shift in how financial data is processed and analyzed. These large-scale AI systems are trained on billions of financial events—including payments, transfers, product interactions, and behavioral signals—transforming raw data into actionable intelligence that can enhance customer service.

The structural shift from traditional fraud detection methods to foundation models allows for contextual interpretation of consumer behavior. For instance, a payment made at midnight carries different implications if it follows several other transactions made within a short timeframe or occurs on an unfamiliar device in a new location. This contextual depth significantly enhances performance across various tasks rather than limiting it to isolated functions.

Revolut’s collaboration with NVIDIA led to the development of PRAGMA—a family of transformer-based foundation models trained on 24 billion events from 26 million user records across over 100 countries. Utilizing NVIDIA’s advanced AI stack—including Hopper GPUs and cuDF libraries—this model outperforms traditional task-specific models in areas such as credit scoring and fraud detection while minimizing reliance on manually crafted features.

Tadas Kriščiūnas, head of group credit data science at Revolut, emphasized the efficiency gained through this approach: “We move from weeks, or even in some cases months, in feature engineering to no time required for it at all.”

The Cost of Fragmentation

The proliferation of specialized models poses ongoing challenges for financial institutions. Each new use case necessitates additional model development and retraining efforts that fail to share contextual insights across systems. This fragmentation ultimately leaves untapped potential value on the table.

Mastercard is actively addressing this issue by developing its own large tabular foundation model for payments. Trained on billions of anonymized transactions, this model is designed to scale significantly while integrating diverse datasets related to fraud detection and customer loyalty.

Supported by NVIDIA’s technologies along with AWS and Databricks capabilities—including the NVIDIA NeMo AutoModel open library—Mastercard’s initiative aims to reduce dependence on multiple AI models across various markets and applications. Preliminary results indicate that this model outperforms standard machine learning techniques in several areas including cybersecurity and personalized services.

Agentic Commerce and Its Implications

A growing number of financial firms—42% according to recent surveys—are exploring agentic AI systems capable of executing transactions autonomously. These systems can manage subscriptions, route payments, and facilitate purchases based on a comprehensive understanding of transactional contexts.

Stripe exemplifies this trend by leveraging NVIDIA’s platform to build foundation models that grasp the full context behind transactional behavior rather than merely responding to isolated signals. Last year alone, Stripe blocked nearly $112 billion in fraudulent transactions while achieving an average reduction in fraud rates by 38%.

The proprietary nature of transaction data gives companies like Stripe a competitive edge that is difficult for others to replicate. With proven architecture already established and infrastructure ready for deployment, financial institutions are well-positioned to capitalize on these advancements.

Scaling Through Ecosystem Partners

NVIDIA’s Build Your Own Transaction Foundation Model developer example is now available for customers through Amazon Web Services (AWS) using Amazon SageMaker HyperPod or Nebius AI Cloud powered by NVIDIA accelerated computing. This resource supports the entire lifecycle—from deployment through multi-node training—to managed inference operations.

Financial services firms can collaborate with partners like EXL, Infosys, GFT IT Consulting, and Thoughtworks to tailor this developer example for their unique needs. For instance, EXL is integrating transaction foundation models into its EXLerate.ai platform aimed at unifying siloed financial data into a scalable intelligence layer powered by proprietary transaction information.

Thoughtworks is assisting institutions in operationalizing these models within complex banking environments while ensuring necessary governance structures are established. GFT IT Consulting is also embedding transaction foundation models into its solutions like Wynxx—a platform utilized by over 100 financial institutions focused on secure AI adoption.

What This Means

The introduction of transaction foundation models marks a pivotal moment for financial institutions seeking to enhance their AI capabilities. By moving away from fragmented systems toward unified architectures capable of contextual understanding, organizations can unlock new levels of insight into consumer behavior while improving efficiency across various applications. As more firms adopt these transformative technologies, the landscape of financial services will continue evolving rapidly towards smarter decision-making processes driven by advanced AI systems.

For more information, read the original report here.