Revolut has launched Revolut Research, a dedicated AI division aimed at building a native foundation model called PRAGMA in collaboration with NVIDIA to power financial services.

The leadership emphasizes a build-from-scratch philosophy—training a single unified foundation model on global operational data to enable smarter features, faster deployment, and improved safety and customer experience.

The goal is a single, unified model for financial behavior across products and markets, moving away from multiple narrow models for tasks like fraud screening, credit scoring, and customer support.

Early deployments of PRAGMA on historical data show major performance gains, validating the efficiency of the unified foundation approach.

Executives reiterate a build-from-scratch approach with a unified foundation model to capture real-time financial behavior, and they aim to differentiate Revolut by collaborating with global tech leaders.

The initiative is positioned in a broader AI-and-hardware ecosystem, with links to Quantum Zeitgeist and NVIDIA highlighting external collaboration.

Revolut has not disclosed the division’s financials or its NVIDIA collaboration, but the move signals intensified internal AI investment amid rising competition in digital banking.

Revolut processes billions of cross-border transactions for over 80 million customers across 40+ markets, building a massive global dataset to enhance model performance.

PRAGMA will train on Revolut’s global operational data, creating a self-reinforcing loop as the dataset and model improve together, particularly for fraud detection and risk evaluation.

The sheer scale—over 80 million customers in 40 markets—provides a continuous data flywheel to refine AI-driven features and predictions.

This initiative signals a strategic shift toward in-house research and proprietary foundation models designed for global financial scale.

A detailed benchmark of PRAGMA’s performance is documented in a NVIDIA-hosted case study.