Why Financial Institutions Are Converging on Transaction Foundation Models to Build Their Own Intelligence
NVIDIA Blog · Pahal Patangia · 2026-06-02
NVIDIA's blog describes how financial institutions including Revolut, Mastercard, and Stripe are replacing siloed task-specific AI models with transformer-based transaction foundation models trained on billions of financial events to improve fraud detection, credit scoring, and personalization at scale.
Extraction
Topics: financial-aitransaction-foundation-modelstransformer-modelsfraud-detectionenterprise-ai
Claims
- 65% of financial institutions now use AI, with nearly 90% deploying or assessing it according to NVIDIA's 2026 State of AI in Financial Services report.
- Revolut's PRAGMA foundation model, trained on 24 billion events across 26 million user records, outperforms strong task-specific models across credit scoring, fraud detection, and product recommendations.
- Transaction foundation models eliminate weeks or months of manual feature engineering by learning unified behavioral representations.
- Stripe blocked close to $112 billion in fraud last year and delivered an average 38% reduction in fraud rates using transaction foundation models.
- Mastercard is developing a proprietary large tabular foundation model designed to scale to hundreds of billions of transactions across fraud, authorization, chargeback, and loyalty data.
Key quotes
We move from weeks, or even in some cases months, in feature engineering to no time required for it at all.
Even fractional improvements like a 0.1% uplift in authorization can translate to massive incremental gross merchandise value and substantial cost reductions.
Transaction data is the proprietary history that competitors can't replicate.