European banks turn to open models for proprietary AI

European banks, payment companies and financial data providers showcased deployments at GTC Berlin using NVIDIA accelerated computing and open weight models. The projects span transaction intelligence, customer assistants, financial research and fraud detection, with institutions emphasizing control over proprietary data and infrastructure.

Revolut is rebuilding its technology stack around PRAGMA, a foundation model serving an initial 80 million customers and engineered to scale 3x. Starling Bank developed its Starling Intelligence layer on Civo’s U.K. cloud using its own transaction data. Lloyds Banking Group built ATUM to turn banking transactions into reusable intelligence, while SBS launched a platform that lets banks run generative and agentic models without moving data outside their infrastructure.

bunq’s assistant Finn now uses open weight models deployed on Nebius, matching leading closed models at 60% of the cost per answer. NatWest is developing speech technology for British accents, dialects and financial terminology. Modulr is building specialized payment operations models, and Santander is adopting NVIDIA Kumo models for structured financial data.

LSEG is working with NVIDIA to analyze historical tick data up to 10x faster. HSBC reports 5x faster data processing supporting client insights and risk management. In offline testing, Stripe recorded 42x faster data pipelines and 92% cost savings for its Radar fraud-detection platform using NVIDIA GPU-accelerated Ray Data.