At GTC Berlin, NVIDIA said Revolut is rebuilding its technology stack on a single foundation model, PRAGMA, powered by NVIDIA accelerated computing, and is scaling it in production on Blackwell Ultra GPUs using the cuDF toolkit and Dynamo-Triton software, initially serving 80 million customers and engineered to scale threefold. Pavel Nesterov, Revolut's head of AI, said the bank processes billions of transactions and that PRAGMA answers whether foundation models can operate at that scale and across domains, describing it as built in collaboration with NVIDIA. The same vendor post reported other European financial firms, including Starling Bank's Starling Intelligence agentic model trained on Civo's U.K. cloud with its own transaction data, Lloyds' ATUM transaction foundation model, bunq's agentic assistant Finn on open weight models at a claimed 60% of the cost per answer, and Santander as an early adopter of NVIDIA's Kumo open weight models for structured financial data.