NVIDIA highlighted European financial firms deploying open-weight models and accelerated computing for banking, payments, research and fraud workflows.
At GTC Berlin, NVIDIA described a range of European financial institutions using open-weight models and accelerated computing for production systems. The examples span banking, payments, financial research and fraud detection, with organizations adapting models and data pipelines to specific operational needs.
Revolut is rebuilding its technology stack around PRAGMA, a foundation model developed in collaboration with NVIDIA. The company is using NVIDIA cuDF and Dynamo-Triton software on Blackwell Ultra GPUs; NVIDIA says the system is serving an initial 80 million customers and is engineered to scale threefold. Revolut’s head of AI, Pavel Nesterov, said the model is intended to handle the company’s transaction volume and varied domains.
Other institutions are developing models and services around their own data. Starling Bank developed Starling Intelligence using its transaction data on Civo’s U.K. cloud. Lloyds Banking Group built ATUM, a transaction foundation model accelerated with NVIDIA RTX PRO 6000 Blackwell GPUs. SBS launched a platform intended to let banks run generative and agentic models without their data leaving their own infrastructure.
The examples also include customer service, payments and market analysis. Neobank bunq runs its Finn assistant on open-weight models deployed on Nebius; NVIDIA says the setup matches the quality of leading closed models at 60% of the cost per answer. Santander is adopting NVIDIA’s Kumo family for structured financial data, while Modulr is building payment-operations models based on Nemotron-3-Ultra-550B. NatWest is developing speech technology for British accents, dialects and financial terminology.
NVIDIA also cited data-processing work: LSEG is working to analyze historical tick data up to 10 times faster, and Stripe reported 42-times-faster data pipelines and 92% cost savings in offline testing for its Radar fraud-detection platform. These are company-reported examples, not a single standardized comparison. Together, they show how financial firms are applying models and accelerated infrastructure to distinct workloads while retaining control over proprietary data and deployment choices.