European financial institutions scale open AI models on proprietary infrastructure

Traders Union
All News
Financial Institutions
European banks scale open AI

Highlights

Revolut launched its PRAGMA foundation model on NVIDIA Blackwell Ultra GPUs for 80 million customers and plans to scale usage threefold.
Starling Bank, Lloyds Banking Group, and bunq deployed open weight AI models on sovereign cloud or proprietary GPU infrastructure to enhance transaction intelligence and cost efficiency.
Stripe reported 42 times faster data pipelines and 92% cost savings for its Stripe Radar fraud-detection using NVIDIA GPU-accelerated Ray Data in offline tests.

Production deployments broaden across banking and payments

NVIDIA Newsroom said at GTC Berlin that financial institutions across Europe are moving beyond pilot projects and putting AI systems into production on infrastructure they control, using open weight models and NVIDIA accelerated computing.

Revolut is rebuilding its technology stack around a single foundation model called PRAGMA. The company says the system runs with the NVIDIA cuDF toolkit and Dynamo-Triton software on NVIDIA Blackwell Ultra GPUs, serving an initial 80 million customers and designed to scale threefold.

Pavel Nesterov, head of AI at Revolut, said the company processes billions of transactions and saw a question over whether foundation models could work at that scale across domains. He said PRAGMA, built in collaboration with NVIDIA, shows that research advances can be translated into production banking systems.

Starling Bank has developed Starling Intelligence on Civo's UK sovereign cloud using its own transaction data, and NVIDIA says it is the first bank to train an agentic model on UK sovereign cloud infrastructure. Lloyds Banking Group has built its ATUM transaction foundation model with NVIDIA RTX PRO 6000 Blackwell GPUs to turn billions of banking transactions into reusable intelligence across bank-wide uses.

SBS has launched a banking sovereign intelligence platform aimed at letting banks run generative and agentic models without data leaving their own environments. Neobank bunq has also shifted its assistant Finn to open weight models, saying deployment on Nebius matches the quality of leading closed models at 60% of the cost per answer.

Regional push centers on cost control, data sovereignty and faster analytics

Other institutions are applying the same approach to specialized workloads. NatWest is building technology intended to recognize a broad range of British accents, dialects and financial terms with NVIDIA speech models, including Nemotron 3.5 ASR, while Modulr is developing models for distinct parts of payment operations based on the NVIDIA Nemotron-3-Ultra-550B open model.

Santander is using the NVIDIA Kumo family of open weight foundation models for structured financial data to build intelligence for core banking workloads. LSEG is working with NVIDIA to help institutions analyze large volumes of historical tick data up to 10 times faster with the Polars GPU engine, accelerated by NVIDIA cuDF on NVIDIA RTX PRO 6000 Blackwell GPUs.

RavenPack has built a dataset generation pipeline on Bigdata.com that turns plain-language research hypotheses into labeled datasets ready for backtesting through a fine-tuned NVIDIA Nemotron model. NVIDIA also says its RTX PRO 6000 Blackwell Server Edition GPUs and cuDF can speed data processing across multi-terabyte regulated and siloed sources, shortening the time from data to decision in areas such as fraud detection.

HSBC provides near-real-time client insights using Starburst Enterprise's GPU Engine with data processing that NVIDIA says is five times faster. Stripe, meanwhile, reports 42 times faster data pipelines and 92% cost savings for its Stripe Radar fraud-detection platform with NVIDIA GPU-accelerated Ray Data in offline testing, supporting more continuous model training and fraud detection improvements.

Our earlier report on the Bank of England’s push for “bare metal” recovery planning explained why UK regulators want banks to be able to restore critical services on independent systems if core technology is compromised. It highlighted concerns that AI is increasing the scale and sophistication of cyber attacks, prompting supervisors to demand stronger end-to-end resilience and clearer recovery options beyond perimeter defences.

This material may contain third-party opinions, none of the data and information on this webpage constitutes investment advice according to our Disclaimer. While we adhere to strict Editorial Integrity, this post may contain references to products from our partners.