European Financial Firms Build Proprietary AI on Open Models at GTC Berlin
NVIDIA says European banks and financial firms are deploying open weight models and its accelerated computing hardware to build proprietary AI systems on their own infrastructure.
ひとことで言うと
What did NVIDIA announce about European financial institutions and open models at GTC Berlin?
NVIDIA says European banks, payment firms and data providers including Revolut, Starling Bank, Lloyds, bunq, NatWest, Santander, HSBC and Stripe are using open weight models and NVIDIA accelerated computing to build proprietary AI for banking, fraud detection and financial research, running on their own infrastructure.
要点
- NVIDIA says Revolut is rebuilding its technology stack on a single foundation model called PRAGMA, using NVIDIA cuDF and Dynamo-Triton software on Blackwell Ultra GPUs to serve 80 million customers.
- NVIDIA says Starling Bank trained an agentic model, called Starling Intelligence, on U.K. sovereign cloud infrastructure via Civo, using the bank's own transaction data.
- NVIDIA says Lloyds Banking Group built a transaction foundation model called ATUM using RTX PRO 6000 Blackwell GPUs, and bunq's Finn assistant now runs on open weight models deployed on Nebius at 60% of the cost per answer versus closed models.
- NVIDIA says Modulr is building payment-specific models on the Nemotron-3-Ultra-550B open model, and Santander is an early adopter of NVIDIA's new Kumo family of open weight models for structured financial data.
- NVIDIA says Stripe reports 42x faster data pipelines and 92% cost savings in offline testing of its Radar fraud-detection platform using NVIDIA GPU-accelerated Ray Data.
NVIDIA said a group of European banks, payment companies and financial data providers are deploying open weight AI models on its accelerated computing hardware to build proprietary intelligence systems running on their own infrastructure, according to announcements made at GTC Berlin on October 21, 2026.
Foundation models built in-house
NVIDIA said Revolut is rebuilding its technology stack around a single foundation model called PRAGMA, using the NVIDIA cuDF toolkit and Dynamo-Triton software on NVIDIA Blackwell Ultra GPUs. The company said the model is already servicing 80 million customers and is engineered to scale three times that size. Revolut's head of AI, Pavel Nesterov, said PRAGMA demonstrates that the gap between a research breakthrough and a production banking system can be closed.
NVIDIA also said Starling Bank trained an agentic model, which it calls Starling Intelligence, on U.K. sovereign cloud infrastructure provided by Civo, using the bank's own transaction data — a step NVIDIA described as a first for a bank training on sovereign cloud. Separately, NVIDIA said Lloyds Banking Group built a transaction foundation model called ATUM, accelerated with NVIDIA RTX PRO 6000 Blackwell GPUs, to convert banking transaction data into reusable intelligence across the bank. NVIDIA said SBS launched a sovereign intelligence platform that lets banks run generative and agentic models without their data leaving their own systems.
Open models and cost reductions
NVIDIA said neobank bunq's agentic assistant, Finn, now runs on open weight models deployed on Nebius, which the company said matches the quality of leading closed models at 60% of the cost per answer. NVIDIA said NatWest is developing speech recognition technology for British accents, dialects and financial terminology using NVIDIA's speech models, including Nemotron 3.5 ASR. The company said Modulr is building specialized payment-operations models based on the NVIDIA Nemotron-3-Ultra-550B open model, and that Santander is an early adopter of NVIDIA's new Kumo family of open weight foundation models for structured financial data.
Research and fraud detection
NVIDIA said it is working with LSEG to help financial institutions analyze historical tick data up to 10 times faster using the Polars GPU engine, accelerated by NVIDIA cuDF on RTX PRO 6000 Blackwell GPUs. The company said RavenPack built a dataset generation pipeline on Bigdata.com, running on a fine-tuned NVIDIA Nemotron model, that converts a plain-language research hypothesis into a labeled, backtest-ready dataset.
On fraud detection, NVIDIA said HSBC is using Starburst Enterprise's GPU Engine to process data 5x faster for near-real-time client insights. The company said Stripe reported 42 times faster data pipelines and 92% cost savings in offline testing of its Stripe Radar fraud-detection platform, using NVIDIA GPU-accelerated Ray Data to continuously optimize its machine learning training.
Source: NVIDIA Blog, "European Financial Institutions Embrace Open Models to Build Proprietary Intelligence," published October 21, 2026.
よくある質問
- What is PRAGMA?
- According to NVIDIA, PRAGMA is a single foundation model that Revolut is using to rebuild its technology stack, powered by NVIDIA accelerated computing and engineered to scale to three times its initial 80-million-customer base.
- Which banks built their own foundation models?
- NVIDIA says Starling Bank built Starling Intelligence on U.K. sovereign cloud infrastructure, and Lloyds Banking Group built ATUM, a transaction foundation model, using NVIDIA RTX PRO 6000 Blackwell GPUs.
- How is NVIDIA hardware being used for fraud detection?
- NVIDIA says HSBC uses Starburst Enterprise's GPU Engine for 5x faster data processing on client insights, and Stripe uses NVIDIA GPU-accelerated Ray Data to speed up its Radar fraud-detection pipelines by 42x in offline testing with 92% cost savings.
- What open weight models are financial firms using?
- NVIDIA cites Nemotron 3.5 ASR for NatWest's speech recognition work, Nemotron-3-Ultra-550B for Modulr's payment models, a fine-tuned Nemotron model for RavenPack's dataset pipeline, and the new Kumo family of open weight models for Santander's structured financial data work.