European Financial Services Leverage NVIDIA-Powered Open Models for Scalable Proprietary AI

Image: NVIDIA · Source

Major European banks and fintech firms adopt NVIDIA accelerated computing and open weight models to develop proprietary AI, enhancing scalability, cost efficiency, and fraud detection within their own infrastructure.

At GTC Berlin, leading European financial institutions demonstrated production-scale artificial intelligence systems built on NVIDIA accelerated computing, leveraging open weight models deployed on their own infrastructure. Notably, payment giant Revolut is reconstructing its technology stack around a single foundation model called PRAGMA, powered by NVIDIA's cuDF toolkit, Dynamo-Triton software, and Blackwell Ultra GPUs. Designed to serve their initial 80 million customers with a scalable architecture targeting threefold growth, this initiative marks a significant advance in deploying AI at massive transaction scale. Pavel Nesterov, Revolut's head of AI, emphasized PRAGMA's role in bridging the gap between research breakthroughs and practical, large-scale banking applications.

Other European banks are also embracing AI innovations: Starling Bank developed "Starling Intelligence," an agentic AI model trained on U.K. sovereign cloud infrastructure using their proprietary transaction data. Lloyds Banking Group created ATUM, a transaction foundation model that transforms billions of banking records into reusable intelligence, accelerated by NVIDIA RTX PRO 6000 Blackwell GPUs to improve bank-wide operations. SBS introduced a sovereign intelligence platform enabling banks to run generative and agentic models without their data leaving their premises, addressing key data sovereignty concerns.

Open weight models offer these organizations the ability to build proprietary intelligence with full control over their data and costs. For instance, neobank bunq deploys its agentic assistant Finn on open weight models with NVIDIA acceleration, achieving comparable quality to leading closed models but at 60% lower cost per query. NatWest is developing AI technology capable of recognizing diverse British accents and financial terminology using NVIDIA's Nemotron 3.5 ASR speech model. Modulr and Santander also utilize NVIDIA's Nemotron and Kumo open models respectively to develop specialized AI tailored for payments and core banking workloads.

In addition to customer-facing AI, accelerated computing transforms financial research and fraud detection. London Stock Exchange Group (LSEG) utilizes NVIDIA's cuDF and Polars GPU engine on RTX PRO 6000 Blackwell GPUs to analyze vast amounts of historical tick data up to ten times faster. RavenPack leverages fine-tuned Nemotron models to generate labeled datasets supporting market trend analysis. HSBC applies RTX PRO 6000 Blackwell GPUs and NVIDIA cuDF to enable near-real-time client insights, enhancing decision-making and fraud risk reduction. Stripe experiences dramatic improvements with GPU-accelerated Ray Data, reporting 42x faster data pipelines and 92% cost savings in their fraud detection platform, continuously optimizing machine learning models.

These advancements reflect a broader trend where open models and NVIDIA's accelerated computing infrastructure empower European financial services organizations to develop scalable, proprietary AI solutions. This shift underscores the importance of data sovereignty, operational efficiency, and cost-effectiveness while fostering innovation in banking and payments sectors.