Financial sector companies, including banks, payment systems, and data providers, are increasingly focusing on developing their own Artificial Intelligence solutions. At the recent GTC Berlin conference, these organizations showcased real-world use cases for AI implementations built on NVIDIA’s accelerated computing platforms, leveraging open-source models and deploying them on private infrastructure.
Proprietary Development Efforts
Revolut is scaling its technology platform with a unified foundational model named PRAGMA. Utilizing NVIDIA’s cuDF tools, Dynamo-Triton software, and NVIDIA Blackwell Ultra GPUs, Revolut is deploying PRAGMA to serve its 80 million customers, with plans for threefold growth. Pavel Nesterov, Head of AI at Revolut, highlighted the model’s capability to handle billions of transactions across various domains, stating, “PRAGMA has proven this.” He added that the collaboration with NVIDIA demonstrates how the gap between groundbreaking research and a production-ready banking system can be significantly reduced.
Starling Bank has taken a pioneering step by training an agent-based model on sovereign cloud infrastructure within the UK. Its AI capabilities are powered by “Starling Intelligence,” developed on Civo’s cloud and utilizing the bank’s proprietary transaction data.
Lloyds Banking Group has developed ATUM, a foundational transaction model designed to transform billions of banking operations into reusable, intelligent data. This solution is accelerated by NVIDIA RTX PRO 6000 Blackwell GPUs, enhancing efficiency across banking processes.
Furthermore, SBS has launched a sovereign banking intelligence platform that empowers banks to deploy generative and agent-based models without exposing their sensitive data beyond the corporate network.
Open Models Drive Cost-Effective Innovation
The adoption of open-source models allows European financial institutions to build unique AI capabilities tailored to their specific data while maintaining cost control. NVIDIA’s accelerated computing platform serves as the backbone for these initiatives.
Neobank bunq is now leveraging Finn, an agent assistant powered by open models. This solution, built on NVIDIA accelerated computing and deployed via the Nebius platform, reportedly achieves the quality of leading proprietary models at 60% of the cost per query.
NatWest is developing technology to recognize a wide range of British accents, dialects, and financial terminology, employing NVIDIA’s language models, including Nemotron 3.5 ASR.
Modulr is creating specialized models for various payment operations, building upon NVIDIA’s open-source Nemotron-3-Ultra-550B model.
Santander is among the early adopters of NVIDIA’s new Kumo family of open foundational models for structured financial data, using them to develop specialized AI for key banking functions.
Accelerated Computing Transforms Financial Research
LSEG is collaborating with NVIDIA to analyze vast volumes of historical tick data up to 10 times faster. This is achieved using the Polars GPU engine, running on NVIDIA RTX PRO 6000 Blackwell GPUs and accelerated by NVIDIA cuDF.
RavenPack has established a data generation pipeline on Bigdata.com that converts research hypotheses stated in natural language into a labeled dataset ready for backtesting. This solution utilizes a fine-tuned NVIDIA Nemotron model, enabling investors to identify, analyze, and visualize market trends.
Accelerating Data Pipelines for Enhanced Fraud Detection
RTX PRO 6000 Blackwell Server Edition GPUs and NVIDIA cuDF are instrumental in accelerating the processing of multi-terabyte data from diverse, regulated sources. This significantly reduces the time from data ingestion to decision-making for fraud detection and other critical scenarios.
HSBC is delivering near real-time customer data to support client processes and on-demand decision-making, thereby minimizing risks and fraud. This capability is enabled by a fivefold acceleration in data processing on the GPU Engine of its Starburst Enterprise platform.
Stripe, a global provider of programmable financial services, has reported a 42x acceleration in data pipelines and 92% cost savings for its Stripe Radar fraud detection platform during offline testing with NVIDIA GPU-accelerated Ray Data. This advancement allows Stripe Radar’s machine learning pipelines to continuously optimize model training and improve fraud detection accuracy.
