Transforming Financial Services with NVIDIA Accelerated Computing

At the recent GTC Berlin event, the spotlight was on how banks, payment companies, and financial data providers are revolutionizing their operations through NVIDIA accelerated computing. This deep dive into production-scale intelligence, built on open weight models and powered by NVIDIA’s technology, showcases remarkable advancements that are reshaping the financial landscape.

Revolutionizing Technology with PRAGMA at Revolut

Revolut is making waves in the banking sector by reconstructing its technology stack around the foundation model known as PRAGMA. With the robust support of NVIDIA accelerated computing, Revolut is deploying this model across significant customer bases—initially serving 80 million customers and designed to scale threefold. Pavel Nesterov, the head of AI at Revolut, emphasized the remarkable potential of PRAGMA, stating, “It’s the clearest proof yet that the gap between a research breakthrough and a production banking system can be closed.”

Equipped with tools like NVIDIA cuDF and Dynamo-Triton software on NVIDIA Blackwell Ultra GPUs, Revolut not only processes billions of transactions but also leverages foundation models to orchestrate seamless interactions across different domains.

Training AI Models on Accelerated Infrastructure

In the realm of banking innovation, Starling Bank stands out as the first institution to train an agentic model within the framework of U.K. sovereign cloud infrastructure. This undertaking birthed Starling Intelligence, a sophisticated AI capability developed using the bank’s transaction data on Civo’s U.K. cloud.

Meanwhile, the Lloyds Banking Group has developed ATUM, a transaction foundation model that ingeniously converts billions of banking transactions into reusable intelligence. This model, accelerated by the NVIDIA RTX PRO 6000 Blackwell GPUs, is not only efficient but also facilitates a broad spectrum of bank-wide applications.

Moreover, SBS has launched a banking sovereign intelligence platform enabling banks to implement generative and agentic models, ensuring data security within their confines.

Embracing Open Models for Cost-Effective Solutions

One of the significant game-changers in the financial industry is the adoption of open weight models. These models provide European financial institutions with proprietary intelligence built on their existing data while maintaining cost control. The NVIDIA accelerated computing platform is the backbone supporting these advancements.

For instance, the neobank bunq has integrated an agentic assistant named Finn that operates on open weight models. By deploying on Nebius, bunq boasts a solution that matches the quality of leading closed models at only 60% of the cost per answer.

Furthermore, NatWest is innovating technology aimed at recognizing the nuances of British accents and dialects, developed using NVIDIA’s speech models, including Nemotron 3.5 ASR. This initiative reflects a commitment to enhancing customer experiences through tailored interactions.

Modulr is making strides too by creating specialized models for various aspects of payment operations, leveraging the NVIDIA Nemotron-3-Ultra-550B open model.

Santander is an early adopter of the latest NVIDIA Kumo family of open weight foundation models for structured financial data. This adoption enables the bank to craft specialized intelligence for core banking functions.

Accelerating Financial Research with Advanced Tools

The collaboration between LSEG and NVIDIA is propelling the analysis of large historical tick data to unprecedented speeds—up to 10x faster—with the introduction of the Polars GPU engine. Utilizing NVIDIA cuDF on NVIDIA RTX PRO 6000 Blackwell GPUs empowers financial institutions to analyze significant data volumes effectively.

RavenPack is enhancing its operations by developing a dataset generation pipeline on Bigdata.com. This innovative tool processes plain-language research hypotheses and returns labeled, backtest-ready datasets, powered by a fine-tuned NVIDIA Nemotron model. Investors gain valuable insights to identify, analyze, and visualize evolving market trends.

Enhancing Fraud Detection Through Accelerated Data Processing

In today’s fast-paced environment, timely decision-making is critical, especially concerning fraud detection. The RTX PRO 6000 Blackwell Server Edition GPUs and NVIDIA cuDF are instrumental in accelerating data processing across multi-terabyte-scale, regulated sources. This capability significantly reduces the time needed from data acquisition to actionable insights, particularly in fraud scenarios.

HSBC capitalizes on this technology, providing near-real-time client insights that facilitate customer processes and on-demand decision-making, all while minimizing risk. The implementation of Starburst Enterprise’s GPU Engine enhances data processing speeds by 5x.

Stripe, a leader in programmable financial services, reports a staggering 42x improvement in data pipeline speed and 92% cost savings for its Stripe Radar fraud-detection platform, thanks to NVIDIA GPU-accelerated Ray Data. This enhancement allows continuous optimization of machine learning pipelines, significantly bolstering fraud detection capabilities.

This glimpse into how financial services are leveraging NVIDIA’s accelerated computing technologies demonstrates a dynamic shift towards innovative, efficient, and secure operations. With these advancements, the future of banking and financial services looks promising, bringing speed and intelligence into every transaction.

Inspired by: Source