Europe’s financial services sector is shifting its approach to artificial intelligence, moving away from proprietary systems towards open models powered by NVIDIA technology. At the GTC Berlin conference, Revolut announced plans to overhaul its technology infrastructure using a single foundation model named PRAGMA, which operates entirely on its own hardware. This initiative aligns with a growing trend among European banks, payment processors, and data providers aiming to enhance their AI capabilities.
Revolut, a digital bank with over 40 million customers in Europe, is consolidating its AI functions onto the PRAGMA model, marking a significant reconfiguration of its technical operations. The announcement at NVIDIA’s event reflects a broader rethinking of AI strategies within European financial institutions.
Other organizations in the sector also showcased AI systems built on similar architectures during the conference, emphasizing NVIDIA’s accelerated computing capabilities and the use of open weight models on private infrastructure. This strategy challenges the prevailing closed model approach seen in Silicon Valley, where firms like OpenAI and Google maintain tight control over the AI models and their underlying data.
For European banks, the need for control over AI systems is influenced by regulatory pressures, particularly European Union data protection laws such as GDPR. These regulations create potential legal challenges when handling customer financial data with third-party AI services. By developing their own models, banks seek to ensure compliance and maintain strategic advantages.
Revolut’s PRAGMA model uses NVIDIA’s cuDF, a GPU-accelerated data frame library, enabling rapid processing of large datasets—an essential capability for real-time operations such as fraud detection and credit assessments. The company believes that a model specifically trained on its data will offer better performance than more generalized AI solutions from other tech giants.
This architectural choice also addresses concerns about AI model collapse, where the widespread use of identical base models leads to a loss of competitive differentiation. Open weight models allow banks to develop unique, proprietary AI systems that leverage distinct datasets.
NVIDIA plays a central role in this transition, providing not only hardware but also software frameworks and reference architectures that facilitate AI deployment for financial institutions without needing extensive teams of machine learning engineers. Their recent investment in European AI infrastructure aims to empower banks with the physical resources necessary to implement their software ambitions.
The timing of these developments comes as new partnerships and initiatives are emerging in the European AI space. Following NVIDIA’s backing of several German startups and the German government’s push for accelerated AI adoption, the focus now shifts to how well these production-scale systems can perform—especially in comparison to existing solutions from major players.
However, there are challenges ahead for smaller regional banks that may struggle to invest in the necessary computational resources and talent to develop their own models. This could deepen the divide between AI-native institutions that control their own technologies and those that rely on external vendors.
The landscape is rapidly evolving. While NVIDIA remains a leader in hardware, companies like Amazon and Google are introducing custom chips and migration tools as potential alternatives to mitigate any dependency on NVIDIA’s offerings. New products like Amazon’s TorchNeuron and Google’s PyTorch compatibility layer have been launched to support banks’ diversified AI strategies.
As European financial institutions carve out a distinctive AI path, the stakes are high. U.S. banks are exploring services from tech giants like Microsoft and Google, while European entities invest in building tailored models. This strategic difference could yield significant outcomes depending on the effectiveness of the resulting systems. With Revolut’s PRAGMA model now processing actual customer transactions, its performance will be closely monitored and may influence future developments in AI strategies across the continent.
