At GTC Berlin, NVIDIA presented several case studies from the European financial sector illustrating a shift toward open-weight models deployed in-house, rather than proprietary AI services accessed solely through third-party APIs. The announcement groups banks, payment providers and financial data companies that have chosen to build what NVIDIA calls "proprietary intelligence" — custom AI systems trained or fine-tuned on open models instead of relying exclusively on outside vendors.
Revolut is the most detailed example cited. The fintech company is restructuring its entire technology stack around a single foundation model, internally named PRAGMA, built on NVIDIA's accelerated computing infrastructure. The company reportedly uses NVIDIA's cuDF library, a GPU-accelerated dataframe tool, within the data pipeline that feeds this model. NVIDIA's announcement does not disclose PRAGMA's size, architecture or specific intended use cases.
The broader narrative NVIDIA pushed at the event reflects a pattern increasingly visible among regulated industries in Europe: a preference for models whose weights can be hosted on-premises or in private cloud environments, rather than sending sensitive financial data to external API providers. For banks and payment companies bound by strict data residency and auditability requirements, running open models on owned infrastructure offers greater control over compliance, latency and long-term cost structures.
This remains primarily a marketing showcase for NVIDIA's hardware and software stack, built around customer testimonials rather than new independent product launches. It nonetheless points to a continuing shift in enterprise AI strategy, where large financial institutions increasingly treat foundation models as infrastructure to own and customize rather than a subscribed service. The approach carries its own challenges, from the engineering resources needed to fine-tune and maintain large open models to the governance structures required to validate their outputs in regulated settings. Whether this translates into measurable efficiency gains for Revolut and its peers remains to be shown, beyond the promotional framing of a vendor conference.