Leading European Financial Services Organizations Embrace Open Models to Build Proprietary Intelligence

At GTC Berlin, NVIDIA showcased a long list of European financial firms building AI on open-weight models and their own infrastructure. Revolut says its PRAGMA foundation model is in production serving an initial 80 million customers. Lloyds built ATUM, a transaction foundation model. Starling Bank trained what it calls Starling Intelligence on U.K. sovereign cloud using its own transaction data. Santander is an early adopter of NVIDIA's new Kumo open models for structured financial data.

The common thread is control. Banks want models trained on data they own, running where regulators can see them, at costs they set. That's a different path from renting intelligence from a closed model provider.

We think this is a healthy direction for finance, with an asterisk. Banks sit on exactly the kind of proprietary transaction data that general-purpose chatbots never see, and they operate under rules about where data can live. Open weights let them train on that data without shipping it to someone else. The bunq example makes the economic case: NVIDIA says its Finn assistant, now on open models deployed via Nebius, matches leading closed models on quality at 60% of the cost per answer.

The asterisk is the messenger. This is an NVIDIA blog post written by its regional financial-services sales lead, and every customer story is framed around NVIDIA hardware and software. The numbers, including Stripe's reported 42x faster data pipelines and 92% cost savings for Radar fraud detection, are presented without methodology, and Stripe's figures come from offline testing rather than production. We'd treat them as directional until the institutions themselves publish more.

There's also a quieter tension in the word open. These banks are escaping dependence on closed model vendors, but many of the models here, from Nemotron to Kumo, come from NVIDIA itself, tuned for its GPUs. Swapping one dependency for another is still a dependency. I'm watching whether open weights translate into genuine portability across hardware, or mostly into a different shape of lock-in.

The stakes for customers are real either way. Transaction foundation models that turn billions of payments into reusable intelligence can catch fraud faster, and HSBC and Stripe both point to fraud work. The same models can drive credit decisions, pricing and account flags at enormous scale. When a model serving tens of millions of people gets something wrong, the error is systematic, not individual. Banks building their own models also own the duty to explain them.

Our read: owning the model is the right instinct for regulated finance. Now the industry has to show the governance is as advanced as the GPUs.

Other examples in the post include NatWest building speech recognition for British accents and financial terms with NVIDIA's Nemotron 3.5 ASR, Modulr developing payment-operations models on Nemotron-3-Ultra-550B, SBS launching a platform for banks to run generative and agentic models inside their own walls, and LSEG working with NVIDIA on analyzing historical tick data up to 10x faster.