NVIDIA says Revolut is scaling its PRAGMA foundation model in production for an initial 80 million customers, with capacity engineered to scale threefold—not demonstrated threefold usage. This points to transaction models becoming shared banking infrastructure, rather than isolated applications. https://blogs.nvidia.com/blog/european-financial-services-organizations-open-models-proprietary-intelligence/
NVIDIA also reports that bunq’s Finn matches leading closed-model quality at 60% of the cost per answer, while Stripe achieved 42x faster data pipelines and 92% cost savings in offline testing. These vendor-reported results suggest potential inference and training efficiencies, but Stripe’s offline scope does not establish live fraud-detection improvements. Benchmark baselines, quality evaluation methods and production generalizability remain unverified; NVIDIA’s account is not independent confirmation. https://blogs.nvidia.com/blog/european-financial-services-organizations-open-models-proprietary-intelligence/