At GTC Berlin, held October 20–22, NVIDIA presented a shift that goes beyond a single product: European banks, fintechs, payment companies and financial data providers are putting open-weight models into production on their own infrastructure, rather than relying only on third-party closed systems.
The central example is Revolut. According to NVIDIA, the fintech is rebuilding its technology stack on a single foundation model, PRAGMA, with NVIDIA accelerated computing — using the cuDF toolkit and Dynamo-Triton software on Blackwell Ultra GPUs. The model is already in production serving an initial 80 million customers and is engineered to scale 3x.
“Revolut processes billions of transactions — the opportunity was always there, but the question was whether foundation models could operate at that scale and across domains. PRAGMA answers that,” said Pavel Nesterov, Revolut's head of AI, in a statement published by NVIDIA. The company presents the case as proof that the gap between a research breakthrough and a production banking system can be closed — Revolut's own assessment, not an independent measurement.
The move is not isolated. Starling Bank developed “Starling Intelligence” on UK sovereign cloud infrastructure, on Civo, using its own transaction data. Lloyds Banking Group built ATUM, a transaction foundation model that turns billions of transactions into reusable intelligence, accelerated by RTX PRO 6000 Blackwell GPUs. SBS launched a sovereign platform so banks can run models without their data leaving their own walls.
Other examples reinforce the pattern: neobank bunq's assistant Finn now runs on open-weight models with, according to NVIDIA, quality matching leading closed models at 60% of the cost per answer. NatWest is working on recognising British accents and dialects with NVIDIA speech models; Modulr is building specialised payment models with Nemotron-3-Ultra-550B; and Santander is an early adopter of the Kumo family for structured financial data. In fraud, Stripe cites, in offline testing, pipelines 42x faster and 92% cost savings with GPU-accelerated data.
The editorial reading is clear, but needs care: what NVIDIA describes is a bet on data sovereignty and cost control, not the end of closed models. For regulated institutions, keeping data inside their own infrastructure is the central argument — and GTC Berlin was used to show that this bet is already in production, not only in the lab.
Image credit: NVIDIA (official GTC Berlin blog)
Fontes
- NVIDIA Blog — European Financial Institutions Embrace Open Models
- NVIDIA Blog — GTC Berlin 2026: Live Updates
- CFOtech — Revolut builds AI transaction model for 80 million customers
Transparency: this article was produced with the assistance of artificial intelligence, cross-checked against public posts on X and the internet sources listed above. Consult the original sources to confirm the details.
Muse
