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

NVIDIA used GTC Berlin to present a cluster of European financial services organizations running AI systems in production on its accelerated computing, on their own infrastructure, and largely on open weight models rather than closed APIs. The announcements span banks, payment companies and financial data providers, and cover three broad areas: model training and deployment, cost-efficient open model adoption, and GPU-accelerated data pipelines for research and fraud detection.
Revolut is the headline example. The neobank says it is rebuilding its technology stack on a single foundation model called PRAGMA, built in collaboration with NVIDIA. The system runs on NVIDIA cuDF, a GPU dataframe toolkit, and Dynamo-Triton inference software on NVIDIA Blackwell Ultra GPUs. Revolut describes PRAGMA as serving an initial 80 million customers in production and as engineered to scale three times that. Pavel Nesterov, head of AI at Revolut, framed the question the work answered as whether foundation models could operate at that scale and across domains, and said PRAGMA shows the gap between a research breakthrough and a production banking system can be closed.
On the training side, Starling Bank says it is the first bank to train an agentic model on U.K. sovereign cloud infrastructure. Its intelligence layer, called Starling Intelligence, was developed on Civo's U.K. cloud using the bank's own transaction data. Lloyds Banking Group built a transaction foundation model named ATUM that turns billions of banking transactions into reusable intelligence, accelerated with NVIDIA RTX PRO 6000 Blackwell GPUs and aimed at improving efficiency across bank-wide use cases. SBS launched a banking sovereign intelligence platform intended to let banks run generative and agentic models without data leaving their own walls.
Several firms have moved to open weight models to cut cost while keeping data and answers proprietary. Neobank bunq says its agentic assistant Finn now runs on open weight models, powered by NVIDIA accelerated computing and deployed on Nebius, matching the quality of leading closed models at 60% of the cost per answer. NatWest is building speech technology aimed at recognizing the full range of British accents, dialects and financial terminology, developed with NVIDIA speech models including Nemotron 3.5 ASR. Modulr is building specialized models for distinct parts of payment operations based on the NVIDIA Nemotron-3-Ultra-550B open model. Santander is an early adopter of NVIDIA's new Kumo family of open weight foundation models for structured financial data, using them for core banking workloads.
For financial research, LSEG is working with NVIDIA to help institutions analyze large volumes of historical tick data up to 10x faster using the Polars GPU engine accelerated by cuDF on RTX PRO 6000 Blackwell GPUs. RavenPack built a dataset generation pipeline on Bigdata.com that takes a plain-language research hypothesis and returns a labeled, backtest-ready dataset, running on a fine-tuned NVIDIA Nemotron model; investors can use the output to identify, analyze and visualize market trends.
On the data-pipeline and fraud side, NVIDIA positions RTX PRO 6000 Blackwell Server Edition GPUs and cuDF as a way to accelerate processing across multi-terabyte, siloed and regulated sources, compressing the time from data to decision. HSBC provides near-real-time client insights supporting on-demand decision-making while minimizing risk and fraud, powered by 5x faster data processing on Starburst Enterprise's GPU Engine. Stripe reports 42x faster data pipelines and 92% cost savings for its Stripe Radar fraud-detection platform with NVIDIA GPU-accelerated Ray Data in offline testing, which it says lets Radar's machine learning pipelines continuously optimize training and improve detection.
My reading is that this collection is as much a positioning statement as a product announcement. For a developer, the technical thread is that the same GPU stack — cuDF for dataframe work, Ray Data for pipeline throughput, Triton/Dynamo for serving, and open Nemotron-family models for the model layer — is being reused across training, inference, research and fraud. That reuse is what makes the claimed speedups plausible in aggregate: regulated firms usually struggle less with model choice than with moving large, siloed datasets through compliance and preparation steps. A likely trade-off is that open weight models shift work onto the adopting firm: fine-tuning, evaluation and serving infrastructure become internal responsibilities rather than a vendor's, which is precisely where the sovereign-cloud framing from Starling and SBS fits. It is also worth noting that Stripe's figures come from offline testing, and the other numbers are vendor-attributed, so independent production benchmarks would be the next thing to watch. The usual caution applies to comparisons against closed models: quality parity at lower cost per answer depends heavily on task mix, and 60% of a closed model's cost is a claim about bunq's specific workload.
Why it matters: European banks and payment firms are signaling that production AI can run on open weight models inside their own infrastructure, which affects teams choosing between closed APIs and self-hosted stacks. For developers, that means demand for GPU-data-pipeline skills (cuDF, Ray Data), inference serving (Triton) and model customization is likely to rise, and vendors selling into finance will face pressure to show comparable cost and latency numbers. Whether these gains hold outside vendor testing is, in my inference, the question that will decide how quickly the rest of the sector follows.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
At GTC Berlin, banks, payment companies and financial data providers are showing what production-scale intelligence looks like — built on NVIDIA accelerated computing, powered by open weight models and running on their own infrastructure. Revolut is rebuilding its technology stack on a single foundation model PRAGMA, powered by NVIDIA accelerated computing. Using the NVIDIA cuDF […]