European banks adopt NVIDIA-powered open models: Stripe cuts fraud pipeline costs 92%
NVIDIA Blog · rss · 2026-10-22
At GTC Berlin, NVIDIA detailed how European financial institutions build proprietary intelligence on open-weight models running on its accelerated computing:
- Revolut is rebuilding its stack on a single foundation model, PRAGMA, using cuDF and Dynamo-Triton on Blackwell Ultra GPUs, serving 80 million customers and engineered to scale 3x.
- bunq's assistant Finn now runs open-weight models on Nebius, matching closed-model quality at 60% of the cost per answer.
- Lloyds built the ATUM transaction foundation model; Starling Bank trained an agentic model on UK sovereign cloud; Santander adopted NVIDIA Kumo models for structured financial data.
- Data and fraud: Stripe Radar saw 42x faster data pipelines and 92% cost savings in offline tests with GPU-accelerated Ray Data; HSBC got 5x faster processing; LSEG analyzes tick data up to 10x faster with the Polars GPU engine.
More from Infra
- Locals: Open-Source Apple Silicon Engine Runs Local LLMs Up to 3x Faster — SumitGup · 2026-10-22
- DeepSeek's new funding draws hardware-heavy investors: CATL, SMIC, Geely among backers — teortaxesTex · 2026-10-22
- Open models now over 20% of tokens: deep dive on whether open source hurts GPU demand — dotey · 2026-10-22
- Samsung upgrades Xian plant to 286-layer NAND, SK Hynix seeks 240 layers in Dalian — rohanpaul_ai · 2026-10-22
- SGLang-TorchTPU launches with Google and Meta, hitting 770 tok/s decode on TPU — BanghuaZ · 2026-10-22
- RL environment task prices have gone bananas, says insider on the data market — jia_seed · 2026-10-22