NVIDIA Accelerated Computing Powers Next-Gen AI for Revolut, Starling, and Lloyds

Major financial institutions including Revolut, Lloyds Banking Group, and Starling Bank are moving generative AI into production-scale operations using NVIDIA’s accelerated computing platform. By deploying proprietary foundation models on private infrastructure, these firms are bypassing the limitations of closed-model providers to achieve significant gains in transaction processing, fraud detection, and operational efficiency.

What was announced

Revolut is rebuilding its technology stack on PRAGMA, a foundation model powered by NVIDIA Blackwell Ultra GPUs and the cuDF toolkit. The system is currently servicing 80 million customers and is engineered to scale up to three times that capacity. Lloyds Banking Group has introduced ATUM, a transaction foundation model that converts billions of banking transactions into reusable intelligence, accelerated by NVIDIA RTX PRO 6000 Blackwell GPUs to improve efficiency across bank-wide use cases.

Starling Bank has developed "Starling Intelligence," an agentic model layer trained on Civo’s U.K. sovereign cloud infrastructure using the bank’s own transaction data. In the payments sector, Stripe reported 42x faster data pipelines and 92% cost savings for its Stripe Radar fraud-detection platform during offline testing with NVIDIA GPU-accelerated Ray Data. Neobank bunq has transitioned its AI assistant, Finn, to open weight models running on Nebius, matching closed-model quality at 60% of the cost per answer.

Other implementations include NatWest’s use of Nemotron 3.5 ASR to recognize diverse British accents and financial terminology, and LSEG’s work with the Polars GPU engine to analyze historical tick data 10x faster. Santander is also adopting the NVIDIA Kumo family of foundation models for core banking workloads, while SBS has launched a banking sovereign intelligence platform to allow institutions to run models without data leaving their own infrastructure.

The companies involved

Revolut, headquartered in London, has grown from a digital wallet into a global financial super-app, maintaining a massive presence in the fintech market with over 250 reports in our archives. Lloyds Banking Group remains one of the UK’s "Big Four" clearing banks, managing a vast legacy infrastructure that it is now modernizing through foundation models like ATUM. Starling Bank, a leading UK challenger bank, has focused heavily on proprietary technology and was the first to train agentic models on sovereign cloud infrastructure.

NVIDIA, the hardware giant behind the Blackwell architecture, has become the primary infrastructure provider for the financial sector's pivot toward "sovereign intelligence." These companies represent a cross-section of the market, from agile neobanks to established incumbents, all converging on a similar hardware-accelerated path. The scale of these operations is reflected in their market footprint, with Revolut and Lloyds being among the most frequently covered entities in the fintech space, reflecting their influence on industry standards.

What FF News has reported before

FF News has closely followed the intersection of AI and banking infrastructure. We recently reported on how Starling Bank Unveils AI Agent Capable of 17-Second Financial Crime Investigations, showcasing the speed of their new intelligence layer. Our coverage of Santander Slashes AI Costs by 99% Using Specialized NVIDIA Infrastructure highlights the economic drivers behind this hardware shift. Additionally, we have tracked Revolut’s broader strategic moves, including its stance on Apple NFC Access and its expansion into the telecom sector with Revolut Mobile in Germany.

What this means

The shift toward open-weight models and "sovereign intelligence" marks the end of the experimental phase for banking AI. By running models on their own infrastructure, banks are solving the dual problems of data residency and the "black box" nature of third-party LLMs. This move puts immense pressure on traditional software-as-a-service providers who cannot match the cost-efficiency of in-house foundation models. The real question for the sector is no longer about AI capability, but about the availability of specialized hardware like Blackwell GPUs. As institutions like Stripe and bunq prove massive cost savings, the competitive gap will widen between those owning their intelligence stack and those renting it.