Starling Bank Taps Nvidia and UK Sovereign Cloud Provider Civo to Build Proprietary Transactional AI Model
Starling Bank Taps Nvidia and UK Sovereign Cloud Provider Civo to Build Proprietary Transactional AI Model #
Starling Bank announced on 14 October 2026 that it is developing a Transactional Foundation Model (TFM) in partnership with Nvidia, to be trained and hosted exclusively on infrastructure operated by Civo, a UK-owned and UK-headquartered cloud provider. The bank says the project makes it the first financial institution to train and host such a model within a UK sovereign cloud.
The TFM will draw on more than 15 billion anonymised transactions from Starling’s own systems to identify patterns in customer financial behaviour. The bank calls the resulting model “Starling Intelligence,” describing it as a shared AI layer beneath its entire suite of AI-powered products, intended to enable new customer experiences over time.
The choice of Civo over a major hyperscaler reflects UK government concerns about cloud concentration risk. Harriet Rees, Starling’s Group Chief Information Officer and a UK Government AI Champion, co-authored the UK Government’s Financial Services AI Adoption Plan, published in July 2026. That plan flagged growing concentration risk from the financial sector’s dependence on a small number of global cloud and AI model providers, and called on government and industry to broaden their partnerships in the interest of systemic resilience. Using Civo keeps all training data within UK jurisdiction.
Nvidia has partnered with several major UK institutions on similar foundation-model initiatives. Starling’s use of a sovereign cloud for training distinguishes its approach from those comparable efforts.
Starling said Starling Intelligence is intended to underpin tools including fraud detection, scam prevention, and personalised spending insights, rather than function as a standalone product. The bank already offers AI-powered features, including a scam-detection tool and an agentic business banking assistant, and the TFM is designed to provide a common, proprietary backbone for those capabilities. Production deployment is targeted for the first half of 2027.
Copy link