Starling Builds UK-Sovereign AI Model With Nvidia and Civo

Starling is developing a transactional foundation model with Nvidia that will be trained and hosted entirely within a UK sovereign cloud, as the digital bank looks to deepen its use of artificial intelligence while reducing reliance on a small number of global infrastructure providers.

Starling Builds UK-Sovereign AI Model 

The model will be trained on more than 15 billion anonymised transactions from Starling’s own data estate, with the aim of creating a richer understanding of customer behaviour and financial patterns.

Starling plans to put the model into production in early 2027, initially to strengthen fraud detection before extending it into the bank’s existing AI assistant.

Over time, the model is intended to become the foundation of a broader intelligence layer known as Starling Intelligence.

Transactional Data Becomes the Training Advantage

The key distinction is that the model is being trained specifically on banking activity rather than on general-purpose datasets.

By analysing sequences and patterns across billions of transactions, Starling expects the model to support a range of applications, from identifying suspicious behaviour to producing more personalised financial guidance.

The bank says the intelligence layer could eventually underpin multiple AI services, including proactive budgeting support through Starling Assistant.

The model is being developed with Nvidia’s financial services team using its Transaction Foundation Model blueprint and associated accelerated computing tools.

Sovereign Cloud Addresses Concentration Risk

Just as significant is where the model will run.

Starling is working with UK cloud provider Civo to ensure that training and hosting remain within UK jurisdiction.

The approach reflects concerns about the financial sector’s dependence on a limited number of global cloud and AI providers.

Those risks were highlighted in the UK Government’s Financial Services AI Adoption Plan, co-authored by Starling group chief information officer Harriet Rees, which called for greater supplier diversity to strengthen resilience and improve optionality.

Rees said Starling wants to support the development of domestic cloud infrastructure while reducing concentration risk.

AI Strategy Moves Closer to Core Banking Infrastructure

The project shows how banks are beginning to treat AI less as a standalone productivity tool and more as a strategic infrastructure layer.

For Starling, the immediate applications are fraud and customer assistance, but the longer-term ambition is broader: one intelligence layer trained on proprietary transactional data and embedded across the bank’s operating model.

That creates potential advantages, but also raises the stakes around governance, resilience and data control.

The wider significance is that AI competition in banking may increasingly depend not only on model capability, but on who controls the data, the infrastructure and the jurisdiction in which that intelligence is built.