Starling Bank to Develop UK-Based AI Banking Model Using 15 Billion Transactions

Digital bank partners with NVIDIA and British cloud provider Civo to develop new artificial intelligence technology for fraud prevention, personalised financial services and greater control over customer data.

British digital bank Starling has announced plans to develop an artificial intelligence banking model trained on more than 15 billion anonymised financial transactions, in a move designed to strengthen fraud detection, improve customer services and reduce dependence on overseas technology providers.

The project, announced on 14 October 2026, will see Starling work with technology company NVIDIA and UK cloud infrastructure provider Civo to create what it describes as the first banking transactional foundation model to be trained and hosted on a UK sovereign cloud.

The initiative represents a significant investment in domestic financial technology infrastructure and reflects growing concerns about the banking industry’s dependence on a relatively small number of international cloud computing companies.

Starling says the new system will form the foundation of its future artificial intelligence capabilities, with initial applications expected to enter production in early 2027.

 

New AI System to Analyse Banking Transactions

At the centre of the project is a Transactional Foundation Model (TFM), a specialised artificial intelligence system designed to identify patterns and relationships within financial transaction data.

Unlike general-purpose AI models trained on large collections of text and publicly available information, the system will focus on banking activity.

Starling intends to train the model using more than 15 billion anonymised transactions from its own systems.

The bank says this will allow it to develop a deeper understanding of customer behaviour and create more sophisticated financial services.

The technology will underpin a new intelligence platform called Starling Intelligence, which is intended to connect the bank’s existing AI applications and support the development of additional services.

Potential uses include identifying suspicious financial activity, improving fraud prevention and providing customers with more personalised financial information.

Fraud Detection Among First Planned Applications

Starling expects to introduce the transactional model into production in early 2027, initially concentrating on improvements to fraud detection.

Financial institutions already use automated systems to identify unusual payments and potentially fraudulent transactions.

However, advances in artificial intelligence offer opportunities to analyse more complex behavioural patterns and identify suspicious activity that might be missed by conventional systems.

A model trained specifically on banking transactions could potentially distinguish more effectively between legitimate customer behaviour and transactions requiring further investigation.

Starling also plans to use the technology to enhance Starling Assistant, its existing AI-powered financial assistant.

Future developments could include more personalised budgeting recommendations and proactive financial guidance based on customer spending patterns.

These applications remain planned developments rather than services already delivered by the new model.

Starling Seeks Greater Control Over AI Infrastructure

A central feature of the project is the decision to train and host the model within the UK.

Starling will use infrastructure operated by Civo, a British cloud and AI computing company, alongside NVIDIA’s technology for processing data and training artificial intelligence models.

The bank says the arrangement will allow the model and associated data to remain within UK jurisdiction.

This approach addresses growing concerns about concentration risk in financial services, where a relatively small number of major international technology providers supply much of the industry’s cloud computing infrastructure.

Dependence on a limited group of suppliers can create operational risks if technical failures, cyber incidents or other disruptions affect multiple financial institutions simultaneously.

Harriet Rees, Group Chief Information Officer at Starling and a UK Government AI Champion for Financial Services, said:

“We also want to take a leadership role in the development of home-grown cloud infrastructure, which is essential to European companies’ ability to diversify their suppliers and reduce concentration risk.”

The decision follows recommendations in the Government’s Financial Services AI Adoption Plan, published in July 2026, which identified the importance of improving resilience and reducing excessive dependence on a small number of technology suppliers.

NVIDIA Technology to Support Development

The project will use NVIDIA’s financial services technology and software tools to develop the transactional model.

These include NVIDIA’s Transaction Foundation Model developer blueprint, CUDA-X software libraries and NeMo AutoModel training technology.

The tools are intended to support the processing of large datasets and the development of specialised AI models.

Civo will provide the computing infrastructure, including high-performance graphics processing hardware and systems designed to support demanding AI workloads.

Mark Boost, Chief Executive of Civo, said:

“AI will only deliver lasting value for the UK economy if it’s built on secure, resilient, sovereign infrastructure.”

The companies believe the project could demonstrate how banks can develop advanced artificial intelligence capabilities while maintaining greater control over their infrastructure and data.

What Does This Mean for UK Banking Customers?

For consumers, the immediate impact is likely to be limited because the new model is not expected to enter production until 2027.

However, the longer-term implications could be significant.

More sophisticated transaction analysis may allow banks to identify potential fraud earlier, develop more relevant financial products and offer personalised services based on individual customer behaviour.

It could also help financial institutions develop systems that respond more effectively to unusual account activity.

Nevertheless, the use of large-scale transaction data in artificial intelligence raises important questions about privacy, transparency and governance.

Although Starling says the model will use anonymised data, banks developing AI applications must continue to meet applicable data protection, operational resilience and financial regulatory requirements.

The effectiveness of the new system will also depend on the quality of its training data, testing and ongoing supervision.

A Potential Boost for Britain’s Financial Technology Industry

Starling’s decision to work with a UK-based cloud provider could have wider implications for Britain’s financial technology sector.

The UK has a substantial fintech industry, but many financial services businesses rely on technology infrastructure supplied by multinational companies.

Developing more domestic computing capacity could provide additional options for banks and other regulated businesses seeking greater control over where their information is stored and processed.

The Starling project also demonstrates how financial institutions are moving beyond general-purpose AI chatbots towards specialised models developed for particular business functions.

If successful, the technology could provide a foundation for further innovation in fraud prevention, financial management and customer services.

For now, the initiative represents a significant commitment to AI development rather than proof of improved financial outcomes.

With production planned for early 2027, the next stage will be demonstrating whether the model can deliver measurable improvements in banking security and customer experience.

Source and further reading:

Starling Bank – Starling Becomes First Bank to Train AI Model on UK Sovereign Cloud, 14 October 2026

Image: ChatGPT