Starling to train banking AI model on UK sovereign cloud

Starling is building a transactional foundation model with NVIDIA on Civo’s UK cloud infrastructure, becoming the first bank to train and host such a model in a UK sovereign cloud. The project is designed to strengthen fraud detection, improve customer experiences and reduce reliance on a small number of global cloud providers.

Why it matters: - Starling’s project ties advanced banking AI to UK-based infrastructure, a move aimed at improving control over data and reducing supplier concentration risk. - The model is meant to become a shared intelligence layer for future Starling products, which could sharpen fraud detection and personalize customer services. - The effort aligns with UK government guidance that urges broader cloud and AI partnerships to improve resilience.

What happened: - Starling is developing a transactional foundation model with NVIDIA support. - The model will be trained and hosted on Civo’s UK sovereign cloud infrastructure. - Starling says the project makes it the first bank to train a banking model inside a UK sovereign cloud. - Starling intends to move the model into production in early 2027.

The details: - The model will use proprietary Starling data from more than 15 billion anonymised transactions. - Starling plans to turn the model into a single intelligence layer called Starling Intelligence. - Starling says that layer will support new AI experiences across the bank. - The model is being built with NVIDIA’s financial services team. - The work uses NVIDIA’s Transaction Foundation Model developer blueprint. - NVIDIA CUDA-X libraries will handle GPU-accelerated data processing. - NVIDIA NeMo AutoModel will be used for model training. - Civo’s infrastructure includes NVIDIA Hopper GPUs and planned Vera Rubin deployments. - Civo says its platform provides low-latency compute and advanced liquid cooling for real-time reasoning, autonomous systems and agentic AI. - The reusable intelligence layer is designed to identify patterns in anonymised transaction data and support multiple applications. - Starling plans to use the model first to improve fraud detection. - Starling then wants to use the model to augment Starling Assistant, the UK’s first agentic AI financial assistant. - Starling says the intelligence layer will help Starling Assistant deliver more personalised features, including proactive budgeting guidance.

Between the lines: - The project reflects a wider push by UK financial firms to keep sensitive AI workloads on domestic or sovereign infrastructure. - Harriet Rees, Starling’s Group CIO and UK Government AI Champion for Financial Services, co-authored the Financial Services AI Adoption Plan published in July 2026. - That plan warned about dependence on a small number of global AI and cloud providers and recommended broader partnerships. - Civo and NVIDIA are positioning sovereign infrastructure and accelerated computing as compatible with performance and innovation, not a trade-off.

What’s next: - Starling will continue training the model on its anonymised transaction data. - The bank expects to launch production use in early 2027. - Early use will center on fraud detection before the model expands into more customer-facing AI features. - The Starling Intelligence layer is expected to become the foundation for new applications across the bank.

The bottom line: - Starling is betting that sovereign cloud infrastructure can power frontier banking AI while keeping tighter control over data and supplier risk.