Starling Bank intends to construct what it claims will be the UK's inaugural banking artificial intelligence model, trained and operated on domestic sovereign cloud infrastructure. The initiative aims to lessen the sector's dependence on major international technology corporations. Working alongside NVIDIA and domestic cloud operator Civo, Starling is building a transactional foundation model (TFM) powered by data from over 15 billion anonymised transactions.
Unveiled on 14 October, the undertaking targets fraud detection and enhanced personalised banking experiences, with rollout to live systems anticipated in early 2027. The move also reflects broader apprehension regarding supplier concentration in the financial sector, where reliance on just a handful of cloud and AI vendors could create systemic vulnerability.
Starling Develops AI Model on UK Sovereign Cloud
Starling's proprietary transaction records will serve as the training foundation for the model, enabling it to detect behavioural patterns in customer finances. Distinct from general-purpose language models, this TFM addresses financial transactions specifically. Starling plans to deploy it as the cornerstone of a broader AI initiative termed Starling Intelligence.
Initially, the model will enhance fraud prevention capabilities before being embedded into Starling Assistant, the bank's agentic AI financial advisor, to deliver more tailored financial recommendations. NVIDIA's Transaction Foundation Model blueprint will guide development, incorporating its GPU-driven data handling and training apparatus. Civo will furnish the training infrastructure, deploying NVIDIA GPUs within UK territory.
Harriet Rees, Group CIO at Starling and designated UK Government AI Champion for Financial Services, emphasised the project's contribution to strengthening domestic infrastructure. "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," Rees stated.
Sovereign Cloud Investment Addresses Concentration Risk
The infrastructure strategy aligns with guidance from the UK Government's Financial Services AI Adoption Plan, which Rees co-authored and released in July 2026. The document flagged the financial sector's vulnerability stemming from its heavy reliance on a limited pool of international cloud and AI suppliers as a potential operational hazard.
The plan advocates for broadening vendor relationships and cultivating in-house AI development to grant organisations greater optionality and shield them from disruptions at major suppliers. Regulatory bodies have taken parallel steps in response to these risks.
In July, the Bank of England, Prudential Regulation Authority and Financial Conduct Authority commenced oversight of four critical third-party technology suppliers—Amazon Web Services, Google Cloud, Microsoft and Oracle. This framework targets the potential fallout from service interruptions at technology vendors whose infrastructure supports significant portions of the UK financial system.
For Starling, partnering with a domestic cloud operator expands its infrastructure alternatives. Nevertheless, UK-based hosting alone does not guarantee full technological autonomy, particularly given the model's continued dependence on NVIDIA hardware and software.
Mark Boost, CEO of Civo, remarked: "AI will only deliver lasting value for the UK economy if it's built on secure, resilient, sovereign infrastructure. Starling is leading the financial sector by training and hosting its Transactional Foundation Model on a homegrown cloud. Pairing NVIDIA compute with Civo's UK sovereign cloud shows that banks don't have to trade performance or innovation for control of their data, or depend on a handful of global providers."
Production Deployment Planned for 2027
Starling anticipates moving the transactional foundation model into production during early 2027. The initial phase will concentrate on strengthening fraud detection, with subsequent expansion into Starling Assistant's feature set. Since the project remains under development, tangible results regarding fraud mitigation and customer experience improvements remain unproven.
From a cloud and data centre perspective, the infrastructure choice carries broader implications. Starling is opting to engineer a bespoke AI model using UK-resident compute while maintaining reliance on NVIDIA's platform. The initiative illustrates a potential pathway for financial institutions to broaden their infrastructure vendor base as AI becomes increasingly embedded in their operations. Whether this strategy achieves its intended outcomes will hinge on the model's real-world performance and Starling's capacity to manage it reliably at production scale.
