Starling Bank will train a transactional foundation model on 15 billion anonymized transactions, running the workloads on UK sovereign cloud infrastructure from Civo and Nvidia hardware.

The challenger bank calls the model a TFM (transactional foundation model): a large AI model trained on proprietary payment and account data to detect patterns in customer behavior. Starling expects it to eventually form a "Starling Intelligence" layer that underpins new customer services.

By choosing a UK-based sovereign cloud — infrastructure where data storage and processing stay within national borders — Starling aligns itself with the UK Government's Financial Services AI Adoption Plan, which warns financial firms against depending on a small number of global providers.

What does the deal change for Starling?

Harriet Rees, group CIO at Starling and UK Government AI champion for financial services, framed the move as both a product and a supply-chain play. "We've made huge strides with our proprietary AI tools, launching the UK's first agentic AI financial assistant and rolling out new features every week. Now we want to connect these tools with a single intelligence layer that's trained entirely on the specific needs of our customers," she said.

She added: "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."

For Civo, the win validates its pitch that sovereignty and performance aren't mutually exclusive. "AI will only deliver lasting value for the UK economy if it's built on secure, resilient, sovereign infrastructure," said Mark Boost, CEO of Civo. "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."

Which Nvidia tools are involved?

The training stack isn't just GPUs. Starling will draw on several Nvidia components:

• Nvidia GPUs for the training compute itself.
• The Nvidia Transaction Foundation Model developer blueprint, a pre-built reference architecture for transaction models.
• Nvidia CUDA-X libraries for GPU-accelerated data processing.
• Nvidia NeMo AutoModel to manage model training.

The blueprint approach shortens the path from raw transaction data to a working model, since Starling doesn't build the training pipeline from scratch.

Civo's edge buildout looms behind the deal

The announcement lands weeks after Civo unveiled plans for 40 edge data centers across the UK, starting with an 8MW facility in Hertfordshire. The company said those sites will power sovereign AI compute in the UK, including deployments of Nvidia Vera Rubin NVL72 systems — the rack-scale platform Nvidia positions for large-scale AI inference and training.

That infrastructure matters here. A TFM trained on 15 billion transactions is a serious compute workload, and Civo's sovereign capacity gives Starling room to scale training and hosting without moving data outside UK borders.

For the wider UK banking sector, the deal is a reference point. Regulators and government policy have pushed financial institutions to weigh concentration risk in cloud contracts. Starling is now the visible test case for a mid-sized bank running foundation-model training on domestic infrastructure — with Nvidia silicon inside a homegrown cloud, rather than a hyperscaler's.

Whether other banks follow depends on execution: training quality, time to deployment for the intelligence layer, and whether the "Starling Intelligence" services deliver measurable customer outcomes. The groundwork, at least, is now public.