Starling Bank to train UK-sovereign transactional foundation model on Civo cloud with NVIDIA
Starling is building a transactional foundation model (TFM) with NVIDIA support, which it says is the first banking model trained and hosted in a UK sovereign cloud. The model will run on infrastructure from UK-owned provider Civo and be trained on more than 15 billion anonymised transactions. Starling plans to put it into production in early 2027, first to improve fraud detection and then to extend its Starling Assistant agentic AI financial assistant. The article presents this as product and business news with no security incident, vulnerability or threat actor involved. Starling ties the approach to the UK Government's Financial Services AI Adoption Plan (July 2026), which flags reliance on a few global AI and cloud providers as a concentration risk.
Technical Details
- Built on NVIDIA's Transaction Foundation Model developer blueprint, using CUDA-X libraries for GPU-accelerated data processing and NeMo AutoModel for training.
- Civo's sovereign infrastructure uses NVIDIA Hopper GPUs, with Vera Rubin deployments planned.
- All data stays within UK jurisdiction, on cloud infrastructure owned by a UK-headquartered company.
Linha do tempo
-
Starling Bank to train UK-sovereign transactional foundation model on Civo cloud with NVIDIA
ler a análise desta fonte ocultar
Starling is building a transactional foundation model (TFM) with NVIDIA support, which it says is the first banking model trained and hosted in a UK sovereign cloud. The model will run on infrastructure from UK-owned provider Civo and be trained on more than 15 billion anonymised transactions. Starling plans to put it into production in early 2027, first to improve fraud detection and then to extend its Starling Assistant agentic AI financial assistant. The article presents this as product and business news with no security incident, vulnerability or threat actor involved. Starling ties the approach to the UK Government's Financial Services AI Adoption Plan (July 2026), which flags reliance on a few global AI and cloud providers as a concentration risk.
Technical Details
- Built on NVIDIA's Transaction Foundation Model developer blueprint, using CUDA-X libraries for GPU-accelerated data processing and NeMo AutoModel for training.
- Civo's sovereign infrastructure uses NVIDIA Hopper GPUs, with Vera Rubin deployments planned.
- All data stays within UK jurisdiction, on cloud infrastructure owned by a UK-headquartered company.