Santander has developed specialised banking AI models with NVIDIA, reducing serving costs in an experimental use case while maintaining performance.
Banco Santander has announced new progress in its AI collaboration with NVIDIA. With NVIDIA’s support, the Santander AI Lab, which explores and tests emerging AI technologies before wider deployment across the bank, has developed capabilities to evaluate AI models and agents for banking applications, identify the most suitable model for each task, and deploy it efficiently. As part of this work, Santander has been moving from large general-purpose models towards smaller models specialised for specific banking tasks. In selected experimental use cases, this approach has reduced serving costs from around EUR 5 per interaction to approximately EUR 0.03 while maintaining performance.
Model evaluation and specialised models
Through the Lab, Santander has built Model Arena, a tool for evaluating open-weight models against banking tasks and selecting the right model for each workload, and is specialising models using NVIDIA NeMo Customizer. Comparisons in customer service showed that smaller, specialised models can perform well on targeted banking tasks. Santander has also built specialised intelligence for core banking workloads on NVIDIA Kumo, a family of open-weight foundation models for structured data that helps turn structured data into the predictive insights needed for agentic services.
The Lab is industrialising inference within Santander’s own architecture using NVIDIA Dynamo-Triton, the approach behind the serving cost reduction. It is also extending Model Arena to design, simulate, evaluate, and supervise banking agents, alongside performance engineering and adversarial testing of conversational AI in controlled environments. In one experimental customer service use case, Santander improved conversation quality while cutting costs by more than 50% by deploying a more efficient AI model.
Three-year collaboration framework
For the next phase, Santander and NVIDIA have established a three-year collaboration framework covering foundation models, AI agents, fraud and anti-money laundering capabilities, real-time decisioning, and advanced simulation technologies, including a banking digital twin. Santander is progressing from open-weight models towards specialised models and governed banking agents, designed to provide greater control, safety, and lower operating costs at scale.
Iñaki Bernal, Chief Artificial Intelligence Officer at Santander, described NVIDIA as a key player in the infrastructure, software, and research behind current AI advances, adding that, combined with Santander’s scale across businesses, customers, and geographies, the bank aims to turn those advances into practical banking capabilities. Aser Blanco, Global Banking Lead at NVIDIA, stated that banks now need to turn their data into usable, proprietary intelligence, which depends on training and fine-tuning rather than model size. According to Blanco, cost per intelligence output has become a key metric, and Santander is demonstrating this with smaller models specialised on its own data and served efficiently at the scale of a global bank.
