Santander and Nvidia develop AI capabilities for banking

Santander announced the work on Oct. 19, describing it as part of the next phase of its collaboration with Nvidia.[1][2] The companies are developing capabilities to evaluate AI models and agents for banking, with the aim of helping Santander choose a model that fits a task and deploy it efficiently.[2] Santander has also established a three-year framework for the collaboration.[1] The available information does not specify a rollout schedule, product pricing or which banking teams will use the capabilities first. The announcement identifies NVIDIA AI infrastructure as part of the effort.[1]

Santander says it is moving away from relying on large general-purpose models and toward smaller models specialized for specific banking tasks.[2] The bank says that, in selected experiments, this approach reduced serving costs from €5 to about €0.03 per interaction while maintaining strong performance.[2] It also reports that deploying a more efficient AI model improved conversation quality while lowering costs by more than 50%.[2] These are Santander-reported results; the evidence does not provide details on the experiments’ tasks, measurement periods or comparison baselines.[2]

Santander constructed a Model Arena to evaluate open-weight models against banking tasks and determine the right model for each workload.[2] It is using Nvidia NeMo Customizer to specialize models.[2] Blanco described Santander’s approach as using smaller models specialized on the bank’s own data, tested against real banking tasks and served efficiently enough for a global bank’s scale.[2] The approach makes task-specific evaluation and operating cost central considerations for model selection.[2] The evidence does not specify which models were tested or identify the workloads used in the Model Arena.[2]

The collaboration’s stated areas include foundation models, AI agents, fraud and anti-money-laundering capabilities, real-time decisioning, and advanced simulation technologies such as banking digital twins.[2] Blanco said cost per intelligence output is a key metric in the new AI environment.[2] PYMNTS Intelligence found that 95% of surveyed large financial-services firms had widely deployed or embedded newer AI tools in data and technology processes.[2] Its research also cited 25% adoption for AI in API orchestration and integration, while 30% of financial-services companies identified data quality and fragmentation as their main obstacle to further AI deployment.[2]