Santander Nvidia Partnership Builds Model Evaluation Tools for Banking AI

Banco Santander announced on October 19, 2026, a new phase in its collaboration with NVIDIA to develop evaluation tools for AI models and agents used in banking. The work centers on the Santander AI Lab, which tests emerging technologies before wider deployment across the bank.

Santander AI Lab Introduces Model Arena for Banking Tasks

The Santander AI Lab created the Model Arena platform to compare open-weight models on specific banking workloads. Teams run evaluations to identify the strongest model for each task and then specialize it with NVIDIA NeMo Customizer. The approach moves away from large general-purpose models toward smaller systems tuned on bank data.

Model Arena now extends to agent design, simulation, evaluation, and supervision. It incorporates performance engineering and adversarial testing to measure reliability under real conditions. NVIDIA Kumo foundation models support structured data workloads and enable predictive insights for agentic services.

Inference runs on Santander infrastructure through NVIDIA Dynamo-Triton. Early tests showed serving costs falling from roughly five euros per interaction to three cents while preserving output quality. One customer-service application recorded more than 50 percent lower costs alongside higher conversation quality scores.

Three-Year Collaboration Framework Covers Core Banking Areas

Santander and NVIDIA established a three-year framework that includes foundation models, AI agents, fraud and AML detection, real-time decisioning, and advanced simulation. The simulation work encompasses a banking digital twin for scenario testing. Santander chief AI officer Iñaki Bernal noted NVIDIA supplies the combined infrastructure, software, and research while Santander provides scale for testing.

NVIDIA global banking lead Aser Blanco stressed the focus on cost per intelligence output achieved by fine-tuning smaller models on proprietary bank data. The partnership aligns with Santander targets of more than one billion euros in combined AI-driven revenue and cost savings between 2026 and 2028. The bank already recorded 35 million euros in AI value during the first quarter of 2026 and projects over 200 million euros for the full year.

Previous AI deployments delivered measurable gains in customer service and operations. The current effort builds on those results by industrializing model selection and deployment processes across additional business lines.

Next Steps Focus on Bank-Wide Scaling

Santander plans to extend proven Model Arena capabilities and agent supervision tools across the organization. Additional initiatives under the three-year framework will advance fraud and AML detection, real-time decision systems, and the digital twin project. The bank continues to prioritize smaller, specialized models that deliver required performance at lower operating cost.

These developments occur as Santander pursues a group profit target above 20 billion euros by 2028. The AI program forms one component of that broader financial plan.