# Santander and NVIDIA to develop banking-specialized AI

> Santander and NVIDIA agree three-year collaboration to develop banking-specialized AI, improve services and cut costs
> By Covadonga Fernandez · Technology · 2026-10-19

Santander has sealed a three-year agreement with NVIDIA, one of the technology giants driving global artificial intelligence development, to advance its application to the banking business. The collaboration spans everything from customer service to fraud and money laundering prevention, and includes the development of agents capable of performing tasks and tools for making real-time decisions.

The agreement expands the work both companies already carry out at the Santander AI Lab, the laboratory where the bank tests new technologies before rolling them out across the group. The institution aims to turn NVIDIA's capabilities into concrete banking services, with models tailored to its needs and costs that allow them to be used at scale.

The first tests are already delivering results, according to the statement published by Santander. In an experimental customer service case, the bank improved the quality of conversations and cut costs by more than 50%. In other experimental uses, the execution cost has fallen from around 5 euros per interaction to around 0.03 euros, while maintaining good results. These figures correspond to specific tests and not to its services as a whole.

## Specialized models for banking

One of the main lines of work consists of moving from large general-purpose AI models to smaller ones specialized in banking tasks. To this end, Santander has created Model Arena, a platform that compares models and makes it possible to determine which is most suitable and efficient for each use.

The bank works with open-weight models, whose trained parameters can be used and adapted under the terms of their licenses. With NVIDIA tools, it fine-tunes them for specific functions and checks their performance before integrating them into its systems.

This strategy places the cost of each task resolved among the criteria for choosing the technology. At an institution that needs to handle numerous operations, the usefulness of a model also depends on whether it can run with the efficiency needed to extend it to different businesses and countries.

Santander also uses NVIDIA Kumo, a family of models for structured data, such as data organized in tables and records. The goal is to turn that data into predictive information that can be used in the bank's core processes and in services with AI agents.

## Agents and simulations before deployment

The collaboration includes the development of [banking agents](https://www.santander.com/es/sala-de-comunicacion/notas-de-prensa/2026/10/santander-desarrolla-con-nvidia-capacidades-de-ia-que-mejoran-la-calidad-del-servicio-y-reducen-los-costes), systems capable of chaining actions to complete a task. Santander is expanding Model Arena to design them, simulate their behavior, evaluate their results and monitor their operation.

The laboratory also puts conversational AI through stress tests in controlled environments. The bank seeks to verify how it responds before extending its use, with oversight and security mechanisms. The three-year framework envisages advanced simulation technologies, such as a banking digital twin. This virtual representation would make it possible to test processes and study their behavior before introducing changes to real systems.

Iñaki Bernal, Santander's head of Artificial Intelligence, stresses that NVIDIA contributes infrastructure, software and research. The bank's contribution is its scale of customers, businesses and geographic presence to turn that technology into banking applications.

For Aser Blanco, NVIDIA's global head of Banking, progress depends on training and adapting models to the tasks they must solve. Santander, he argues, is working with smaller models, specialized with the bank's own data and evaluated in real banking situations.
