The Pulse
Santander and Nvidia Set Three-Year AI Plan for the Bank
Santander and Nvidia have agreed to a three-year collaboration covering banking AI, including models, agents, fraud prevention and simulation. Early experiments reported by the companies suggest smaller, specialized models can reduce servin
AI.info Team · Written with AI · fact-checked ·
Santander’s AI rollout starts with smaller models
Santander and Nvidia have agreed to a three-year collaboration framework to develop and expand AI capabilities for banking. Announced on October 19, 2026, the work spans foundation models, AI agents, fraud and money-laundering prevention, real-time decisions and simulation technologies, including banking digital twins. The effort builds on work at Santander AI Lab, where the companies have developed capabilities to evaluate models and agents for banking tasks.
The lab’s Model Arena evaluates open-weight models against banking tasks to help identify which model fits each workload. Santander is also adapting models with Nvidia NeMo Customizer. Other developments include specialized intelligence for core banking processes using Nvidia Kumo, a family of models for structured data, and the use of Nvidia Dynamo-Triton to bring capabilities into the bank’s technology infrastructure. Santander is moving from large general-purpose models toward smaller models designed for specific banking tasks.
Model Arena is also being used to design, simulate, evaluate and supervise banking agents. Santander says it is testing conversational AI in controlled settings to assess and improve results before any wider deployment across the group. The framework identifies areas for continued work; it does not mean that every proposed system is already deployed throughout the bank.
Tests put the cost question in numbers
In selected experiments, serving costs fell from about €5 per interaction to approximately €0.03 as the bank moved toward smaller, specialized models, according to the announcement. The companies also described an experimental customer-service use case in which conversation quality improved while costs fell by more than 50%. Those figures refer to tests, not a bank-wide result, and the announcement does not specify how many interactions or services the cost comparison covers.
Nvidia’s global banking lead, Aser Blanco, described the approach in a statement included with the announcement:
“Santander is showing what that looks like in production with smaller models, specialized on the bank’s own data, evaluated against real banking tasks and served efficiently enough to run at the scale of a global bank.”
Blanco’s statement focuses on task-specific models and the cost of running them. The experiments offer examples of possible efficiency gains, but they do not establish how much the partnership will save across Santander or how many services will use the tools.
Santander’s separate AI value target
The Nvidia framework forms part of Santander’s broader AI work, but the bank has not assigned it a separate financial target. In a June 22, 2026 update, Santander set a goal of generating more than €1 billion in total business value from AI between 2026 and 2028, combining additional revenue and cost reductions. The update reported €35 million in AI business value during the first quarter and said the bank was on track to exceed €200 million for the full year.
The June figures describe Santander’s overall AI target and reported results, not the expected contribution of the Nvidia collaboration. Neither the framework nor the experimental cost examples specify how much business value will come from Nvidia’s tools. The work described so far ranges from model evaluation and model specialization to controlled testing of agents; the announced next step is continued collaboration across the listed banking applications.