BNP Paribas is gaining broader access to Google Cloud’s AI ecosystem, headlined by Gemini Enterprise and the Gemini models. A five-year deal is set to accompany the bank as it ramps up its capabilities in agent-based AI.
CIB in the Lead
It is Corporate & Institutional Banking (CIB), the division dedicated to large corporations, that kicks things off. The Gemini models will bolster LLM@CIB, the bank’s in-house AI assistant already used by more than 65,000 employees.
As for agents in the strict sense, the initial workstream focuses on preparing credit analysis dossiers, where AI will support the teams rather than replace them, at least for now. The bank is also eyeing other domains: sales, trading, research, structuring.
And there is already a use case running somewhere else in the group. At Nickel, the consumer payments subsidiary, a tool named Nickel Assist relies on a Gemini model to help about 200 customer advisers respond to clients more quickly, drawing on internal procedures.
A Reaffirmed Multi-Cloud Strategy
The deal does not imply a 100% shift to Google Cloud. BNP Paribas has long pursued a multi-cloud and multi-model approach. It sits alongside other providers, notably IBM Cloud (a ten-year contract renewed for 2025) and Oracle, tied to Exadata Cloud Customer hosted in the group’s data centers.
Marc Camus, the group’s CIO, notes that the bank prefers to retain control, choosing model and infrastructure on a project-by-project basis to fit its security, quality, and budget needs. The arrival of Google Cloud in the equation, he says, simply broadens the menu.
Like always, AI adoption is accompanied by a strong emphasis on control. BNP Paribas frames the new contract within its internal rules for data security and governance. Each agent will be identified individually with access limited to only the resources necessary for its task.
BNP Paribas also plans to leverage this collaboration to bolster its internal AI training program, covering both broad foundational training for a wide audience and more specialized capabilities, such as designing specialized agents.
