BNP paribas scales AI with Google cloud, sees fast gains

BNP Paribas entered the Google Cloud deal from a position that was already distinctly hybrid rather than all-in on public cloud. Prior to the September 2026 deal, BNP Paribas operated a hybrid/multicloud model that included IBM Cloud hosted in its own data centres alongside on-premises infrastructure, and by 2025, BNP Paribas had not placed client data or production workloads containing sensitive data in public cloud, maintaining strict security and data-governance rules to protect banking confidentiality and operational resilience.

That matters because the 2026 agreement did not mark a reversal of BNP Paribas’s cloud posture; it marked an extension of it. This practice continued under the 2026 agreement, which expanded Google Cloud access within these controls without replacing them, and the 2026 agreement extended this model by adding Google Cloud as an additional infrastructure and AI provider without displacing existing infrastructure or governance.

Gemini Enterprise for Financial Services is being sold not as a generic model endpoint but as a packaged system for regulated work: Google introduced Gemini Enterprise in October 2025 as “a governed environment for employees to use and build agents across workplace data,” and the finance edition adds preconfigured domain capabilities rather than forcing banks to assemble workflows themselves. According to TNGlobal and Newswav, that package combines models, agents, skills, connectors and governance controls, with a Google-managed Financial Research agent, more than 50 specialized skills and 13 connectors to market data, news and regulatory sources for capital-markets and corporate-banking tasks, including examples such as risk management teams executing portfolio shock analyses in under five minutes with automated duration-hedging strategies.

What makes the offer enterprise-grade for banks is the compliance scaffolding around the automation: Newswav says the central research agent can automate workflows while providing confidence scores, methodologies, data snapshots and source citations, while the platform supports 13 secure integrations with licensed providers including FactSet, LSEG, Moody’s, MSCI, PitchBook, S&P Global and SEC Edgar. Technology Magazine adds that the foundational technology operates entirely within an organisation’s private cloud perimeter, customer data, client files and prompts are never used to train base models, and Gemini Enterprise connects enterprise systems, data and security controls within a single centralised environment; onboarding workflows, for example, use multi-format document ingestion to evaluate risk.

European banks are broadening AI adoption because narrow efficiency pilots no longer look sufficient against the larger payoff from institution-wide deployment. Payments Wrap Up found that “30% of European banks cite cost reduction as their primary AI driver – nearly twice as many as those prioritising customer experience,” yet “those focused on customer experience were 40% more likely to achieve transformational gains,” helping explain why banks are doing the harder work of scaling AI across functions, especially as “implementation gaps between markets now exist in every functional area, with a maximum delta of 35 percentage points” and early movers are already seeing compounding returns.

What makes that broader shift plausible is that governed cloud-era AI can be rolled out in stages and still satisfy the control demands of regulated operations. On The AI in Business Podcast, Mark described a “crawl, walk, run type of methodologies” approach to “let's focus on these 10 applications first… create some initial wins… prove the value… and then let's scale it out to prove its value,” while Perforce found “Organizations with formal governance report 94% trust in AI, compared with just 51% relying on ad hoc approaches,” that “66% of organizations are already using AI in infrastructure and configuration workflows, including provisioning, drift detection, and compliance management,” that “52% of organizations have fully automated audit trails,” and that “Organizations with fully standardized internal developer platforms (IDPs) reach 92% confidence in AI outputs,” while “44% of IDP-mature organizations run AI workflows fully autonomously, compared to 26% of organizations,” showing why banks see measurable gains beyond pilots.

The payoff is no longer hypothetical because banks are measuring AI against operating results that matter to managers. Mark McNoli of Citi said, “we’re already effectively using it in our development cycle” and “where we are we’re seeing benefits something like 30 to 40% increase in time to deployment” where AI is used, while another large bank said it tracks “adoption metrics” and “actual outcomes instead of potential outcomes,” including a mobile app where “100 performers were created in the first month” and tasks were completed in about half the manual time.

Those metrics are large enough to move budgets because they translate into service deflection and supplier revenue at scale. The same bank said chatbot “Penny” cut tickets by 60% in four months, while Alphabet reported Google Cloud revenue “jumped for 48% to a whopping $17.7 billion,” above roughly $16.2 billion expected; even with capex projected to “reach as much as $185 billion,” investors treated demand as real, with Mark Mahaney raising his target to $450 from $420 and Ivan Feinseth lifting his to $485 from $415 after BNP Paribas signed its five-year Google Cloud partnership.