AI push forces banks to ditch old cores

The timing of the modernization wave is less about fashion than exposure: AI and real-time banking are revealing that cores built for batch processing cannot carry adaptive, always-on services. Leonardo Collado of Visa’s Pismo said AI is exposing “the structural limits of systems originally designed for batch processing and slower settlement environments,” adding that legacy infrastructure cannot support real-time contextual personalization and that “you can’t just place new tech on old stack,” making end-to-end rebuilds and cloud migration a competitive requirement rather than an upgrade option.

The pressure is operational as much as technical, because old estates consume the money and control capacity banks need to compete in faster, more connected markets. Collado said “roughly 70% of IT budgets are often devoted to maintaining outdated systems,” leaving institutions with “limited capacity to build new products or customer-facing capabilities,” while Gulf Times noted that after the Saudi Central Bank began licensing open banking providers in March 2026, covering both account information and payment initiation, every new external connection increased the strain on brittle point-to-point architectures behind the interface.

What is broken is not just an old core, but the layers built around it. Banks long avoided changing monolithic mainframe-based cores because of cost, security, and scalability fears, so they kept adding new technology on top. That approach now fails because many legacy architectures were built for product and ledger processing, not always-on digital journeys. The result is a patchwork that fragments customer and operational data across separate repositories and workflows, making it harder to build a complete customer view, automate decisions, or give employees one coherent screen to work from.

Integration is where the damage compounds. Once systems are deeply intertwined, even a core replacement can turn into dozens of dependent projects, hidden capabilities, and manual governance work that slow delivery and automation. PYMNTS said service, fraud, and lending teams often operate on incomplete context, and Tearsheet added that a modern core on top of fragmented data does not get a bank very far. KPMG’s 2026 Banking Technology Survey found 71% of banking executives say they need platform modernization to bring new or enhanced products and services to market, while TechBullion noted that 60% of companies experience more cyberattacks during integration phases and cited IBM’s estimate of a $4.45 million average breach cost.

The clearest sign this shift is real is that banks are no longer talking about modernization as a future aspiration; they are disclosing live, multi-year rebuilds of the core itself. Briefglance reported that Valley National Bank has been overhauling its core banking architecture as a “high-stakes, multi-year project,” alongside a “massive ‘cloud-first’ migration” that moved “over 80% of its data center capacity to the cloud” and created a centralized enterprise data hub, a foundational re-platforming that the bank says has already delivered nearly $100 million in cost savings, about $20 million in new recurring revenue, more than 100,000 hours of employee productivity, a 200% improvement in application performance, 3x faster deployment cycles, and a 50% reduction in server footprint.

That same pattern is visible across other institutions, with named platforms, migration timelines, and operating results replacing abstract strategy. Regions said it partnered with Temenos in 2023 to move from legacy core systems to a SaaS-based modern core and was roughly two years into that transition by August 2026, while SACOMBANK said its modernization reached stable production go-live with provisioning cut from about two days to six to eight hours and infrastructure reduced from eight servers to four; Deutsche Bank’s Private Bank, meanwhile, announced plans to cut its core banking systems from 15 to two modern, cloud-based platforms, with products migrated progressively from 2027 onward through a phased parallel run.

The evidence says AI is no longer sitting in pilot labs or customer-service demos; it is moving into the operating core of the enterprise. CGI’s 2026 Voice of Our Clients Canada research, based on 278 in-depth discussions with business and IT executives across 179 Canadian organizations, found that 64% of Canadian executives say their organizations are applying AI to core business and operational processes, while 61% say they are quantifying results from AI implementation, showing that AI is being managed as an operational capability with measurable performance expectations.

What blocks that shift is not lack of interest but weak foundations: CGI found 65% cite security and cyber risk mitigation as a key goal in optimizing cloud adoption and value delivery, 51% cite data sovereignty and localization, 49% say legacy systems significantly challenge digitization, data and AI strategies, and 85% plan to modernize more than 20% of their core business application portfolio over the next two years. In financial services, PYMNTS Intelligence similarly found 95% of firms have broadly deployed or embedded newer AI in data and technology functions, yet only 25% have adopted AI in API orchestration and integration, while 30% identified data quality and fragmentation as their leading obstacle to further AI deployment.