The AI boom is putting significant pressure on the infrastructure that supports bank modernisation. As hyperscalers and major AI buyers absorb server capacity, chip supply and related infrastructure, fulfilment timelines are extending drastically and procurement is becoming increasingly difficult to manage. As a result, banks are facing a multifaceted issue: They need enough compute to support current operations but also need to build a foundation that supports AI workloads with scalability in mind.
Supply chain pressure for critical components
The supply challenge is not limited to one component or vendor. AI’s high demand is consuming server providers, chip makers and other key infrastructure suppliers at the same time, lengthening lead times and making normal refresh cycles difficult to manage. This bottleneck is putting pressure on banks to think differently about timing, sequencing and planning. In today’s landscape, buying hardware is no longer just an IT ask; it is a strategic necessity.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
For institutions, the point is simple: If lead times are stretching, then there must be a deliberate focus on identifying areas for simplification, deciding which ones can wait for modernisation and which need to be prioritized. In the past, infrastructure refresh was viewed as a predictable cycle, but in today’s capabilities race, swift action is required to keep up.
Stretching existing capacity
One practical response we’ve seen involves institutions making better use of what is already readily available within their network. Many large institutions currently operate with a significant infrastructure across hybrid environments, leveraging both cloud and native data centres. In today’s market, even these large institutions are forced to stretch their existing model by creating space, where possible, to temporarily increase capacity. This often includes simplifying applications, reducing duplicated efforts and leveraging AI tools to identify inefficient code or underutilised systems. In practical terms, they are creating headroom and temporarily delaying the need for hardware that is currently unobtainable or out –of budget.
This is a key approach because institutions are not starting from a blank slate. The infrastructure is largely in place and much of it remains deeply integrated to core operations. The question is not whether everything must be replaced, but how institutions can make the current infrastructure work while supply remains tight.
Regulation and resilience
The regulatory angle is an important consideration but should be viewed as a part of broader resilience discussion for financial institutions across the board. The Boston Fed recently published a paper highlighting the risks of relying too heavily on outside technology service providers, especially when these outages impact customer access and transaction continuity. The FDIC’s 2025 Report on Cybersecurity and Resilience reinforces the need for institutions to manage supply-chain risk and understand where single points of failure can emerge across a bank’s technology stack. The message is to not abandon the cloud, rather, it is to avoid overconcentration and keep enough control in place to protect critical services.
For banks, the risk is not just focused on technical failures that impact customers; it is also imperative for operational and strategic modernisation. If a bank is overly dependent on a narrow set of providers, its ability to maneuver during interruptions or when the market shifts, is handicapped — as we’re currently seeing with AI. That is why resilience is becoming part of infrastructure planning as a strategic initiative rather than a risk-management exercise.
The hybrid model
A hybrid of both native data centres and cloud services remains the practical model for the majority of financial institutions. Most large banks aren’t starting from scratch. They’ve already developed long-standing investments in data centers, mainframes and legacy systems that are difficult, and often costly, to move wholesale. The cloud offers speed, automation and geographic flexibility, but is not the right answer for every workload. Institutions should prioritise core or low-latency systems, use cloud services where it adds clear value and avoid forcing every application into the same operating model.
With most large institutions managing a mix of old and new infrastructure, this mix is exactly what allows them to keep modernising while holding on to mission-critical legacy systems. In addition, some applications are self-built and are overly complicated to re-engineer or move to the cloud. Other applications may be better suited for the cloud because they benefit from speed or have the need for scalability. Overall, hybrid models are the go-to approach because they cater to the complex, differing needs of an institution’s range of applications.
The required innovation to support AI growth
The growth required to support AI makes this balance far more complicated. For instance, many institutions are experimenting with agentic AI to support customer service needs – an AI-service model that can quickly grow from a few hundred test agents to thousands. This level of growth is only possible with the proper infrastructure capacity and operating model, often requiring a mix of on-premise and cloud-based data services. In addition, banks need to manage privacy, jurisdiction and governance regulatory requirements that directly impact where AI tools can operate and how the data is handled. The real challenge is not deploying these AI tools; it’s effectively integrating them into production where business, risk and compliance functions are able to sustain.
This is the inflection point where infrastructure planning becomes a key piece of institution’s modernisation discussion. If banks are committed to supporting larger AI footprints, a foundation that supports room for growth without losing control is paramount. This issue goes beyond procurement issues; it impacts how a bank is organised and how much operational change it can absorb while maintaining operational oversight.
Infrastructure planning as a strategic capability
Longer lead times are forcing banks to think more strategically about how they buy, extend and use their infrastructure. Some institutions are looking at ways to simplify their application estates and improve efficiency amidst hardware supply issues. Others are using longer-range planning with more diversified vendor strategies to reduce the risk of delay. The banks that closely manage their approach to infrastructure planning will likely save money in the long term, as well as preserve flexibility and extend the value of the assets they’ve already deployed.
It is no longer enough to treat infrastructure buying as a routine back-office IT task. Institutions need to understand the workload at hand, how long the current infrastructure can support it, what can be simplified and where future demand will come from. In a constrained market, the institutions that plan early and work cohesively across technology, risk and resourcing strategy will have the foundation for modernisation.
The next phase of banking modernisation will not only be defined by the cloud or on-premise alone. It will be determined by whether institutions can build a resilient infrastructure that supports AI, protects customer data and offers enough control to adapt as conditions change. Treating infrastructure, cloud and governance as components of a larger modernisation strategy will be key for institutions to ensure they are not left behind in the current and future financial services landscape.
John Carey, Managing Director at AArete
