AI projects are stalling everywhere in the banking industry.
Boards have approved the budgets, pilot projects are running, there’s a roadmap full of use cases, and the models are working … but not much is changing.
When it’s time to plug AI into payments, credit decisions and customer journeys, performance stalls as integrations get messy and compliance flags issues.
The ambition is there, but the reality is that they’re short on the foundation needed to make it real.
Research from our global report, Cloud-led innovation in the era of AI: The new rules for driving value with cloud, shows that although a staggering 98% of banking and investment organizations say the rise of AI has increased their need for cloud investment, only 14% consider themselves to be at the highest level of cloud maturity.
The bottleneck isn’t AI; it’s the cloud behind it
There’s a tendency to treat AI itself as the hard part. While implementing and managing models and algorithms can be complex, it’s rarely what holds things back. The issue is that AI is only as effective as the environment it runs in.
For example, a fraud-detection model can flag a suspicious transaction in milliseconds, but it may take minutes to pull the right data. And if the system can’t trigger an action in real time, or can’t act without manual intervention. the value is lost. The insight exists, but the platform can’t act on it.
To help organizations avoid these bottlenecks, cloud has become the execution layer of AI — the place where decisions are made, workflows are triggered and systems scale under pressure.
However, after years of cloud migration, many banks are still treating cloud as a place to host applications, not a platform to run the business and carry the weight of AI.
Legacy infrastructure is not solely to blame
It’s easy to blame legacy systems — and yes, they’re part of the problem. But migrating to the cloud doesn’t help much if data is still scattered across systems and applications remain tightly coupled, with fragile integrations between them. When AI enters the picture, it will hit the same walls.
This is why so many AI programs feel stuck. They’ve been deployed but not fully embedded, and it shows in the data. Half the organizations surveyed for our report say the need to modernize applications and data platforms is holding back cloud innovation.
So, AI doesn’t break because models fail but because the systems around them can’t support their reliance on real-time data, event-driven decisions and continuous change.
What “modern cloud” means in the AI era
Modernization is often discussed in the wrong way. Rather than just moving workloads from one place to another, it’s meant to change how entire systems behave.
When customers open their banking app, check their balance and get a personalized offer in real time, based on market conditions, risk signals and transaction history, several things have to happen instantly:
- The data has to be available and up to date.
- The systems need to talk to each other without interruption.
- Decisions have to trigger actions automatically.
- Everything must be governed and auditable.
In other words, cloud needs to operate as three things at once in the era of AI: a real-time data backbone, a platform where intelligent workflows run end to end, and a controlled environment where decisions can scale safely.
This can be achieved only through cloud-native design — event-based, modular systems, driven by application programming interfaces, that can evolve without disrupting what’s around them.
Architecture is now a regulatory decision
Banking is a highly regulated industry. As AI becomes embedded in core workflows, regulators are scrutinizing both the models and the environments they run in. Banks must comply with regulations such as the European Union’s Digital Operational Resilience Act and India’s Digital Personal Data Protection Act.
An NTT DATA report on AI leaders in banking and financial services shows that 62.5% of these organizations flag cross-geography data privacy and sovereignty as top governance concerns.
This is why, more broadly, 99% of organizations in our cloud research expect private cloud adoption to increase, driven by security, sovereignty and compliance needs, while sovereign cloud adoption is projected to rise by 50% in the next two years.
If banks’ architectures aren’t designed for regulatory-grade control, their AI strategies won’t scale, regardless of how strong their models are.
Why cloud and AI strategies must converge
We often see AI and cloud being treated as separate conversations. AI is the responsibility of innovation teams, while infrastructure teams look after cloud. But several things happen when AI and cloud strategies evolve in isolation:
- AI ambitions outpace what the platform can handle.
- Cloud investments lack a clear business purpose.
- Scaling becomes slower, more expensive and harder to control.
Our research shows that leading organizations start with the use case and build the architecture with AI workloads in mind from day one. Data, platforms and governance are aligned up-front.
This produces a very different result: AI that doesn’t need to fight its environment to scale but is instead supported by it.
How leaders are moving forward (and why their approach works)
As banking gets smarter, agentic AI systems are becoming more widespread. In lending, for instance, documents can now be collected, verified, assessed and approved with minimal human intervention because an AI system is orchestrating the entire workflow and making decisions — continuously and at scale.
These agentic systems are already triggering fraud investigations when anomalies appear, orchestrating complex lending and onboarding workflows, and personalizing customer interactions in real time.
Any one of these AI workflows might now involve half a dozen applications, multiple data sources and, increasingly, autonomous agents that make decisions along the way. Trying to manage all this manually or through disconnected systems quickly becomes difficult.
Some banks are already modernizing in layers to fix what matters most, where it matters most. Our global AI research shows that 62.5% of banking and financial services AI leaders use hybrid deployment models — combining plug-and-play solutions with selective co-innovation — compared with just 29.2% of banking and financial services laggards.
This usually leads to a mix of hybrid cloud and targeted application changes, with a strong platform foundation to bring everything into one place:
- A clear, end-to-end view of what’s happening in every system
- Governance built into how decisions are made and executed
- Continuous visibility of cost as AI workloads scale
- Automation that keeps things moving without constant intervention
Fix the foundation or limit your AI future
Most banks aren’t operating at a level where cloud is the execution layer of AI.
To get there, banking leaders need to start thinking about cloud differently. They should see their infrastructure as the foundation for how the bank runs and grows, leading to faster innovation and better margins.
Only then will they maintain their competitive edge in the era of AI.
This article was co-authored by Kane Stavens, Managing Director: Asia Pacific at NTT DATA.
WHAT TO DO NEXT
Access the NTT DATA report, Cloud-led innovation in the era of AI: The new rules for driving value with cloud, to see how your organization’s cloud maturity measures up.