AI pilots are becoming common across banking. The institutions that pull ahead will be those that turn experimentation into industrial operating capability.

Jason Cao believes banking is approaching a structural turning point. As chief executive officer of Huawei's Digital Finance Business Unit, he works with financial institutions navigating digital infrastructure modernisation, operating model redesign and artificial intelligence (AI) deployment. In his view, the next phase of banking transformation will be more fundamental than improving existing processes through incremental automation. The underlying architecture of banking itself is beginning to change.

The force driving that shift, he argues, is not AI alone. It is the emergence of a real-time economy in which customer expectations are increasingly shaped by immediacy. Digital commerce, payments, fulfilment and interaction cycles are accelerating in ways that challenge traditional operating assumptions. Institutions designed around human-paced workflows may increasingly find themselves structurally misaligned with customer behaviour.

That creates a more consequential question than whether banks should adopt AI. The strategic issue is whether institutions can evolve operating models capable of functioning at machine speed without compromising governance, resilience or control. Technology adoption, in this context, becomes inseparable from operating design.

Cao's argument is grounded less in futurist speculation than in execution. He speaks about infrastructure choices, engineering capability and production deployment rather than generic transformation narratives. Pilot programmes are relatively easy. Industrialising AI inside core banking environments is materially harder.

That distinction shapes his wider thesis. The competitive divide in banking may increasingly emerge not between institutions experimenting with AI and those that are not, but between banks capable of scaling intelligent systems operationally and those that remain trapped in fragmented pilots.

How does banking change when the economy moves at machine speed?

Cao's starting point is behavioural rather than technological. "We are moving to a real-time economy," he said.

For much of the past decade, digital banking transformation focused on improving customer access. Mobile applications, digital onboarding, self-service servicing and faster payments reshaped convenience and accessibility. Those changes were significant, but they largely operated within traditional institutional structures where human workflows still governed many internal processes.

Cao believes the next phase is structurally different. Customer expectations are increasingly shaped by industries where speed is measured in seconds rather than business days. Cao pointed to a simple example from China: ordering a suitcase after midnight and receiving it within 29 minutes. Commerce does not pause. Fulfilment does not wait for operational queues. Banking operating models built around human-paced workflows increasingly come under pressure in such an environment.

That creates tension for institutions still organised around human-paced operating models. "If banking services remain mainly human-based, you cannot provide that kind of machine speed," he said. His argument is not that human judgement becomes irrelevant. It is that institutions relying excessively on human intervention for routine operating execution may struggle to remain competitive.

AI becomes strategically relevant in that context because it enables new operating possibilities. The real shift, in Cao's telling, is not automation for its own sake, but the evolution of banking towards AI-native operating architectures capable of supporting machine-speed execution.

What foundations define the AI-native bank?

Cao is sceptical of narratives that treat AI as a modular capability that can simply be layered onto existing institutions. In his view, building an AI-native bank depends on deeper architectural choices.

"You need the computing power, the model, the agent capability and the engineering capability," he said. The list reflects an ecosystem view rather than a narrow software perspective. AI capability is not a single product decision. It is an integrated operating stack.

One of the more strategic questions concerns control. Many global institutions began AI experimentation through external closed-model environments. Cao sees an alternative path emerging, particularly among Chinese financial institutions that have adopted open-source model strategies.

"A lot of leading banks in China have taken a different approach. They use the open-source model," he said. The appeal is not ideological. It is operational. Data remains inside the institution rather than being pushed into external environments. Model behaviour becomes more transparent. Explainability improves. Regulatory conversations become easier, including in conversations with regulators. Institutions can also refine domain-specific models using their own business knowledge rather than relying solely on generic foundation models. Long-term economics may also become more manageable at scale.

"It avoids the black-box problem," Cao said. He also argued that open-source ecosystems may accelerate innovation because broad communities contribute continuously to model improvement, narrowing the gap with proprietary alternatives.

For banks, AI capability is inseparable from governance, compliance and operational accountability. Foundational architecture choices therefore shape far more than performance.

Why do AI pilots fail to become operating capability?

Cao draws a sharp distinction between experimentation and execution. "Pilots are much easier," he said.

AI pilots are now widespread across the industry. Customer service experiments, productivity assistants, internal copilots and narrow automation projects have become increasingly common. Many generate early enthusiasm. Far fewer mature into scaled operating capability.

Cao believes the missing discipline is engineering. Referencing what he described as the "Missing Middle" problem, his argument is that many institutions can innovate or pilot successfully, but far fewer can industrialise AI at production scale.

"The difficult part is getting it to scale," he said. "The key to getting it to scale is engineering capability." This is the central argument in his leadership perspective.

The problem is not model capability alone. Scaling AI inside production banking environments means balancing multiple operating demands simultaneously. Speed matters. Performance matters. Quality matters. Security matters. Economics matter. Systems integration matters. Each pressure becomes harder as deployment moves closer to live production.

"Quality. Security. Cost. Speed. You have to get all these things together," Cao said. His framing shifts the discussion away from AI as a conceptual innovation and towards industrial execution discipline. The most important capability may not be the model itself, but the institution's ability to operationalise it reliably.

What does industrial AI deployment look like in practice?

Cao grounds his argument in live production deployment rather than abstract theory.

The significance lies not simply in automation. Many institutions can automate narrow workflows. Cao's argument concerns industrial reliability inside meaningful operational processes. AI becomes strategically relevant when it performs consistently in production, not merely in controlled experimentation.

This reinforces his wider thesis. The conversation around AI in banking often centres on use cases. Cao is more interested in repeatable operating capability. One successful deployment matters less as an isolated case study than as evidence that institutional scaling is achievable when engineering discipline is present.

Can institutions buy AI capability, or must they build it?

Cao is clear that technology acquisition alone is insufficient. "We do not just give them the boxes," he said.

The challenge, in his view, is institutional capability building. AI at scale requires more than infrastructure procurement or vendor implementation. It demands engineering literacy, operating discipline and organisational adaptation.

"We help them build this engineering capability," he said. That capability spans technical and organisational dimensions. AI deployment cannot remain confined to isolated technology teams if it is expected to reshape operating models.

"Technical teams, business teams and data teams need to sit together," Cao said. He described this less as conventional training than institutional enablement. Cross-functional workshops, collaborative problem-solving sessions and engineering methodology transfer are intended to help institutions build internal AI capability rather than remain dependent on external implementation support.

The formulation reflects a cross-functional execution model rather than a traditional technology project structure. That matters because production AI affects workflows, decision-making, governance and customer experience simultaneously.

His perspective treats AI less as a technology procurement decision and more as institutional transformation through capability accumulation. The long-term winners, in that framework, may be institutions that internalise engineering discipline rather than outsource strategic competence.

Why does AI become a competitive urgency?

Cao does not frame AI as another optional innovation cycle. He sees a strategic divide emerging.

"In future, there will be only two types of banks: AI banks and others," he said.

The underlying point is clear. AI capability may increasingly shape structural competitiveness rather than incremental productivity.

"The fast mover will definitely have a very, very big competitive advantage," he said. Cao's argument extends beyond customer-facing innovation. Institutions that accelerate engineering transformation itself may shorten modernisation cycles, improve cost efficiency and move faster than slower incumbents still constrained by legacy execution models.

Speed matters because capability compounds. Institutions that industrialise AI earlier may benefit from faster learning cycles, stronger engineering maturity and earlier operating redesign. Competitive advantage in this context is not simply about deploying better tools, but about redesigning the institution itself around faster decision-making and more scalable operating models.

That does not imply indiscriminate deployment. Cao's own emphasis on governance, architecture and engineering discipline argues against careless acceleration. Institutions that move quickly without operating discipline may simply scale risk faster. The competitive advantage lies in building institutional capability faster and more effectively than peers, not in rushing isolated pilots into production.

Organisational inertia may be the greater strategic threat. Institutions that continue treating AI as fragmented experimentation may discover that competitors have already moved towards structurally different operating models. By the time the competitive shift becomes visible in market outcomes, the underlying capability gap may already be difficult to close.

For Cao, the real question is no longer whether banks will adopt AI. It is whether they can evolve quickly enough to remain relevant in an economy increasingly operating at machine speed.