China’s financial sector has moved through three distinct eras: branch-centric banking, digital banking and now machine-speed finance enabled by artificial intelligence (AI). The first era was defined by the transition from offline services to online platforms, followed by the migration to mobile ecosystems. These shifts were propelled by changing customer behaviour, with users rapidly embracing digital experiences and redefining their expectations of financial services.
The development of mobile payments and e-commerce platforms was about directly responding to customers’ expectations for frictionless, real-time interactions. The resulting platforms and ecosystem business models that emerged have fundamentally reshaped how value flows through the economy.
The rise of machine-speed banking

A new era is now coming into view with the emergence of an intelligent real-time economy, one in which retail, logistics and payments increasingly operate at machine speed. For banks, this means new categories of financial products and operating models across the economic chain to meet evolving customer demands—whether corporate clients’ desire for velocity-based financing and real-time settlement, or retail users’ expectations of streamed income and invisible, contextualised payments.
At the heart of this shift is the deepening role of AI. Across the sector, there is a firm belief that AI is not a shortcut but a multiplier. It strengthens existing digital foundations—personalisation, decision accuracy, automation and speed—while widening the gap between existing digital leaders and laggards. AI lowers the barrier to entry and raises the performance ceiling, creating a structural divide between AI-first banks and everyone else.
Competition has evolved in tandem. The competition between banks—expressed in the number of branches, products, asset size and relationship managers—morphed into a battle of mobile apps, and has now become a race between banking AI agents.
The next shift will be bigger still: customer AI versus bank AI. While customers use their own AI assistants to search for products, compare banks and analyse wealth plans, banks will rely on their own intelligent systems to design, price and deliver services. In this emerging paradigm, the traditional two-party banking relationship gives way to a four-party dynamic: customer, advisor, bank AI and customer AI.

This shift exposes gaps in data security, regulatory compliance, high-value scenario implementation, business-value visibility, engineering experience, data quality, AI talent, sustainable compute economics and architectural planning. These challenges can be summarised in a unified framework—also known as the “4S” constraints—which sets out the conditions under which modern financial AI can operate:
4S constraints
Sovereignty over data and models
Security against evolving threats
Scalability of AI across millions of users
Sustainability of compute economics
Banks face real challenges in modernising legacy infrastructure to address the 4S constraints and meet the more complex requirements of the AI age, such as real-time data processing, widespread AI adoption and continuous risk authorisation.
However, valuable lessons emerge from Chinese banks’ experiences. Supported by open-source foundation models, hybrid cloud environments and domain-specific financial models, mature institutions have moved away from fragmented, siloed builds. Instead, they have pivoted to AI-native architectures designed around five principles:
AI-native architectures designed around five principles
These principles operate as an integrated system, and their practical impact is measurable. One leading bank, for example, built a risk-analysis agent across nine risk domains. Using only a foundation model, effectiveness was around 12%. Integrating structured business knowledge raised its effectiveness level to 57%, while targeted model engineering pushed it even higher to 86%.
The lesson is clear: AI’s real value in banking is realised through domain intelligence engineered around proprietary data, workflows and risk logic.
AI’s real value in banking is realised through domain intelligence engineered around proprietary data, workflows and risk logic.
The foundations of agentic banking
China’s banks have shown that building agentic banking begins with strong digital foundations: modernising the core, re-architecting data and redesigning infrastructure.
All of China’s large and mid-sized banks have already completed their legacy-to-cloud core modernisations, and smaller regional institutions are now navigating the same transition. Two national banks—with more than 700m and 650m customers respectively—completed their core migrations within about five years. Regional banks serving 30m to 60m users have done so in two to three.
Moreover, though cloud technologies were not invented in China, banks there have adopted them for core applications at such a speed and completeness that their experiences now serve as unique reference points for institutions elsewhere. These modern, cloud-based cores provide the elasticity, throughput and deployment discipline that AI needs to operate as part of the institution’s operating fabric. However, modern cores alone do not spell intelligence, as AI also depends on access to real-time data, unified semantics, fine-grained access control, audit logging and continuous feedback loops.

Chinese banks have responded by building cognitive data systems structured around four pillars:
Real-time processing, which gives the institution reflexes, capturing behavioural changes and business signals the moment they occur
All-domain collaboration, which dissolves departmental barriers
Converged knowledge, which provides the AI model with the semantic spine it needs to transform raw data into reusable financial expertise which provides the AI model with the semantic spine it needs to transform raw data into reusable financial expertise
Experiential intelligence, which turns knowledge into intuitive interactions
This data architecture changes how risk, fraud and creditworthiness are perceived. Risk becomes a reflex rather than a report; fraud detection evolves from static rules to behavioural anomaly recognition; and creditworthiness becomes dynamic and context-aware. The institution no longer waits to discover risk because it anticipates it.
Infrastructure must evolve at the same pace, especially in today’s cyber landscape. As adversaries increasingly adopt AI-driven attack strategies and quantum capabilities emerge, institutions must deploy cryptographic systems that can withstand both classical and quantum attacks. Real-time anomaly detection, autonomous threat containment and self-healing network fabrics have become essential components of modern digital resilience. For the entire industry, four uncompromising outcomes have become structural requirements for participating in the real-time economy: zero downtime, zero wait times, zero trust and zero touch. These outcomes rest on four core properties of modern infrastructure:
Four core properties of modern infrastructure
Reliability, meaning organisations can operate through failure rather than simply recover from it
Availability, meaning perfect data integrity across distributed consensus
Autonomy for infrastructure to manage itself through digital twins and AI-driven observability
Security becomes adaptive, behavioural and collaborative, with defensive agents sharing threat intelligence across institutions
The institution no longer waits to discover risk because it anticipates its emergence right from the get-go.
Core modernisation, unified data and resilience infrastructure are the digital foundations that allow AI to scale and become the institution’s execution engine.
Huawei’s engineering statecraft and ecosystem orchestration

China is the world’s first large-scale cashless economy, and the digital-finance path it pioneered is now being replicated globally—from Brazil’s Pix to Singapore’s PayNow. The scale and coherence of China’s banking transformation did not emerge from technology alone, but rather reflect a long period of engineering discipline and ecosystem building, in which Huawei has been one of the central actors.
Over the past 16 years, Huawei has worked closely with China’s financial institutions to modernise cores, build cloud-native architectures and deploy AI across high-value financial systems. It has contributed from a technical perspective, and also in terms of the underlying technological architecture, operations and organisation.
As such, Huawei’s role has been to turn AI from a promising capability into an industrial one.
As such, Huawei’s central contribution has been to turn AI from a promising capability into an industrial one. At the centre of this contribution is what can be described as engineering statecraft—a systems-level effort of shaping the entire AI stack into a coherent, governed and scalable capability. Huawei has organised this into four interconnected disciplines:
Four interconnected disciplines
Model engineering transforms open-source foundation models into domain-specific intelligence through layered model matrices, fine-tuning and version governance, giving banks proprietary capability rather than outsourced AI.
Compute engineering ensures that intelligence can operate at scale, optimising the full inference chain through caching, acceleration, scheduling and concurrency tuning, so that AI remains economically viable under real-world resource constraints.
Agent engineering turns AI from a conversational interface into a business executor, using unified intent routing, task orchestration, modular skills, streamlined context and full-chain auditing to embed intelligence directly into workflows.
Security engineering integrates compliance, access control and threat response into every layer of the model, compute and agent stack, ensuring that intelligence remains safe, governed and deployable across core financial systems.
Together, these engineering disciplines can determine a bank’s intelligence level, its ability to scale AI economically and evolve AI beyond conversation into an executor of business operations.
Huawei’s Ronghai Program brings together more than 180 partners across 30 categories, forming an ecosystem with consulting, service, integration and solution capabilities that can be reused across financial scenarios. The ecosystem operates through a “4-Win” model that ensures that all stakeholders—financial institutions, independent software vendors, local system integrators and Huawei—benefit.
Building a 4-Win model to create new value
Financial institutions, for example, are able to achieve transformation at industrial scale with new solutions, while their partners are empowered to reach new markets. Local system integrators are given opportunities to deepen their role in delivery and expand their solution portfolios, while Huawei focuses on strengthening the infrastructure that makes large-scale AI possible.
Huawei’s engineering and ecosystem capabilities provide the foundation needed to translate the banks’ five principles into operational reality and spur groundbreaking innovations, such as new financial scenarios, operating models and pathways for delivering real-time intelligence. Its expertise builds the missing middle that ensures these advances have the stability, governance and performance they need for enterprise-scale deployment. At the same time, its ecosystem prevents fragmentation, allowing thousands of scenarios, partners and applications to grow in parallel on a coherent technical base.
By riding on this engineering capability and ecosystem scale, China has entered a new age of industrialised AI-native banking.
Explore more content

Article
Laying the foundations for an AI-driven banking revolution
Banks have been among the earliest adopters of digital innovations, whether it is computers in the 1950s or big data platforms at the turn of the century.

Article
Opening new realms of innovation
Intellectual property (IP) rights have a long history, dating back to ancient Greek chefs who were granted year-long “patents” for inventing new dishes. Since then, IP laws...

Article
Advanced wearables paving the way to better health
The invention of the “Manpo-kei” (10,000 steps) pedometer in 1965 heralded the beginning of a wearable technology industry that is worth about US$100 billion today.

