The China AI Innovation Study Tour 2025 took delegates to CloudWalk Technology, Douyin, ZhongAn Insurance and Shanghai Pudong Development Bank (SPDB) in Shanghai; MYbank, Ant Group, PingPong and the Zhejiang University–Hangzhou Global Scientific and Technological Innovation Centre (ZJU-HIC) in Hangzhou; OneConnect, Lexin and WeBank in Shenzhen; and Huawei’s Songshanhu campus in Dongguan. Each visit combined a walk-through with an extended briefing and question-and-answer session. The sessions examined operating models, data and technology architecture, compliance and risk, and how artificial intelligence (AI) is embedded into day-to-day processes. In the closing discussion, Professor John Gong of the University of International Business and Economics in Beijing, framed the week within China’s longer trajectory towards technological independence and indigenous innovation, linking the institutions visited to wider national priorities. The programme offered a composite view of how platforms, regulation and talent work together at scale. Platforms and horizontal integration Chinese institutions repeatedly presented platform models that integrate services across industries and channels. Douyin explained how short-video engagement flows directly into retail, with livestream commerce reaching RMB 2.7 trillion (about $380 billion) in gross merchandise value in 2023. The briefing connected recommendation at scale with merchant tooling and logistics, showing how media and commerce co-exist on one platform. ZhongAn described a fully digital insurance model embedded into online journeys. Products sit inside e-commerce and mobility platforms, allowing micro-protection to be offered at the point of need. The company serves more than 500 million customers and has issued over 10 billion policies annually across categories such as health, pet and travel insurance, illustrating how distribution follows users rather than channels. PingPong showed the infrastructure side of platformisation. Focusing on business-to-business cross-border payments since 2015, it processes more than $100 billion annually for over one million merchants in 200 markets. Its briefing set out linkages with marketplaces such as Amazon and regional platforms in Southeast Asia, and how bank partnerships sustain settlement corridors. WeBank presented a consumer and small and medium-sized enterprise (SME) banking model that is native to platforms. Products such as Weilidai and Weiyedai are distributed inside WeChat and QQ, which the bank said account for around 40% of originations. This allows an entirely digital operating model while serving more than 420 million customers. OneConnect’s session clarified the distinction between a parent group walk-through and its own briefing. While the exhibition at Ping An International Financial Centre showcased group capabilities, the discussion focused on OneConnect’s platform solutions for banks and insurers, including customer interaction, risk and claims automation delivered as modular services. CloudWalk set the context for platform thinking from an infrastructure perspective. Its CloudWalk Operating System integrates hardware, algorithm factories and data into one operating environment, moving from narrow uses such as facial recognition into cross-scenario deployment for banking, airports and urban security. The presentation placed AI alongside national infrastructure, not just single applications. SPDB’s research institute briefing linked platform logic to strategy. Li Lin, head of the institute, described the banking sector’s transition across three dimensions — from traditional to modern banking, from finance to financial technology and from technology back into finance — emphasising that new platform models require different strategic assumptions. Data, supply chains and privacy-preserving computing Data was presented as the connective tissue of platform operations. MYbank described the “three-one-zero” model — three minutes to apply, one second to approve and zero human intervention — supported by the “bird systems”. Tomtit uses satellite remote sensing to verify farmland plots and estimate yields for rural lending. Lark processes SME documents with AI to raise credit limits and speed decisions. Goose maps supply chains to identify overlooked distributors and suppliers, while Cuckoo forecasts cashflows to shape product recommendations, including for wealth management. The MYbank team discussed how government-backed standardisation and data ecosystems enable these tools, and clarified that fraud losses are treated separately from non-performing loans. They also noted that about two-thirds of staff are in technology roles, with an average employee age of 31, reflecting the maturity needed for credit and risk work. WeBank described more than 115,000 models trained on 126 petabytes of data, combining internal histories with third-party datasets to strengthen SME profiling and consumer risk. These models underpin product journeys that complete in seconds inside WeChat, demonstrating how data and distribution move together. Ant Group’s anti-money-laundering (AML) team set out how privacy-preserving computing and federated learning allow collaborative analytics without exposing raw data. The briefing walked through sanctions screening, customer due diligence and transaction monitoring feeding a case-management platform for review and reporting, embedding privacy by design. Lexin added a value-chain view from consumer finance. Its teams use simulation to forecast the operational impact of onboarding new funding partners before launch, modelling transaction flows, funding distributions and customer segments to reduce risk and accelerate decision-making. Douyin’s session covered data infrastructure for merchants and advertisers, linking recommendation quality to commercial outcomes. Tools built around its data platform help creators and brands target, transact and fulfil within the same environment, emphasising that platform data supports the whole value chain. Compliance, risk and trust at scale Compliance appeared as an engineered feature of growth rather than an external constraint. Ant’s AML centre presented a product ecosystem built on three pillars — sanctions screening, customer due diligence and transaction monitoring — with alerts escalating to manual review before regulatory reporting. The emphasis was on operating at platform scale while maintaining structured controls. OneConnect discussed risk tools embedded in partner workflows. Micro-expression analysis is used in credit interviews: the system captures facial movements, benchmarks a baseline and compares responses to financial questions, with unusual results routed for human review. The team also described vehicle identification and image recognition to detect duplicate or fabricated claims. WeBank highlighted how fraud detection sits alongside lending systems but is accounted for separately from non-performing loans. Blockchain was presented as a way to strengthen auditability and reporting. The bank’s in-house, cloud-native core supports these controls at high throughput. SPDB’s research institute offered a strategic lens for risk. Li characterised the operating environment as “BANI” — brittle, anxious, nonlinear and incomprehensible — and argued that uncertainty should be handled as both risk and opportunity. Traditional strategy frameworks, it was said, do not capture the speed and shape of AI-driven change. ZhongAn’s cloud-native operations showed how compliance and speed can co-exist. The team linked rapid product design and launch cycles to a core system that maintains controls across categories, which is necessary when products are embedded into third-party platforms. Across sessions, delegates observed a common pattern: detection, escalation and reporting were built into products and processes. Trust was treated as a prerequisite for scale, not a by-product of it, and risk systems were described as part of the service architecture. Artificial intelligence as process-native, from finance to robotics Institutions presented AI as part of the operating fabric. OneConnect showed an AI use case for code writing that generates frameworks and process flows for stock or portfolio-selection models, integrating the output into structured development. Wu Lei, its chief AI officer, also said more than 80% of calls are managed by AI that blends rules and large language models, with escalations for exceptions. Lexin’s LexinGPT underpins more than 50 AI agent applications across requirement preparation, coding, testing and deployment. The company reported monthly code output in the hundreds of thousands of lines and cumulative AI-reviewed code in the millions of lines in 2024, validated by simulation frameworks before implementation. WeBank listed more than 200 internal AI applications across marketing, fraud detection and service. A live demonstration showed a credit journey inside WeChat completing in under five seconds, illustrating how model-driven decisions are embedded into everyday user flows. DeepRobotics broadened the picture beyond finance. Founded in 2017, the company demonstrated three lines of robots, including the flagship X30. Robots climbed stairs, navigated narrow spaces and performed inspection tasks. Use cases covered warehouse and industrial inspection, firefighting support and public-safety scenarios. The company said its systems operate in more than 40 countries, with partnerships that include Amazon and Singapore Power. Delegates discussed costs, competitiveness and regulatory acceptance with the DeepRobotics team. Questions also explored whether quadrupeds could replace police dogs in airports. The company emphasised applications in security, search and rescue and autonomous inspection, where robots can replace humans in hazardous or high-cost environments. Huawei’s campus visit joined these threads with national infrastructure. Presentations covered AI applied to smart cities, networks and cloud platforms, and a 19-level capability maturity framework for structured improvement. Jason Cao, CEO of Huawei Digital Finance, framed the technology giant as a symbol of technology resilience and indigenous innovation, with full-stack capabilities from hardware to cloud. People, training and collaborative research The people dimension recurred in every briefing. Douyin highlighted a workforce with an average age of 27, noting the energy and stamina required to operate a platform that combines content and commerce at massive scale. Rapid iteration cycles depend on teams that can learn and adjust quickly. MYbank emphasised maturity in risk roles, with an average age of 31. Executives linked this profile to the judgment and experience from going through actual business and credit cycles needed for credit decisions that rely on alternative data and automated processes. The bank also noted that around two-thirds of its workforce is in technology roles. Huawei showcased a 19-level capability maturity framework that captures lessons from operations and turns them into structured improvement and talent progression. The approach supports repeatable improvement across technical and managerial tracks and underpins product quality over time. Ant, MYbank and WeBank each described internal academies and programmes to spread AI literacy beyond specialist teams. The aim is to embed data and model thinking throughout operations so that adoption is organisational, not just technical. At ZJU-HIC, managers outlined a collaborative model that connects research to commercial outcomes. The centre focuses on synthetic biology, advanced materials, instrumentation and equipment. It has filed more than 15,000 patents, licensed over 4,700 and incubated around 192 companies, supported by proof-of-concept funds, shared laboratories and office space. Delegates asked how resources are allocated across the four ecosystems. Zhang Chen Chen head of technology transfer at ZJU-HIC said its Future Plus and ZJU Plus incubators cover all four areas rather than specialising. She also described joint labs with European universities and partnerships with more than ten global institutions, as well as innovation competitions where winning projects gain funding and incubation. Technology stacks, regional roles and what can travel WeBank’s technology leadership team described an in-house, cloud-native core built on a proprietary database and modular microservices. The platform is designed for high throughput at 50,000 transactions per second, with products such as Weilidai for unsecured credit, Weiyedai for micro and small enterprises and Weichedai for auto loans serving nearly three million customers. ZhongAn’s cloud-native insurance platform enables product design and launch in days, which is necessary for micro-protection embedded into large consumer platforms. The system supports more than 500 million customers and very high policy volumes. Ant Group presented OceanBase, its distributed database, which has replaced legacy relational systems and supported peak transactions during shopping festivals for over a decade. The team said about 60% of Chinese banks have adopted OceanBase, underlining the mainstreaming of domestic infrastructure. OneConnect discussed exporting technology to Southeast Asia and Africa, noting that localisation for compliance is essential. The briefing covered solutions for digital know-your-customer, risk and customer interaction and the need to align with local rules. PingPong linked its global growth to licensing and bank partnerships, highlighting relationships with institutions such as Citi and CIMB. Expansion plans include Latin America, the Middle East and Africa, built on the same infrastructure approach. Regional roles anchored these choices. Shanghai combined finance and commerce, Hangzhou brought together fintech and research through Ant, MYbank, PingPong and ZJU-HIC, Shenzhen focused on digital banks and consumer finance with OneConnect, Lexin and WeBank, and Dongguan, through Huawei’s campus, highlighted national technology infrastructure. Together, the regions form a coordinated ecosystem. Culture, inclusion and social responsibility Product-embedded responsibility appeared across institutions. Ant Forest has engaged more than 650 million users in carbon-saving behaviour, converting digital points into tree-planting projects. The initiative links everyday product use to environmental outcomes. WeBank described AI-enabled accessibility features that allow more than 6,000 hearing-impaired customers to access services. The team presented inclusion as part of mainstream product work rather than an adjunct programme. MYbank’s clarifications on fraud and non-performing loans framed customer protection within credit policy. Separating fraud losses from loan performance supports accurate risk measurement and consumer outcomes. ZhongAn’s micro-protection products meet specific needs at low friction, increasing uptake among first-time insurance users. Embedding products into digital journeys reduces barriers to adoption. Ant’s briefing covered tap-to-pay via near-field communication inside Alipay and the interoperability of Alipay+ wallets for international visitors in China. Both were presented within a framework that emphasised consumer protection and regulatory alignment. Huawei’s campus highlighted green design alongside technology, presenting sustainability as part of its operating culture. Across institutions, these elements reinforced trust by aligning commercial products with social and policy objectives. Five days, one operating logic Across five days, the programme progressed from AI as operating system to AI as operating habit. In Shanghai, CloudWalk, Douyin, ZhongAn and SPDB showed how infrastructure, commerce and finance are being re-organised around platforms and risk frameworks. In Hangzhou, MYbank, Ant and PingPong demonstrated alternative data, privacy-preserving computation and cross-border payments operating at national scale, and ZJU-HIC connected research to commercialisation. In Shenzhen, OneConnect, Lexin and WeBank illustrated process-native AI in customer interaction, underwriting, fraud and software development. The final day at Huawei framed these elements within full-stack technology and capability maturity. A single pattern was visible across very different institutions. Services are organised as platforms that integrate horizontally; data moves through supply chains and consumer journeys under privacy-preserving controls; compliance is engineered into products; AI sits inside processes; talent is developed continuously; and core systems are owned, cloud-native and built for high throughput. The context is specific. Shanghai, Hangzhou, Shenzhen and Dongguan play different roles; overseas expansion requires localisation; and success depends on structures that link entrepreneurship with regulatory direction. Huawei’s role symbolised technology resilience and indigenous innovation, while SPDB’s strategic framing and DeepRobotics’ demonstrations showed how finance and real-world automation are converging with AI. The result is an operating logic where growth and governance are designed to work together. That, more than any single exhibit or demonstration, is what the China AI Innovation Study Tour 2025 made clear.
