"AI Competition" in the Annual Reports of Listed Banks: By 2025, the six major banks will have invested over 130 billion yuan in financial technology, with scenario implementation and risk challenges coexisting

Daily Economic News reporter | Liu JiaKui    Daily Economic News editor | Wei Wenyi

As the 2025 A-share listed bank annual report season wraps up, a set of figures sketches a brand-new picture of the financial industry’s intelligent transformation: Industrial and Commercial Bank of China’s full-year investment in financial technology reached 28.59B yuan; China Merchants Bank claims that its AI (artificial intelligence) applications replaced more than 15.56 million human hours within one year; and Ping An Bank’s large-model application scenarios doubled within one year, growing to nearly 400⋯⋯

A reporter from the Daily Economic News (hereinafter referred to as “Daily Economic News reporter”) noted that in 2025, six major state-owned banks—including Industrial and Commercial Bank of China, Agricultural Bank of China, Bank of China, China Construction Bank, Bank of Communications, and Postal Savings Bank—collectively invested more than 130 billion yuan in financial technology, up further from 125.46B yuan in 2024. Behind the huge spending, a deeper shift is under way: artificial intelligence has moved from the technology sections that look ahead in annual reports to becoming a key yardstick for measuring a bank’s core competitiveness.

Meanwhile, across the ocean, JPMorgan Chase is painting another AI vision: CEO Jamie Dimon has positioned artificial intelligence as a “transformative technology on par with the printing press and the steam engine,” and announced that it will invest more than $2 billion every year to build an “all-AI collaborative enterprise.” This Wall Street financial giant is not satisfied with single-point applications; instead, it is trying to deeply integrate AI into every fine capillary of the organization.

On one side is domestic banking’s systematic, large-scale AI investment and scenario deployment; on the other is a comprehensive intelligent overhaul pursued by international financial giants using an ecosystem mindset. This financial-intelligence wave spanning the Pacific is quietly changing every core link—from credit approval and risk pricing to investment decisions.

However, behind this hot wave of AI investment and vision lie the deep waters of data governance, the real risks of “model hallucinations,” and compliance challenges brought by algorithmic “black boxes,” all testing the depth and sustainability of this transformation. The AI journey in finance is not only showing immense potential, but also entering a crucial phase that requires more wisdom and prudence.

Strategic upgrade: the strategic sprint from “digital” to “intelligent digital”

After reviewing the performance reports of 2025 listed banks, Daily Economic News reporters found that “artificial intelligence” has jumped from technology outlook sections to become a key performance indicator for measuring future core competitiveness. The focus of this race is shifting from “whether to apply AI” to “how deeply to apply it and how strong the system is,” showing clear characteristics of systematic, large-scale deployment.

Backed by abundant resources, the state-owned big banks are building “heavy-duty infrastructure” for AI transformation. In its annual report, Industrial and Commercial Bank of China stated that it has fully upgraded its four-year “Digital ICBC” (D-ICBC) strategy to “Intelligent Digital ICBC” (AI-ICBC). Its core “Gongyin Zhiyong” large model has been deployed across more than 30 business domains, landing more than 500 application scenarios. China Construction Bank disclosed that AI technology has been scaled up to empower 398 scenarios within the group. Bank of China, meanwhile, has built a BOCAI large-model capability platform, with more than 400 intelligent assistant deployments in total.

Joint-stock banks and city commercial banks, in contrast, have shown greater agility in the speed and breadth of scenario rollout. At a performance briefing, China Merchants Bank disclosed that its AI application scenarios have reached 856. Over the full year, AI replaced more than 15.56 million human hours, equivalent to forming more than 8,000 full-time-equivalent human-efficiency levels. More importantly, AI is moving from a “cost center” to an “efficiency engine.” For customer managers, the bank’s intelligent assistants help increase the average number of effective outbound calls per person by 14% and raise the average transaction scale per customer by 20%. Ping An Bank’s large-model application scenarios surged from “more than 200” to “more than 390” within one year. The share of code generated by AI has already exceeded 30%. Citic Bank has built a “large model + small model” collaborative mode, and as of the end of 2025, more than 120 scenarios had been deployed with the large model.

From “AI-first” to “AI-native,” leading banks are trying to embed intelligence deeply into the organization’s very fabric and build new competitive barriers.

A senior banking-industry research source told Daily Economic News reporters that the concentrated disclosure of AI results in 2025 annual reports indicates that China’s banking industry’s digital transformation has entered a “deep-water zone” centered on intelligent decision-making and process reshaping. Behind this is an inevitable choice to seek efficiency and growth through technology amid the backdrop of ongoing narrowing net interest margins. AI investment is no longer just a budget item for the technology department; it is strategic investment directly tied to core operating indicators such as cost reduction and efficiency gains, risk control, and revenue improvement.

Deep application: the efficiency revolution in risk control, inclusive finance, and operations

After years of exploration, AI’s use in banking has long gone beyond early intelligent customer service and facial recognition payment, penetrating the core of business operations and demonstrating disruptive potential in both improving efficiency and controlling risk.

In the “heart” of risk management—credit and anti-fraud—AI is achieving a qualitative shift from “rule-based judgment” to “intelligent sensing.” Traditional risk control relies on historical data and static rules, making it hard to deal with complex and ever-changing new types of risk. But an intelligent risk-control system centered on machine learning and graph computing can process massive amounts of heterogeneous data in real time. For example, Postal Savings Bank has built an end-to-end anti-fraud model system; in the first half of 2025, it cumulatively protected more than 100k potential victims’ accounts. China Merchants Bank’s online risk-control platform approved corporate credit volumes of nearly 600 billion yuan in 2025, up 44% year over year. With AI-assisted post-loan risk warnings, the lead time compared with traditional manual modes averages 42 days earlier.

In the inclusive finance sector, AI is破解ing the classic problems of “hard-to-get financing” and “expensive financing” for small and micro enterprises by analyzing alternative data. Many banks use AI models to integrate companies’ tax, invoices, supply-chain, and even water-and-electricity data—building a credit “profile” for small and micro enterprises that lack traditional collateral, enabling rapid credit granting.

Intelligent operations and customer service are the most direct expression of AI reducing costs and improving efficiency. China Merchants Bank’s intelligent assistant for more than 10,000 Golden Sunflower customer managers has become a smart companion in daily work. Ping An Bank uses generative AI (AIGC) to assist in creating marketing content; just this alone saved about 60 million yuan in expenses in 2025. In the operations back office, AI “digital employees” are taking over large quantities of repetitive work. Citic Bank has used AI to promote intensive processing of corporate account opening, information changes, and other businesses, increasing intensive operational effectiveness by more than 2 times.

“The success of AI in these areas is mainly because it solves the massive volumes of data that humans in traditional financial models struggle to handle, complex patterns that rules cannot fully cover, and the real-time response needs under high concurrency,” the aforementioned banking-industry research source analyzed. These mature applications form the bank’s AI “main base,” and their value is directly reflected in cost savings, risk reduction, and improved customer experience.

He believes that current applications are more about “optimizing existing processes,” and the next stage of competition will focus on how to use AI to “create new processes,” and even “create new businesses”—moving from “improving internal efficiency” to “creating external revenue.”

Overseas reality: breakthroughs from process optimization to value creation

While domestic banks focus on using AI to optimize internal processes and customer service, international financial giants represented by JPMorgan Chase are extending AI into more disruptive areas: investment decision-making itself.

In the venture capital (VC) and private equity (PE) fields, AI is reshaping the underlying logic of deal sourcing and due diligence. The traditional approach—relying on relationship networks and industry research (such as platforms like Wind and Bloomberg)—is being changed. For example, Sequoia Capital has long developed internal AI tools to automate the scanning of global startup data, academic papers, patents, and news. At fixed times each day, it sends the investment team an initial analysis brief to potential targets, improving the breadth and efficiency of deal screening.

In the client-facing wealth management and investment banking sectors, AI is moving from back-office assistance to front-office service. JPMorgan Chase applied for a trademark as early as 2023 for a product called “IndexGPT.” It uses generative AI technology to automatically analyze and select securities investment advisory tools based on the themes or focus areas input by clients. The model is built on a general large model foundation and trained using JPMorgan Chase’s unique large-scale private data such as macroeconomic and company research, with the goal of providing personalized investment portfolio advice.

In addition, in the lending business, using AI to provide more refined risk grading and pricing for clients is already a fairly mature practice overseas.

The aforementioned banking-industry research source interpreted that AI practices by overseas financial institutions reveal two key trends: first, AI applications are moving from “optimizing internal processes” toward “creating external value,” directly intervening in core value-creation areas such as investment advice and product design; second, leading institutions are using their unique, high-quality data moats (such as transaction data and deep research) to train vertical-domain large models, building new and difficult-to-replicate competitive moats. Compared with this, domestic financial institutions still have room to develop in using AI to directly drive investment decisions and provide in-depth intelligent investment advisory services—perhaps the high ground that will need to be tackled in the future.

Hidden obstacles ahead: the tests of data governance, AI hallucinations, and talent shortages

Beyond mature applications such as anti-fraud and intelligent customer service, the financial industry is cautiously pushing AI toward more cutting-edge and more core areas, trying to unlock new value and enabling AI to play the role of an “analyst,” or even a “junior decision-maker,” in more complex financial activities.

Daily Economic News reporters learned that in intelligent public-opinion analysis and market risk alerts, some institutions are already training AI to real-time capture and analyze large volumes of unstructured data such as news, research reports, social media, and even satellite images, in order to detect “signals” of risks that may affect the market or specific companies. For example, the “Oriental Brain” artificial intelligence platform of Orient Securities can process nearly 70k pieces of market public-opinion information per day on average, automatically identifying corporate entities and categorizing negative public-opinion.

In intelligent post-loan management and asset preservation, AI is being used for continuous, automated risk monitoring of existing loans. By analyzing companies’ operating data, judicial information, and changes in public opinion, models can give early warnings of potential risks, turning from passive response into proactive management. Some banks have tried using large models to assist in generating post-loan review reports, significantly shortening the writing time.

Even more disruptive exploration is happening in the core area of trading and investing. In the field of quantitative investing, beyond optimizing existing trading strategies, the more front-line exploration is to develop “virtual traders” that can autonomously learn market microstructure and independently execute portions of trading instructions. It has been reported that JPMorgan Chase has released its AI quant trading platform, supporting intelligent integration of high-frequency trading and multi-factor strategies. In client trading (such as foreign exchange and interest-rate derivatives trading), AI is also being studied to provide traders with real-time best price quotes and hedging-strategy suggestions.

However, despite a broad outlook, deep application of AI in the core areas of finance still faces constraints. Data governance, the issue of large-model “hallucinations,” and the shortage of compound talents are three hurdles that financial institutions must overcome.

First is the challenge of data governance. High-quality, standardized data is AI’s “fuel.” However, financial data involves highly sensitive personal privacy and commercial secrets, and it is often scattered across different business departments, forming “data silos.” KPMG experts point out that financial institutions commonly face challenges such as difficulty coordinating multi-source heterogeneous data and difficulty enabling internal data flow and sharing.

Second is the “hallucination” problem of large models and the risk of reliability. The “hallucination” issue inherent to large language models can be fatal in financial decisions that require zero mistakes. A researcher at Postal Savings Bank of China, Lou Feipeng, stated that if “hallucinations” occur in the risk-management domain, it may lead to the bank being unable to understand the logic of risks, and thus unable to take effective response measures.

Third is the shortage of compound talent and the pain of organizational change. Compound talent that both deeply understands complex financial business logic and is proficient in AI algorithms and engineering is extremely scarce. At the same time, traditional banking organizational culture—emphasizing rigor and a hierarchical system—creates deep tension with the agile development mode required by AI, which demands rapid iteration and tolerance for error-and-trial.

The aforementioned banking-industry research source summarized that future competition in the financial industry will be a contest of a comprehensive ecosystem: “technology—data—governance—talent.” Only institutions that can build high-quality data assets first, establish a trustworthy AI governance framework, and successfully transform their organizations and culture will win long-term advantages in this profound “intelligent-digital” revolution.

(Editor: Qian Xiaorui)

Keywords: