How AI Became the Operating System of China’s Investment Banks

Guotai Haitong put more than 260 AI systems into production, then began financing the companies building the technology it runs on.

At the World Artificial Intelligence Conference (WAIC) in Shanghai, Guotai Haitong laid out something more revealing than a product demo. Its systems ran across investment banking, wealth management, institutional services, research, compliance and corporate operations, one connected layer rather than a shelf of separate tools. The firm says more than 260 AI applications are now in production across six business areas. None of it is a pilot.

For years, banks treated AI as a feature: chatbots on the website, trials in the back office. That period is ending. The models are moving into the workflows that generate revenue, allocate expertise and control risk, which turns AI from something employees use into something the institution runs on.

Investment banks are knowledge factories. They gather information, turn it into judgment and sell that judgment through research, advice, underwriting and trading. Put a model inside each of those steps and you change the speed and cost of the whole line, from screening companies and preparing pitches to monitoring portfolios and reviewing communications.

The pressure is not confined to China. >Bloomberg reported in October that OpenAI had recruited more than 100 former investment bankers to train AI on the financial models behind restructurings and initial public offerings, aiming the technology squarely at the apprenticeship that has always produced Wall Street’s deal-makers.

A Bank Rewired Around AI

Guotai Haitong’s numbers describe a build, not an experiment. In April 2025, it won regulatory approval for a customer-facing large language model. By 2026, it had stood up a private environment linking models, computing capacity and proprietary data under a single system of governance. The 260-plus applications run on top of it.

Production changes what a firm learns. Every deployment teaches it something about data quality, model error, access controls and whether employees actually use it. Those lessons compound. Over time the plumbing and the habits, the unglamorous parts, become the asset a competitor cannot copy by buying the same software.

Lingxi is the part clients see. Inside the firm, its AI agents support more than 6,000 financial advisers and have handled about 388,000 client interactions. Other systems help researchers synthesize information, connect corporate clients to financing and treasury services, and let control teams watch service quality and compliance in near real time. A banker maps an industry faster. An adviser holds more accounts without dropping any. A research team tests more hypotheses before it is published.

That is the line between buying software and redesigning a bank. The models carry context across departments, hold institutional knowledge that used to leave when people did, and shorten the distance between a question and an action.

The Bank on Both Sides of the Boom

Guotai Haitong is also financing the industry whose technology it is installing. The firm says it has invested in 305 projects across AI, semiconductors and other frontier technologies and helped 107 companies list on Shanghai’s STAR Market. That puts it on both sides of the AI cycle, as a user of the technology and an intermediary directing capital toward the companies building it.

That role is becoming more important as technology companies look across mainland China and Hong Kong for capital. Yu Weijun, Business Director, President of the Investment Banking Business Committee and General Manager of the Investment Banking Department at Guotai Haitong Securities Co., Ltd., has examined the logic of A-share companies listing in Hong Kong and Hong Kong-listed companies returning to the A-share market.

Yu argues that reforms to A-share follow-on financing have made domestic funding channels more efficient. With H shares often trading at a discount, raising capital is no longer the main reason for an A-share company to seek a Hong Kong listing. The larger objective is to build an international capital platform and establish a presence across both markets.

The logic can also run in reverse. Hong Kong-listed AI companies returning to the A-share market may have greater scope for a valuation rerating, Yu said, while an A-share IPO can also provide support for the company’s H-share price.

The technology boom is also changing the composition of China’s listed markets. 

Technology companies have accounted for more than 30% of A-share IPO fundraising in each of the four years since 2022, while Hong Kong’s listing pipeline is concentrated in technology and healthcare.

As stronger conditions in the technology sector feed through to listed-company earnings, Zhang sees the potential for a “Davis double play” of earnings growth and valuation expansion. He also points to improving corporate profitability, household wealth reallocation and renewed overseas inflows as forces that could support a broader revaluation of Chinese equities.

For investment banks, the implication is direct. A larger technology sector creates more companies to finance, more businesses to research and more assets to distribute. It also raises the value of the AI systems banks use to identify prospects, assess risk and advise clients.

The New Cost of Competing

Banks have always competed on relationships, research and execution, and AI reaches all three. It surfaces prospects earlier, keeps advisers in more continuous contact with clients, and lets research teams work through a wider field of filings, calls and market signals.

Access to the models buys almost nothing on its own. The durable advantage sits in proprietary data, workflow design, governance and whether employees adopt the tools. Investment firms are already building AI-ready data systems and running research agents through multi-step analysis.

The scoreboard will read in efficiency, reduced risk and investment performance.

The risks move closer to the center too. A weak chatbot annoys a customer. A flawed model embedded in advice, compliance or deal work produces suitability, privacy and accountability failures at scale. The controls have to move as fast as the systems they govern.

For a bank, the strategic asset is the institution it rebuilds around the technology. Models are becoming widely available. The advantage lies in integrating them with proprietary data, regulated workflows and the habits of thousands of employees.

A firm already running AI in production has a head start. That head start is becoming the cost of admission.