Baidu’s transition from an internet-search company to an AI business is reaching a new financial test. Chief Financial Officer Henry He says the company’s AI-related operations could soon produce margins comparable to its legacy search business, an assertion that turns a familiar strategic promise into a measurable profitability question. In an interview reported by Bloomberg through The Edge Singapore, He said revenue linked to cloud computing, chips, and applications now makes up more than half of Baidu’s sales.
The claim is consequential because those businesses have very different cost structures from search advertising. Search historically benefited from a mature infrastructure and a business model in which incremental queries could be highly profitable. AI services require expensive computing capacity, continuing model work, product development, and customer support. If Baidu can demonstrate that cloud demand and broader deployment are improving returns on that spending, investors may begin to judge the company less as a search platform under pressure and more as a group of AI businesses with distinct economics.
The operative word, however, is “could.” He’s comments are a forecast, not a reported margin result. The company’s AI revenue category also bundles several activities, including cloud, chips, and applications. That makes the direction of travel clear while limiting what outsiders can infer about the profitability of any individual model, agent, or hardware line.
An AI Revenue Mix With Several Engines
He told Bloomberg Television that Baidu’s AI-driven revenue now exceeds half of total sales. That umbrella includes the company’s cloud services, Kunlunxin chip operation, and applications. Taken together, the figure suggests Baidu has already shifted a substantial portion of its commercial base beyond the advertising business that defined its earlier growth.
The company is trying to turn its stack into a connected system. Models can generate demand for cloud capacity. Cloud customers may need AI infrastructure, model-hosting services, or tools to deploy agents. Chips can offer an alternative or complement to third-party accelerators. Applications, including autonomous driving and consumer services, provide a way to turn technical capability into a service people can actually use.
Those linkages are central to Baidu’s investor narrative, but they also make financial analysis complicated. Strong cloud growth does not automatically prove that every AI product is profitable. A chip business can require different capital commitments than an application business. And a company moving quickly in several layers of the stack must decide how much infrastructure to build before revenue is visible.
Baidu’s own recent history illustrates the tension. EastFrontier reported in August that AI revenue was growing while advertising weighed on quarterly sales. He’s new comments push the argument beyond growth: the company now wants investors to consider whether the newer businesses can eventually match the economics of the old one.
Cloud Demand and the Cost of GPUs
He said new GPU clusters that host AI services could reach full cash payback within two to three years. That estimate rests on demand for cloud infrastructure and chips, and it should be treated as management’s view of future utilization rather than a company-wide forecast. The timing of a cash payback depends on equipment costs, customer contracts, electricity, utilization rates, and competitive pricing.
The question is especially acute in China, where large technology companies are expanding AI capacity while seeking to control capital expenditure. Baidu faces competition from Alibaba, Tencent, Huawei, and other providers that also sell cloud services, develop AI platforms, or support enterprise deployment. The pressure is not simply to build more capacity. It is to use that capacity well enough to earn an attractive return.
He argued that Baidu’s cloud revenue should grow faster than the industry average in coming quarters while spending remains lower than the market expects. That is again a management outlook. It does, however, reveal the financial discipline Baidu wants to emphasize. Rather than announcing unlimited investment in the largest possible foundation model, the company is signaling that it wants to allocate capital across models, chips, cloud, agents, and physical-world applications.
The chip component is part of that picture. Baidu is considering an independent listing for Kunlunxin, according to the Bloomberg report. The unit has been a recurring focus for the company as it tries to show it has a role in domestic AI infrastructure. EastFrontier previously covered Baidu’s intention to pursue a Kunlunxin listing, though any current valuation or listing timetable must be treated cautiously until formal filings provide an update.
A New Way for Investors to Value Baidu
On September 1, Baidu also gained dual-primary listing status in Hong Kong. The change could enable eligibility for Hong Kong’s Stock Connect program, allowing mainland investors to trade the company’s shares if the relevant conditions are met. That development comes as Baidu is asking the market to look beyond a single internet multiple.
He’s argument is that the company’s parts should be valued separately: cloud infrastructure, chips, autonomous driving, agents, and proprietary models each address different markets and carry different risks. That logic is common among companies trying to persuade investors that an older core business masks newer assets. It will only be persuasive if Baidu reports enough evidence that those assets can produce durable cash flow.
Baidu has also authorized a US$5 billion share-repurchase program and adopted a dividend policy, according to the report. Those measures can appeal to investors who want returns while the company spends on AI, but they also sharpen the capital-allocation challenge. Money used for repurchases or dividends is money not used to add computing capacity, fund model research, or support product distribution.
The next phase of Baidu’s AI transition will therefore be judged in practical terms. It is not enough for AI-related revenue to exceed half of sales if margins remain opaque or infrastructure commitments rise faster than demand. He’s forecast establishes a clear benchmark: AI profit margins approaching search. Whether that arrives soon will depend on customer adoption, cloud utilization, chip execution, and Baidu’s ability to make its many AI businesses reinforce one another rather than merely coexist.