Since the large language model wave began, Baidu (BIDU) has been one of the earliest Chinese internet giants to bet on artificial general intelligence and launch its own proprietary large model. It has built a complete full-stack AI layout spanning computing power, foundation models, intelligent cloud, and autonomous driving, and was once regarded by the market as a benchmark player in China's domestic AI sector.
However, the gap between the ideal narrative and actual business performance has continued to widen. On one side, there are persistently high R&D expenses, GPU computing procurement, and large model iteration costs; on the other, the traditional advertising core business is under pressure, the revenue structure of new AI businesses is imbalanced, and user penetration for end products has fallen short of expectations. The market is gradually realizing that Baidu's AI story has not yet translated into stable, high-margin commercialized cash flow, and the stock price has long been grinding along a volatile downward trajectory.
As of market close on November 3, 2026, Baidu (BIDU-W) Hong Kong shares stood at HK$74.7 per share, sitting at a historically low level.
Double Blow to Stock Price and Earnings: Heavy Bet on Full-Chain AI Layout, Yet Growth Remains Weak and Shares Keep Sliding
Baidu's stock price has shown a long-term and pronounced downward trend. After hitting a peak of HK$256.6 per share in 2021, the stock entered a prolonged correction. After a recent rebound, the share price has once again retreated and is currently hovering around HK$75, placing it in the historically low range for this chart period. (Source: Wind Information)
The reason Baidu, which entered the AI spotlight early, has failed to win investor favor and has repeatedly faced sell-offs lies in the growing pains of its transition and the reshaping of its valuation logic. On one hand, business performance has been lackluster. As the core foundation, the traditional search advertising business has struggled for growth under macroeconomic pressure and competition from short-video platforms, while heavily invested new AI businesses, despite impressive growth rates, have not yet generated enough profit support. High computing power and R&D costs have severely weighed on overall profitability, leaving the old and new growth engines in an awkward handoff period. On the other hand, market expectations are insufficient. Facing extremely intense competition in AI large models, investors lack confidence in the prospects and monetization loop for Baidu's AI technology commercialization. Unable to see a clear profit path, the capital market has grown fatigued with its "AI story" and is no longer willing to assign excessively high valuations for upfront investment.
First, looking at the company's business performance: in the first half of 2026, Baidu generated revenue of RMB 63.4 billion, down 2.71% year on year. Net profit attributable to shareholders was RMB 5.764 billion, plunging 62% year on year. On a quarterly basis, second-quarter revenue was RMB 31.3 billion, down 4% year on year; net profit attributable to shareholders was RMB 2.319 billion, down 68% year on year. (Source: Company announcements)
By business segment, in the first half of 2026, Baidu's traditional core online marketing business faced significant downward pressure, with revenue falling sharply from RMB 34.791 billion in the same period of 2025 to RMB 28.138 billion, a year-on-year decline of 19.12%. Its share of total revenue also dropped to 44%, reflecting how the traditional search and feed advertising core is being squeezed by both macroeconomic volatility and fierce market competition. At the same time, Baidu's AI business showed clear "structural imbalance in momentum": although overall core new AI businesses grew 36% year on year, and the underlying intelligent cloud infrastructure, especially GPU cloud revenue, surged 230%, the true monetization core, "AI applications," generated only RMB 5 billion in revenue, with year-on-year growth of just 1%, exposing serious bottlenecks in upper-layer application deployment and commercialization. Meanwhile, the traditional online marketing business fell nearly 19% year on year, evaporating about RMB 6.6 billion, meaning that the roughly RMB 6.9 billion increment from new AI businesses was almost entirely used to fill the gap left by the shrinking old business. This disconnect, where "infrastructure runs fast while applications cannot keep up," leaves the overall AI business lacking strong momentum to drive a leapfrog growth for the company and makes it difficult to fully reverse the pressure on performance.
For Baidu, the essence of the dilemma is that long-term investment consumes cash flow, yet high-value-added, scalable, high-margin AI-native applications have not emerged. The computing power business is more like providing infrastructure to other AI companies and is unlikely to form an exclusive moat. Meanwhile, the consumer-facing applications and enterprise office-native agents that could truly bring long-term premiums and strengthen user mindshare have yet to produce a dominant product. The market may no longer be willing to pay simply for "large model technology," and business performance and commercialization delivery have become the core yardstick determining valuation.
AI Layout: Ernie Bot's Consumer Voice Remains Weak, Enterprise Office Track Falls Notably Behind Tencent WorkBuddy
As the core product of Baidu's AI strategy, Ernie Bot carries the dual commercialization mission of consumer traffic and enterprise services, and it is also the key window through which the market observes Baidu's AI deployment capabilities. But after more than two years of iteration, Ernie Bot's commercialization report card is clearly below the market's initial expectations. Its product positioning has been adjusted several times, its entry points have been integrated repeatedly, and it has never firmly established itself in the top tier of the consumer market. In domestic general large model consumer monthly active user rankings, Doubao and Tongyi Qianwen have long held leading positions, while Ernie Bot's independent app and web user scale, activity, and time spent are all clearly lagging. It has long been absent from mainstream AI application rankings and has not formed the mindshare of a national-level AI assistant.
Behind the consumer setback may be Baidu's wavering product strategy. In the early days, Ernie Bot was promoted independently, but later it was merged with the Ernie assistant inside the Baidu app, while the company simultaneously launched multiple AI office products such as Baidu Dazi and Kuku AI, spreading resources across multiple lines without concentrating on building a unified super entry point. Compared with competitors, ByteDance's Doubao relies on short-video traffic, and Alibaba's Tongyi Qianwen penetrates users through the DingTalk ecosystem, while Ernie Bot is highly dependent on Baidu search for traffic. The efficiency of converting search users into AI assistant users is not ideal, and consumer paid conversion and membership subscription revenue remain limited, making it hard to become a second growth curve.
In the enterprise office AI track, which has the greatest commercialization potential, Baidu's gap is even more obvious. Tencent WorkBuddy relies on the natural collaborative ecosystem of WeCom, Tencent Docs, and enterprise contact lists, deeply embedding itself into daily enterprise approvals, meeting minutes, document collaboration, and team knowledge base workflows. It leads by a wide margin in desktop office AI visits and office agent monthly active user rankings. QuestMobile data shows that WorkBuddy has more than 32 million monthly active users and 85.34 million desktop monthly visits, far ahead of peers. Although Baidu has launched a product matrix including Baidu Dazi and Kuku AI in the office track, Baidu Dazi has only 9.77 million monthly active users, and the gap in desktop interactions is even larger. It falls far short of WorkBuddy in enterprise customer procurement, team collaboration scenarios, and native workflow embedding.
When enterprises purchase AI products, they first look at whether the products can connect with existing office systems, whether they can accumulate internal knowledge bases, and whether they can adapt to organizational permissions. Tencent WorkBuddy is naturally embedded in the WeCom ecosystem, requiring no extra registration or account switching for employees and is ready out of the box. Baidu's office agent products, by contrast, require enterprises to deploy separately and connect internal systems, resulting in higher implementation costs and weaker customer stickiness. Although the Ernie Qianfan platform has a large number of developer customers and considerable enterprise API call volume, most of these are broad-industry calls and outsourced development needs, rather than standardized, high-renewal office productivity products.
Overall, Baidu's AI commercialization presents an imbalanced pattern: strong in computing power, weak in applications; strong in enterprise infrastructure, weak in office applications; and weak in consumer traffic. Computing power leasing is a stable increment, but its ceiling is low and competition is intensifying. The high-value AI applications the market truly expects, whether general assistants for the public or office agents for enterprises, have not formed sufficient competitiveness. Ernie Bot's predicament also reflects Baidu's shortcomings: it possesses top-tier foundation model research and development capabilities, but lacks a product ecosystem that can deeply bind users and embed into organizational workflows, making it difficult to smoothly convert technological advantages into commercial advantages.
Comments