In May of this year, Robin Li, founder of Baidu, announced a new industry metric at the "Baidu Create 2026" conference: DAA (Daily Active Agents).
His explanation was: Tokens may not represent the outcome; they represent cost, not revenue. To measure the prosperity of a platform and ecosystem, we should focus on DAA—how many agents are working for humans and delivering results.
Developers and investors at the event had mixed opinions. Some saw it as Baidu's "new ticket to the game," while others believed that introducing a new concept is not enough to define Baidu's future. Behind the debate lies an undeniable fact reflected in the financial report: Baidu's two growth curves are approaching each other rapidly.
On May 17, Baidu released its first-quarter 2026 results. General business revenue reached 26.0 billion yuan, a 2% year-on-year increase. Among this, AI business revenue was 13.6 billion yuan, accounting for 52% of Baidu's general business revenue, growing for several consecutive quarters. This also marked the first time that Baidu's AI revenue exceeded half of its total general business revenue. In the same period, traditional online marketing revenue was 12.6 billion yuan, down 22% year-on-year.
The flip side of this turning point is the cost. During the period, Baidu's net profit attributable to shareholders was 3.4 billion yuan, a year-on-year decline of more than 55%.
A senior executive at an AI company said, "What is most puzzling is Baidu's falling behind in the large model wave. It invested in AI earlier than anyone else. Many leaders of ByteDance's Seed and Qwen teams came from Baidu, and Baidu also accounts for more than half of the leading figures in autonomous driving companies... Search also has a natural synergy with consumer-facing AIGC applications. Yet Baidu has failed to secure a place in the top tier."
According to data, as of March 2026, the monthly active users (MAU) of AI-native apps in China reached 440 million. Among them, ByteDance's Doubao, Alibaba's Qwen, and DeepSeek had 345 million, 166 million, and 127 million MAUs respectively, ranking as the top three domestic AI-native apps, while Baidu's Wenxin app had fallen out of the top ten.
Baidu has achieved a structural breakthrough led by A, but it faces short-term profit pressure. Baidu needs a more decisive "gear shift": using profits from the old track to pave the way for growth on the new track.
Since the beginning of this year, Baidu has carried out a series of product planning and organizational changes for its Mobile Ecosystem Group (MEG), clearly integrating product lines around a user-centric approach. The standalone Wenxin app now works more closely with internal entry points such as Baidu Search and Baidu Library. Correspondingly, Baidu has also made a series of organizational and personnel adjustments.
For more than two decades, search has been Baidu's foundation. But in the AI era, the way users access information has been reshaped. The shift from "search-click-read" to "ask-question-AI-gives-answer-directly" may seem similar to search. However, the real challenge lies in Baidu's more conservative strategic choices despite its deep AI technology accumulation, and the internal conflict between AI applications and Baidu's cash cow—paid ranking.
At the end of April this year, a Baidu employee who wished to remain anonymous admitted that for daily information gathering, "I use Doubao and DeepSeek more often." To this day, "Baidu's ability to build consumer-facing products is not on the same level as ByteDance's."
A head of vertical large model development at a leading company, analyzed from an industry perspective: "Robin Li made it clear from the beginning of building large models: no open source. This choice has both pros and cons."
Open source vs. closed source essentially reflects different companies' judgments on technology paths and business models. In the view of the aforementioned source, the open-source strategy was key to the rapid iteration of the AI industry over the past two years, while Baidu chose a more closed path from the start. Closed source can protect core technology and build differentiation, but it does suffer in terms of ecosystem expansion speed.
A deeper pressure lies in organizational ecosystem coordination. An industry insider said, "Every layer of Baidu's departments has its own planning. It feels like many teams are stepping on each other's toes—serious product homogenization, scattered resources, and fragmented user experience."
In contrast, within ByteDance's system, Doubao, Douyin, Jimeng, and others have formed tight synergies from tools to platforms.
In contrast to the consumer side, Baidu AI Cloud performs well on the B2B side.
According to third-party reports, in Q1 2026, Baidu's market share in the self-developed GPU cloud market reached 40.4%, ranking first in China. In China's AI application public cloud service market, its share rose to 30.7%. Much of the AI demand from automakers, banks, and state-owned enterprises lands on Baidu's platform. Particularly in the automotive and financial sectors, customers highly recognize Baidu's technical execution capabilities.
However, of the 13.6 billion yuan in AI revenue, smart cloud accounted for nearly 65%, while application-side revenue has declined rather than increased. This suggests that Baidu's current role in the AI value chain is more of an infrastructure provider than a "gold rusher." From another perspective, this reflects Baidu's strategic choice of "building the road first, then running the cars": solidifying the computing power foundation before naturally extending to the application layer.
In terms of technological accumulation, Baidu is not lagging. Public data shows that as of the end of 2024, Baidu had filed over 27,000 AI patent applications globally (22,000 in China, with 12,000 granted), covering full-stack AI fields such as deep learning, natural language processing, and computer vision.
In terms of its self-developed Kunlun chips, Baidu's long-term commitment is also evident. The third-generation Kunlun chip is already running stably on 10,000-card clusters, providing an independently controllable computing power foundation for Wenxin large model training and inference, and will further optimize Baidu's cost advantage over the long term.
But the next test will be whether the high growth of B2B orders can sustain Baidu's overall "AI narrative."
External market changes have also catalyzed a dramatic restructuring of Baidu's organization.
According to media reports, at the end of 2025, Baidu launched its "largest-ever" adjustment, reducing its total employee count from 41,300 in 2022 to 35,900. Positions in AI Cloud and autonomous driving were prioritized, with resources further tilted toward AI.
The most notable move was the establishment of the BMC in May this year. Earlier, in November 2025, Baidu had already set up two parallel departments—the Base Model Unit (BMU) and the Applied Model Unit (AMU)—both under the direct oversight of Robin Li.
According to public information, BMC members are young researchers with a deep understanding of large models. They will coordinate the development of Baidu's large model base models and applied models, reporting directly to Robin Li. The BMC adds a coordination layer above the BMU and AMU, marking the formalization of this organizational structure since the creation of the new model R&D departments last year.
These actions all point to the same goal: shortening decision-making chains and enabling frontline technical judgments to reach top decision-makers faster.
Compared to startups, the most difficult balance for established giants when innovating is often the "legacy burden" of the organization.
A senior executive at a large model company who led teams for over a decade at a major tech firm candidly said, "Currently, the companies doing well in large models tend to have younger employees. How do you retain them? You can't manage them in a traditional, step-by-step, assembly-line fashion. Not long ago, Justin Lin Junyang left Alibaba because he couldn't stand that kind of management style—so he left."
Alibaba's DAMO Academy faced similar structural difficulties and eventually scattered like stars.
"DAMO Academy's early positioning was innovation, but later, driven by cost and profitability pressures, it had to be self-sustaining. Ultimately, business logic drives everything."
Thus, the ultimate driver and outcome still point to the resolute determination of the highest decision-makers.
From establishing the Institute of Deep Learning in 2013, to proposing "All in AI" in 2017, and now to the full-scale implementation of large models and agents, Robin Li's commitment to AI has never wavered for more than a decade. This long-termism is rare among Chinese internet entrepreneurs.
Since the beginning of this year, market education for the agent ecosystem has been completed. Commercialization paths are shifting from consumer-side consumption to coding-based payments, with answers still being explored.
"No one has really found a truly high-frequency commercial growth point yet. Baidu's pressure is that others are rushing to capture entry points first," said an employee who has worked at Baidu for many years. Just as ByteDance captured the Douyin entry point, Doubao thereby gained continuous iterative data and user feedback.
Baidu's opportunity lies more in the combination of its search entry point and agent products. For example, DuMate is embedded directly into the Baidu App, covering hundreds of millions of daily active users. Moreover, Baidu has 20 years of accumulated expertise in Chinese data processing, search logs, and knowledge graphs, which can still form unique competitive moats.
"Qianwen does open source best, with very fine-grained data—such as PDFs and web pages broken down in detail. Zhipu's underlying data is also good," the R&D head noted. "The core remains strategy selection; distillation alone has limited value." With Alibaba adjusting its open-source strategy and Doubao planning to implement paid plans, the degree of open source and commercialization strategies among domestic large model companies has become a pressing issue.
But looking at the financial report, the inflection point in Baidu's revenue structure has already appeared, and the window of opportunity is becoming concrete and urgent. Baidu needs to think clearly not only about its technology and business direction, but also about what kind of organization, culture, and profit-sharing structure it needs to execute its strategy.
In this long-distance race of AI, how will Baidu fulfill its long-standing technological conviction? As Robin Li said in his speech at the conference, "We are not conducting an experiment; we are paving a road."
But for Baidu standing at a crossroads, the most difficult hurdle has never been inside the laboratory.
(Source: China Entrepreneur)
Related News:
Deepline | End of GPU-only strategy? Intel bets big on agentic AI
Comment