Baidu Crosses the AI Rubicon: Revenue Leadership Achieved, Consumer Relevance at Risk
Baidu has reached a structural inflection point — AI revenue now accounts for the majority of its core business for the first time, yet the transition is exacting a severe near-term toll, with net profit falling more than half and its flagship consumer AI product sliding out of China's top ten.
First-quarter 2026 results filed May 17 showed core revenue of RMB 26 billion (US$3.61 billion), up just 2% year-on-year. AI revenue contributed RMB 13.6 billion (US$1.89 billion), representing 52% of the total and marking the first time that figure has crossed the halfway threshold across consecutive quarters of expansion. The counterweight: traditional online marketing revenue fell to RMB 12.6 billion (US$1.75 billion), a 22% decline. Net profit attributable to shareholders dropped to RMB 3.4 billion (US$472 million), a decline exceeding 55% year-on-year.
Introducing DAA: Founder Robin Li Reframes the AI Metric War
At the Create 2026 Baidu AI Developer Conference in May, founder and CEO Robin Li unveiled what he positioned as a new industry benchmark: DAA, or Daily Active Agents — the number of AI agents actively completing tasks and delivering results for users on any given day. Li argued that token consumption, the prevailing measure of large language model traction, captures cost rather than value creation. "What matters," he told attendees, "is how many agents are working for humans and delivering outcomes."
The concept drew a divided response from developers and investors present. Proponents called it Baidu's "new boarding pass" for the agent era; skeptics dismissed it as conceptual rebranding in lieu of concrete user growth. The underlying tension is real: Baidu needs a narrative framework that explains why infrastructure-level AI investment will eventually translate into consumer-facing dominance — a case its current app metrics make difficult to argue.
Consumer AI Metrics Reveal a Competitive Gap Baidu Cannot Paper Over
Data from QuestMobile as of March 2026 illustrates the scale of Baidu's consumer AI challenge. China's AI-native app ecosystem has reached 440 million monthly active users. ByteDance's Doubao leads with 345 million MAU, followed by Alibaba's Qianwen at 166 million and DeepSeek at 127 million. Baidu's Ernie app has dropped entirely out of the top ten.
The irony is structural. Baidu was among the earliest institutional investors in deep learning in China, establishing its Deep Learning Research Institute in 2013 and declaring an "All in AI" strategy in 2017. Multiple founding engineers at ByteDance's Seed AI lab and Alibaba's Tongyi Qianwen team trace their careers to Baidu. The company holds more than 27,000 cumulative AI patent applications globally as of end-2024, covering deep learning, natural language processing, and computer vision across the full stack. Yet its consumer product execution has lagged materially behind leaner, faster-moving competitors.
Industry insiders attribute part of the gap to an early strategic choice: Li explicitly rejected open-source development when Baidu launched its large model program. Closed-source development protects proprietary technology and can build differentiated moats, but it sacrificed the ecosystem velocity that open-source iteration — exemplified by Alibaba's Qwen releases — generated over the past two years. A senior R&D executive at a competing model company told China Entrepreneur Magazine: "The open-source strategy was the key to rapid iteration across the industry. Baidu chose a more closed path from the start. The tradeoff in ecosystem expansion speed is real."
Cloud Dominance Provides a Floor, But Positions Baidu as Infrastructure, Not Application Layer
Where Baidu retains clear leadership is in enterprise cloud. Third-party data for Q1 2026 shows Baidu holding a 40.4% share of China's self-developed GPU cloud market — the largest of any domestic provider — and a 30.7% share of China's AI public cloud services market. Automotive OEMs, state-owned financial institutions, and central government enterprises represent the core customer base, sectors where Baidu's technical delivery track record commands premium positioning.
However, the revenue composition within that RMB 13.6 billion AI figure is revealing: intelligent cloud accounts for approximately 65% of AI revenue, while application-layer revenue has declined. This positions Baidu primarily as an infrastructure provider rather than a direct beneficiary of AI application monetization — a "build the road, then drive the car" strategy, as insiders describe it. The risk is that cloud infrastructure margins face commoditization pressure as domestic GPU supply expands, while the application-layer value capture accrues to companies with stronger consumer distribution.
Baidu's proprietary Kunlun chip program offers a partial hedge. The third-generation Kunlun chip is now running stably in clusters of 10,000 units, providing the compute substrate for Ernie model training and inference while building a long-term cost advantage independent of foreign semiconductor supply chains.
Organizational Restructuring Signals a Race Against Internal Entropy
The competitive pressure has triggered what sources describe as Baidu's most aggressive internal reorganization in its history. By end-2025, total headcount had been reduced from 41,300 in 2022 to approximately 35,900, with AI cloud and autonomous driving roles explicitly protected while legacy functions were rationalized.
The restructuring has continued into 2026 at accelerating pace. In January, Baidu Library and Baidu Netdisk were merged into a new Personal Super Intelligence Group (PSIG), with Vice President Wang Ying reporting directly to Li. In April, Baidu abolished its long-standing alphabetical employee grade system in favor of a unified numerical framework — a structural signal intended to dissolve the boundary between technical and management career tracks.
The most consequential organizational move came in May: the establishment of the Baidu Model Committee (BMC), a coordinating body composed of young researchers with deep large-model expertise, sitting above both the Foundation Model Research Division (BMU) and the Application Model Research Division (AMU) established in November 2025. All three structures report directly to Li. The design compresses the decision chain between front-line technical judgment and the CEO — a response to criticism that Baidu's multi-layered product hierarchy, encompassing Ernie App, Wenxiaoyuan, Ernie Yuger, and multiple overlapping tools, had created internal resource fragmentation and user experience inconsistency.
The personnel cost has been significant. Over the past year, departures have included Zhao Shiqi, Vice President and head of Baidu Search; the head of Baidu Search AI products; and Chen Yingmei, widely identified as Li's first management trainee. The talent drain reinforces a concern Li himself articulated publicly: "Baidu cannot only serve as a Whampoa Military Academy for AI talent."
Agent Products Define Baidu's Offensive Posture for the Second Half of 2026
At Create 2026, Li launched four agent products simultaneously: general-purpose agent DuMate, coding agent Miaoda 3.0, digital human agent Baidu Yijing, and decision-making agent Famou 2.0. DuMate has achieved state-of-the-art performance on multiple international Agent Benchmark evaluations and is capable of operating software, processing files, and integrating enterprise systems. Baidu Yijing is positioned as the world's first full-scenario multi-agent digital human platform.
The strategic logic is coherent: DuMate embedded directly within Baidu App reaches hundreds of millions of daily active users, giving Baidu a distribution channel that pure-play AI startups cannot replicate. Combined with two decades of Chinese-language data processing, search logs, and knowledge graph construction, Baidu retains proprietary data assets that remain competitively defensible even as model capabilities converge across the industry.
The challenge is timing. Agent-based commercial monetization remains nascent across the industry. ByteDance's Doubao is transitioning toward paid coding features; Alibaba has recalibrated its open-source strategy. The high-frequency commercial use case that justifies sustained consumer agent spending has not yet been clearly identified by any major player. As one long-tenured Baidu employee told China Entrepreneur Magazine: "Nobody has found the truly high-frequency monetization growth driver yet. Baidu's pressure is that everyone else is racing to occupy the entry point first."
Impact Assessment: Three Risks Investors Should Monitor
Revenue cliff steepness. Online marketing revenue — still 48% of core sales — is declining at 22% annually. If AI cloud growth moderates as enterprise procurement cycles extend, the crossover arithmetic deteriorates faster than Baidu's current guidance implies.
Consumer application market share recovery. Ernie's exit from the top-ten AI app rankings is not merely a vanity metric. Consumer distribution drives the feedback loops — user data, behavioral signals, iterative improvement — that compound model quality over time. ByteDance's Doubao is demonstrating this flywheel in real time, evolving from a chat tool into a search-adjacent entry point.
Organizational execution under dual pressure. Baidu is simultaneously running a large-scale workforce reduction, a top-to-bottom grade system overhaul, and a new model governance structure — all while competing for AI talent against better-capitalized or faster-moving rivals. Alibaba's DAMO Academy provides a cautionary precedent: innovation mandates eroded by profitability requirements produced structural fragmentation. Whether Li's direct-reporting model for BMC avoids the same fate will be visible in product release cadence by Q3 2026.
Related Coverage: