First-Mover Advantage Fades: Baidu’s AI Took the Earliest Biggest Bets Yet Falls Behind — An In-Depth Review of the 2026 Industry Landscape

Among China’s major internet conglomerates, Baidu is undoubtedly the earliest and largest investor in artificial intelligence. Back in 2013, Baidu Research built a top-tier domestic AI research team. In 2017, it fully rolled out its All-in-AI strategy, advancing two core pillars: the Apollo autonomous driving platform and the Ernie large language model. In 2023, it launched Ernie Bot, China’s first general large model comparable to ChatGPT, firmly securing its position as the leader of domestic AI at that time. Rival analysts widely believed Baidu would maintain a long-term dominant edge, bolstered by search-derived data reserves, self-developed chips, and a full-stack technical framework.
By June 2026, however, the market landscape has undergone a complete reversal. ByteDance’s Doubao, Alibaba’s Qwen, and Tencent’s Hunyuan form a stable top tier of big-tech AI players, while independent unicorn DeepSeek emerges as a disruptive lone wolf competitor. In stark contrast, Baidu, the pacesetter that got a years-long head start, sees comprehensive declines across key metrics: its C 端 monthly active users drop out of the industry’s top ten, alongside falling API call volumes, shrinking developer ecosystems, and slower commercial growth. Its once-unassailable first-mover edge has nearly evaporated. Why did Baidu, which poured unprecedented capital into AI earlier than everyone else, get overtaken by multiple rivals within just three years? A breakdown of Baidu’s strategic moves, missteps, and underlying weaknesses reveals the core logic reshaping China’s AI sector.
I. Baidu’s Unmatched Early Investment & Layout
A timeline review confirms Baidu’s AI preparations outpaced all competitors by a wide margin, with leading investments in talent, capital, and technology from 2017 to 2022.
On talent recruitment, Baidu recruited world-renowned AI scholars including Andrew Ng, Xu Wei, and Yu Kai in 2017, assembling a hundreds-strong deep learning team — the only Chinese firm with elite foreign AI leadership at the time. Meanwhile, ByteDance and Alibaba operated small-scale AI labs, and Tencent’s AI team was less than one-third Baidu’s size. For hardware, Baidu began developing Kunlun chips, building a fully self-reliant stack spanning chips, the PaddlePaddle deep learning framework, Ernie large models, and end-user applications, aiming to create an “Android-style” ecosystem for AI. Its dual-track business roadmap covered autonomous driving and general large models: Apollo launched its open platform in 2017, when domestic EV startups were still in infancy, granting Baidu a multi-year lead. On the large model front, the first iteration of Ernie debuted in 2019 with years of iterative refinement ahead of rivals, and Ernie Bot’s March 2023 launch made it China’s first generative large model available to mass consumers.
Financial outlay was equally aggressive. For multiple consecutive years, Baidu allocated over 20% of its annual revenue to AI R&D, exceeding 10 billion RMB in annual AI spending for both 2023 and 2024, far surpassing dedicated AI budgets at Tencent and Alibaba. Drawing on over two decades of search operations, Baidu possessed an unparalleled corpus of Chinese web text and question-and-answer training data, equipping the Ernie series with native strengths in search-augmented generation and real-time information retrieval. In the first half of 2023, Ernie Bot monopolized the traffic boom for domestic large models, amassing over 300 million registered users within months. Baidu Intelligent Cloud simultaneously secured a flood of enterprise orders for private deployment and custom APIs, fueling market confidence that Baidu would build an insurmountable technical and data moat.
Yet robust theoretical fundamentals failed to translate into sustained market competitiveness, with downward momentum accelerating from late 2024. Q1 2026 QuestMobile data places Doubao at 345 million monthly active users, Qwen at 166 million, and DeepSeek at 127 million to claim the top three spots. Ernie Bot’s MAU has contracted sharply, falling outside the top ten native AI apps. In API market share, Alibaba’s Qwen accounts for 32.1%, ByteDance’s Doubao 21.3%, DeepSeek 18.4%, while Baidu’s Ernie captures less than 10%. The gap in developer ecosystems continues to widen.
II. Stacked Critical Flaws That Eroded Baidu’s First-Mover Edge
1. Erratic Strategic Shifts: Internal Conflict Between Closed-Source and Open-Source Roadmaps
This stands as Baidu’s most pivotal decision-making failure. In 2024, Robin Li publicly insisted open-source large models were a “waste of resources,” doubling down on closed-source paid licensing as the sole viable profitable model. Baidu locked all Ernie model weights, banning commercial secondary fine-tuning by third-party developers, directly driving legions of independent programmers and small technical teams toward open alternatives like DeepSeek and Qwen. At that stage, DeepSeek had rapidly built a global developer base via fully open, free-for-commercial-use weights, while Alibaba also released multiple open Qwen versions, leveraging ecosystem scale to accelerate model iteration.
Crushed by market pressure, Baidu executed an abrupt U-turn in 2025, announcing open-sourcing for the Ernie 4.5 lineup and full free access for C 端 users — a complete reversal of prior strategy. It first alienated developers with closed policies, then rushed open-sourcing long after the critical window for ecosystem cultivation closed. Pricing was equally inconsistent: Baidu rolled out a USD8.5 monthly premium tier in 2023 to pioneer paid AI services, only to hastily slash rates and run limited-time free promotions amid low-cost competition. Unstable pricing undermined long-term trust among enterprise clients and developers. By contrast, DeepSeek maintained unwavering direction from inception: Mixture-of-Experts architecture to cut inference costs, permanently low API pricing, and full open-source commercial licensing. ByteDance and Alibaba also locked in stable playbooks early: free C 端 access to capture traffic, enterprise service revenue for profits, and open-source releases to recruit developers, with no internal strategic friction.
2. A Thin Core Traffic Ecosystem Limits AI Application Conversion
ByteDance, Alibaba, and Tencent operate super traffic foundations Baidu cannot match. ByteDance funnels massive short-video and content traffic from Douyin and Toutiao straight into Doubao, naturally retaining creators, students, and lower-tier market users with minimal customer acquisition costs. Alibaba embeds Qwen within Taobao, Alipay, and Amap, integrating AI natively into e-commerce, mobility, and local transaction scenarios. Tencent leverages WeChat’s 1.4 billion monthly active users and WeCom to weave Hunyuan AI across social, office, and gaming infrastructure. Their AI products function not as standalone apps, but as embedded capabilities within mature, sprawling business ecosystems, enabling seamless scenario rollouts.
Baidu’s only core asset remains search, whose mobile traffic has declined year over year, with Baidu failing to gain traction in high-frequency verticals including short video, social media, and e-commerce. Ernie Bot operates as an isolated app with no secondary large traffic channel to sustain user growth. Integration between search and large models proceeded sluggishly, as Baidu delayed deep Ernie embedding onto its search homepage, wasting its exclusive search-augmentation technical advantage entirely. While Doubao sustained stable daily active user counts in the tens of millions powered by short-video ecosystems, Ernie’s DAU flatlined at a few hundred thousand, permanently ceding consumer brand recognition.
3. Burdens of Full-Stack Self-Development Dilute Resources Across Multiple Fronts
Baidu’s commitment to end-to-end in-house development — spanning chips, frameworks, large models, consumer apps, and autonomous driving — promised full technological autonomy but split massive capital and manpower across fragmented priorities. Mass production of Kunlun chips faced delays, with subpar yield rates and higher costs compared to mature off-the-shelf alternatives. Though PaddlePaddle ranks as China’s top domestic deep learning framework, its third-party plugin compatibility and community ecosystem lag behind open-source rivals. Apollo held an early lead in autonomous driving, yet most automakers opted for in-house R&D, relegating Baidu to a marginal tech supplier unable to generate large-scale cash flow to reinvest in large model refinement.
Heavy spending across multiple capital-intensive lines drained funds needed for continuous GPU investment for large model training. Peer competitors adopted leaner resource allocation: ByteDance concentrates solely on large models and C 端 applications with on-demand scalable compute procurement; Alibaba balances heavy and light assets by pairing Qwen with cloud services; DeepSeek launched as a lean startup laser-focused on optimizing model inference efficiency, with no chip or hardware overhead, channeling all resources into performance and cost tuning for far faster iteration speeds than Baidu.
4. Rigid Organizational Structures Slow Internal Collaboration & Market Responses
Baidu’s AI divisions operate in silos: Shen Dou oversees Intelligent Cloud, handling B 端 Ernie orders; Ping Xiaoli manages Ernie’s consumer product line; the autonomous driving team runs independently. Data, user insights, and technical infrastructure are poorly interconnected across three branches with conflicting KPIs — cloud teams prioritize order profitability, C 端 units chase user volume, and Apollo targets technical benchmarks, preventing unified coordinated action. When Doubao and DeepSeek exploded in growth in late 2024, Baidu’s internal teams suffered from complacency as an early market leader, delaying tactical adjustments by a full quarter. By the time Baidu rolled out subsidies and ecosystem support, user and developer brand loyalty to competitors had already solidified.
ByteDance’s flat organizational structure grants the Doubao team high autonomy, enabling agile tweaks to iteration, marketing, and ecosystem policies. As an independent startup, DeepSeek boasts an ultra-short decision-making chain, executing two rounds of steep price cuts between May and June 2026 to rapidly capture global developer share, leveraging the agility of a smaller enterprise.
5. Conservative Commercialization Misses Opportunities with Small Businesses and Global Expansion
In its early commercial push, Baidu prioritized high-priced private deployments for large government and enterprise clients, dismissing low-margin API demand from small developers and sole proprietors. Yet the market’s fastest growth segment lies with millions of micro-teams, freelance programmers, and cross-border entrepreneurs — a space fully captured by DeepSeek’s ultra-low-cost APIs. In May–June 2026, DeepSeek cut inference pricing to global rock-bottom levels, winning mass adoption among small-to-medium enterprises across Europe and North America and outpacing all domestic big tech in overseas expansion speed. Baidu moved far slower on global rollout, lagging in multilingual optimization and international compliance, with offshore API volumes less than one-third of DeepSeek’s.
Baidu also rushed to launch consumer paid tiers prematurely, imposing fees before mainstream users built willingness to pay and alienating mass audiences. In contrast, Doubao and Qwen retain free core functionality, monetizing via premium memberships and custom enterprise packages. Their large free user bases supply massive interaction data to refine models, creating a positive iterative feedback loop Baidu failed to establish.
III. Four Distinct Competitor Strategies Squeeze Baidu’s Market Space
ByteDance (Traffic-Ecosystem Model): Rather than competing purely on technical parameters, ByteDance leverages Douyin’s massive traffic pool to dominate consumer user numbers, embedding AI into content creation, video editing, copywriting, and livestream workflows. Vast user interaction data iteratively boosts multimodal and conversational model performance, following a mass-market, accessible entertainment-focused path.
Alibaba (Balanced Full-Stack Model): Qwen advances evenly across open-source distribution, cloud services, and consumer scenarios, with closed loops within Taobao and Alipay. It operates three synchronized revenue streams: government/enterprise contracts, small developer open-source access, and C 端 users, delivering the industry’s most stable revenue structure and top API market share.
Tencent (Scenario-Bound Model): Hunyuan AI tightly integrates with WeChat’s ecosystem, corporate office tools, and gaming. Tencent does not chase blanket national traffic, instead capturing high-margin verticals including private-sector enterprise services and gaming AI.
DeepSeek (Cost-Performance Lone Wolf): Unburdened by big-tech traffic overhead, DeepSeek competes purely via strong model performance, extreme price competitiveness, and open-source ecosystems to win global developers and SMEs, with code and long-text reasoning capabilities ranking among the world’s elite and superior overseas traction compared to domestic giants.
These four differentiated strategies occupy distinct profitable niches, compressing the general large model market Baidu once dominated entirely. Baidu lacks ByteDance’s traffic, Alibaba’s commercial transaction scenarios, Tencent’s private social ecosystem, and DeepSeek’s lean, low-cost open-source flexibility, trapping it in a squeezed middle position with pressure from all sides.
IV. Baidu’s Remaining Assets and Long-Term Limitations
Objectively, Baidu retains irreplaceable core strengths. First, its search real-time data powers unique search-augmented generation capabilities no rival can replicate. Second, PaddlePaddle and Kunlun form China’s only fully domestic end-to-end AI hardware-software stack, granting exclusive bidding advantages for government and state-owned enterprise projects requiring secure, localized technology. Third, the full-modal Ernie 5.0 model still ranks among China’s top-performing large models, sustaining steady revenue from large private deployment orders. Baidu recorded 13.6 billion RMB in AI-linked revenue for Q1 2026, maintaining a solid B 端 intelligent cloud base — albeit with far slower growth than peers.
Yet returning to the top tier poses steep hurdles. Established developer ecosystems create high switching costs, with DeepSeek and Qwen having built mature open-source communities. Massive traffic gaps cannot be closed in the short term, as Baidu cannot replicate Douyin or WeChat-scale super platforms. Repairing brand trust damaged by strategic flip-flops and boosting internal organizational efficiency demand years of gradual reform. Moving forward, Baidu will likely retreat to segmented niches: high-security government enterprise contracts and exclusive search-augmented use cases, surrendering leadership of mass consumer markets and global developer ecosystems.
Conclusion
Baidu’s rise and fall in AI serves as a critical industry cautionary tale: early head starts, heavy R&D spending, and technical reserves do not guarantee victory. Sustainable competitive advantage hinges on consistent, aligned strategy, business models tailored to inherent corporate strengths, and rapid market agility. As of June 2026, ByteDance, Alibaba, and Tencent form the big-tech first tier, with independent technical dark horse DeepSeek securing a permanent top spot, while the sector’s original trailblazer Baidu slides into the second tier. This market reset proves AI competition is not merely a race of R&D investment or launch timing. Long-term industry gains belong only to players who integrate technology, traffic, real-world scenarios, and ecosystems into a complete, self-reinforcing closed loop. Reconciling first-mover legacy with adaptive, market-focused execution defines Baidu’s most pressing challenge ahead.