In-Depth Analysis of the AI Large Model Competitive Landscape of ByteDance, Alibaba, and Tencent

Based on the collected information, I will write a systematic and comprehensive analysis report on the competitive landscape of AI large models for you.

---

Current capital investments in the AI sector by China’s three major internet giants show clear differentiation:

Company	Investment Plan	Core Strategy
Alibaba	RMB 380 billion over three years, covering AI+cloud infrastructure[1][2]	Full-stack self-developed, open-source ecosystem
ByteDance	Approximately RMB 150 billion in 2025, projected RMB 160 billion in 2026[1][3]	Extreme ROI, user-driven
Tencent	Hundreds of billions of RMB in 2025, prudent after sufficient GPUs[4]	Ecosystem integration, steady advancement

Alibaba’s capital expenditure in Q3 2025 reached RMB 31.501 billion, a year-on-year increase of 80.1%, with a cumulative RMB 94.789 billion in the first three quarters, a substantial year-on-year increase of 132.7%[4]. Approximately half of ByteDance’s planned RMB 160 billion investment in 2026 will be used for AI chip procurement[3]. In contrast, after Tencent’s capital expenditure surged 386% year-on-year to RMB 36.58 billion in Q4 2024, it decreased quarter by quarter in 2025, with Q3 capital expenditure reaching RMB 12.983 billion, a year-on-year decrease of 24.1%[4].

is committed to becoming an “infrastructure provider” in the AI era, aiming to provide a computing power “power grid” through Alibaba Cloud and build “the Android of the AI era” with Tongyi Qianwen[5]. CEO Wu Yongming stated that large models are the next-generation operating system, and Tongyi will serve as the core carrier for AI popularization.

continues its core competitiveness of “traffic + algorithms”, entering the battlefield from C-end applications. Zhang Yiming has been deeply involved in AI strategy since the second half of 2024, and the Seed team is regarded as the core business, having completed the integration of departments such as the AI Lab[6].

adopts an “ecosystem embedding” strategy, achieving seamless integration of AI capabilities through national-level applications such as WeChat and QQ. Tencent President Martin Lau emphasized that GPUs are currently sufficient, and capital expenditure is becoming more rational[4].

---

According to QuestMobile’s 2025 Q3 China AI Application Market Report[2][7]:

Rank	Application	Monthly Active Users	Features
1	ByteDance Doubao	172 million	China’s first product with 100 million+ DAUs
2	DeepSeek	145 million	Open-source dark horse, outstanding reasoning capabilities
3	Tencent Yuanbao	32.86 million	Dual-model collaboration (Hunyuan + DeepSeek)
4	Alibaba Tongyi Qianwen	30 million+	Surpassed 30 million users in 23 days, setting a growth rate record

Doubao’s DAU has exceeded 100 million, and its marketing costs during user growth are the lowest among all ByteDance products that have exceeded 100 million DAUs in history[6]. This achievement fully demonstrates ByteDance’s “blitzkrieg” capabilities in user acquisition and product refinement.

In November 2025, the number of ad creatives for native AI products in the mainland market reached 891,000 sets[6]:

• Tencent Yuanbao
  : Accounts for 46%, with the most aggressive delivery strategy, topping the App Store Free Chart in March
• Alibaba Tongyi Qianwen
  : Accounts for 34%, with rapid volume expansion after its public beta in November
• ByteDance Doubao
  : Accounts for only 11%, relying on product strength and organic growth

is positioned as an “AI-native super application”, emphasizing general Q&A, creative generation, and other pan-entertainment scenarios, achieving low-cost user acquisition through Douyin’s traffic pool[6].

emphasizes “usability” and “productivity”. Upgraded from the original “Tongyi” app, it was developed in a closed team of over 100 engineers led personally by Wu Yongming[8], with the goal of building an “AI life entrance”.

relies on the WeChat ecosystem to achieve “seamless integration” — users do not need to download a separate app and can access AI capabilities directly within WeChat[9]. Tencent was the first to implement dual-model collaboration of “Hunyuan + DeepSeek”, allowing users to switch models according to scenarios[9].

---

In the Model-as-a-Service (MaaS) market, Volcano Engine’s daily token consumption reaches 50 trillion, and according to Informa’s research, it is already “the largest in China and the third largest in the world”[1]. Alibaba Cloud emphasizes “effective token calls” as its differentiated competitive point.

Dimension	Volcano Engine (ByteDance)	Alibaba Cloud
Token Call Volume	50 trillion per day[1]	Emphasizes quality over quantity
Market Positioning	Internet-native customers	Full industry coverage
Technical Architecture	Doubao large model series	Tongyi Qianwen series
Ecosystem Strategy	Douyin/Toutiao scenario closed loop	Open-source and open strategy

’s revenue from AI-related products has maintained triple-digit year-on-year growth for nine consecutive quarters[2]. Alibaba Cloud has formed a “data flywheel” in Alibaba’s ecosystem scenarios such as e-commerce, logistics, and finance — real businesses generate data, which in turn feeds back to optimize models[5].

has formed unique advantages in scenarios such as short videos and creative content generation through the Doubao model, especially in the video AIGC field. After the release of Sora2, the necessity of ByteDance’s development of AIGC has become more clear[1].

---

Company	Large Model Strategy	Technical Features
ByteDance	Closed-source first, rapid iteration	MoE architecture, reasoning costs reduced to 60% of the industry average[10]
Alibaba	Open-source and open, full-stack self-developed	Qwen3 trained with 36 trillion tokens of data[9]
Tencent	Dual-model collaboration, ecosystem integration	Flexible switching between Hunyuan and DeepSeek[9]

Alibaba’s Tongyi Qianwen firmly follows the open-source route. Qwen2.5 is pre-trained on 18 trillion tokens, and Qwen3 uses approximately 36 trillion tokens, covering 119 languages and dialects[9]. ByteDance’s Seed team has completed internal integration, emphasizing the strategic determination of “achieving success in one go”[6].

In 2025, the “Hundred Glasses Battle” broke out, and AI glasses are regarded as the next-generation key terminal[7]:

Product	Release Status	Pricing	Core Features
Quark AI Glasses	Released in November 2025	Starting from RMB 1,899	Integrated with Alibaba’s ecosystem
Doubao AI Glasses	Expected to be released in 2026	Expected to be under RMB 2,000	Equipped with Doubao Assistant
Xiaodu AI Glasses Pro	Launched in November 2025	-	Real-time translation, photography

ByteDance also launched the “nubia M153” smartphone in partnership with ZTE Nubia, which deeply integrates Doubao’s capabilities[2]. The essence of this “hardware-software integration” strategy is to compete for the “first touchpoint” of user interaction in the AI era.

:

• Alibaba previously pinned its hopes on Quark (120 million DAUs), but the usage rate of AI functions was less than 20%[8]
• Wu Yongming judged: “Alibaba must have an AI-native C-end super entrance”[8]
• ByteDance Doubao is the only independent AI-native app with more users than DeepSeek[1]

:
Lucy Lu, Research Director of IDC China, pointed out: “The boundary between enterprise-level and consumer-level large model applications is becoming increasingly blurred, and enterprise-level applications often rely on data from C-end applications for feedback and iteration”[2].

---

Dimension	Key Elements	ByteDance’s Advantages	Alibaba’s Advantages	Tencent’s Advantages
Computing Power	GPU scale, self-developed chips	Sufficient procurement budget	Alibaba Cloud infrastructure	Sufficient GPUs, prudent investment
Data	Training data, user data	Massive Douyin video data	Full scenarios in e-commerce/finance	Social/content ecosystem
Scenarios	Application implementation capability	Short videos, creative tools	E-commerce, cloud services	Social, office
Ecosystem	Developers, partners	Traffic distribution capability	Open-source community	User reach capability
Commercialization	Monetization capability	Advertising-e-commerce linkage	Cloud service subscriptions	Value-added services

(Benchmarking OpenAI)

• Focus on breakthroughs in basic model capabilities
• Invest in massive computing power and data
• Representatives: DeepSeek, Zhipu AI

(Benchmarking Android)

• Establish a developer ecosystem through open-source strategies
• Provide platforms rather than single products
• Representative: Alibaba Tongyi Qianwen

(Benchmarking TikTok Model)

• Drive technological iteration through user scale
• Rapid productization and implementation
• Representative: ByteDance Doubao

(Benchmarking WeChat Model)

• Integrate AI capabilities into existing super applications
• Achieve “seamless” user experience
• Representative: Tencent Yuanbao

1. Multimodal Capabilities
   : Doubao’s “vision-language-control” three-modal integration solution has achieved industrial-level application[10]; Tencent Hunyuan Image 3.0 ranked first in the global “blind test”[11]

2. Long-Text/Long-Range Reasoning
   : Tongyi Qianwen can process extremely long materials of over 10,000 pages, supporting one-click quick reading of 100 documents[9]

3. AI Agents
   : Evolving from “conversation assistants” to “task executors”, Yuanbao has launched the “task reminder” function[9]

4. Embodied Intelligence
   : Integration of AI with hardware, exploration of new categories such as Doubao smartphones and AI glasses[7]

---

• Mismatch between AI investment and return cycles, short-term financial pressure is inevitable[2]
• User willingness to pay still needs to be cultivated, and the free model remains mainstream
• Insufficient quantitative evaluation of AI effects in the enterprise-level market

• Data diversity and quality remain major bottlenecks[10]
• Insufficient model interpretability and transparency[10]
• Lack of unified technical standards and specifications[10]

• Rise of new players such as DeepSeek, with rapid user growth[7]
• The industry may fall into a vicious cycle of “burning money to gain market share”
• Intensified homogeneous competition makes it difficult to maintain differentiated advantages

---

1. ByteDance
   leads in C-end applications with its core capabilities of “traffic + algorithms”, and Doubao’s DAU exceeding 100 million is a milestone. However, its B-end market still needs to catch up with Volcano Engine’s MaaS layout.

2. Alibaba
   relies on Alibaba Cloud’s computing power base and Tongyi Qianwen’s open-source ecosystem to occupy advantages in technological influence and the B-end market. Its three-year RMB 380 billion investment plan demonstrates its strategic determination to go All-in on AI.

3. Tencent
   adopts a steady strategy to achieve “seamless” integration of AI capabilities through the WeChat ecosystem. Although capital expenditure is becoming more rational, the growth rate of Yuanbao is worthy of attention.

4. Differentiated competition for general-purpose large models
   will revolve around four dimensions: computing power, data, scenarios, and ecosystem. Pure technological leadership is difficult to form a lasting moat; ecosystem building and scenario implementation capabilities will become key.

• Market Structure
  : It is expected to form a “1+N” structure, with Doubao maintaining its leading position, Tongyi Qianwen and Yuanbao forming the second echelon, and new players such as DeepSeek continuing to stir up the market
• Commercialization Breakthrough
  : It is expected that sustainable payment models will emerge in 2026, and AI-native applications will find clear monetization paths
• Battle for Entrances
  : AI glasses may become a key variable, and hardware-software integration capabilities will become a new focus of competition
• Technological Evolution
  : Multimodal capabilities, long-range reasoning, and AI agent capabilities will become the core competitive dimensions of large models

In this marathon-style competition, temporary leadership does not mean final victory. The reshaping of China’s AI large model landscape may have just begun.

---

[1] The Paper - “Alibaba, ByteDance, Who Will Define the AI Battle in 2026?” (https://m.thepaper.cn/newsDetail_forward_32328218)

[2] Securities Times - “2025 China Technology New Trend: Tech Giants Invest Hundreds of Billions in AI, Competing for Super Entrances” (https://www.stcn.com/article/detail/3563827.html)

[3] Huxiu - “ByteDance Won Another Bet” (https://www.huxiu.com/article/4821029.html)

[4] OFweek AI Network - “Divergence in AI Investment: Alibaba ‘Steps on the Gas’, Tencent ‘Steps on the Brake’” (https://m.ofweek.com/ai/2025-12/ART-201700-8420-30676794.html)

[5] DoNews - “Three Advantages of Alibaba on the Road to Artificial Superintelligence (ASI)” (https://www.donews.com/news/detail/4/6315109.html)

[6] Sina Finance - “ByteDance Won Another Bet” (https://finance.sina.com.cn/stock/2025-12-29/doc-inhemmwr8192273.shtml)

[7] Southern Daily - “Starting Price Expected to Be Under RMB 2,000, Doubao AI Glasses Will Be Shipped” (https://news.southcn.com/node_08203b6b14/f4a945c742.shtml)

[8] Sina Finance - “Why Does Alibaba Insist on Fighting the Tongyi Qianwen Battle?” (https://finance.sina.cn/stock/jdts/2026-01-08/detail-inhfqikr5453016.d.html)

[9] 36Kr - “Annual AI Model Inventory” (https://m.36kr.com/p/3626323665863680)

[10] Sina Finance - “2025: The Year of Large Language Models” (https://t.cj.sina.cn/articles/view/6248544856/174713a5802002hvd8)

[11] Tencent Hunyuan Official Website - “Tencent HY 2.0, the Latest Version of Hunyuan Language Model, Officially Released” (https://hunyuan.tencent.com/news/home)