Cover Story: China’s AI Titans Escalate Battle for Control of Digital Gateways

At the end of 2025, China’s artificial intelligence (AI) market is fiercely competitive, having evolved from a race to train large models into a multi-dimensional battle over users, hardware, talent, and capital. This intense competition largely pits the country’s internet giants against each other while pressuring startups to make strategic adaptations. Notably, Alibaba Group and ByteDance are at the forefront, swiftly launching foundational models, aggressively promoting AI consumer applications, and integrating these technologies into smartphones and wearable devices. The biggest goal: controlling the next generation of digital "gateways" through which people access services—search, communication, shopping, and navigation [para. 1][para. 2][para. 3].

AI startups are now seeking to go public to access capital necessary for funding the enormous computing and talent costs. Investors have shifted focus from speculative large model valuations to whether AI models can deliver real, practical applications. Companies like ByteDance, which leads China’s Model-as-a-Service (MaaS) market with a 37.5% share (the market itself surged 421.2% year-on-year in H1 2025 to 1.29 billion yuan or about $184 million), and Alibaba, are vying for enterprise customers [para. 4][para. 5][para. 6].

ByteDance’s Volcano Engine boasts over 100 enterprise customers, each generating over one trillion tokens, while OpenAI and AWS have more than 30 and 50, respectively. A massive talent war underpins these competitive efforts—Tencent, for example, recruited a former OpenAI researcher as chief AI scientist and set plans to add 28,000 tech internships over three years. ByteDance and Alibaba are equally aggressive, with ByteDance posting the most new AI jobs in 2025 [para. 7][para. 8].

Incumbents’ dominance is rooted in their control over digital gateways. Alibaba, for instance, rebranded its chat app Tongyi as Qwen and integrated it into various services, rapidly surpassing 30 million monthly active users (MAU), while Ant Group’s LingGuang app and Afu health assistant quickly amassed millions of downloads and users, aided by integrations with services like Alipay. Such growth requires significant investment, driving up marketing and user acquisition costs and squeezing out smaller competitors [para. 15][para. 16][para. 17][para. 18].

ByteDance has taken the lead in user numbers, with its Doubao assistant reaching 172 million MAU by December 2025 and seeing daily token usage over 50 trillion. Success is attributed not just to ByteDance’s existing traffic from Douyin, but to innovations in multimodal features and deep cloud integration. ByteDance also connected Doubao to e-commerce, streamlining AI-driven shopping [para. 21][para. 22][para. 23][para. 24].

As the giants fight for dominance, startups are shifting focus from mass-market consumer apps to specific industries and foundational model innovation due to escalating costs. Several “AI Tigers” startups have exited the foundation-model race; others like Zhipu AI, MiniMax, and Moonshot persisting through IPOs despite massive losses (e.g., Zhipu AI lost 2.36 billion yuan on 191 million in revenue in H1 2025). IPO activity in 2026 is expected to swell, testing Hong Kong’s financial market [para. 34][para. 35][para. 36].

Both giants and startups recognize that, despite current differences, ongoing advances at the model level drive real AI competitiveness. While U.S. export controls limit hardware access, China leverages its vast technical talent and burgeoning open-source model community, highlighted by DeepSeek becoming the top model provider globally by token volume in 2025. As companies search for applications where they can become indispensable, the race to improve AI models and capture their value shows no signs of ending [para. 48][para. 49][para. 50][para. 51][para. 52][para. 53].