From transportation networks to e-commerce supply chains, industries exporting China’s strengths share a common trait: their success hinges on proving technology can handle real-world, high-demand scenarios. This means surviving peak traffic loads and integrating seamlessly into users’ daily routines—key factors shaping adoption and growth.
In 2024, artificial intelligence innovation saw a shift as tech companies, rather than academic institutions, took the lead in developing large language models (LLMs). Firms like Google, Microsoft, and OpenAI abroad, along with Baidu and Alibaba in China, moved swiftly to dominate the field. The near-simultaneous launches of ChatGPT and Baidu’s Ernie Bot highlighted how Baidu’s sustained investment—over 20% of its budget directed to R&D during the mobile internet boom—secured its competitive edge.
Corporate control of AI development has created an ecosystem where companies manage everything from foundational research to real-world applications. This competition follows a familiar strategy from the internet era: prioritize scale and visibility over immediate profits. However, while large models generate attention, their real value lies in enabling specific, practical applications.
Like 5G, which became meaningful through applications like live streaming and e-commerce, large models will only matter if they support similarly impactful use cases. Currently, the gap between AI’s technical potential and its user base limits its commercial viability. Baidu CEO Robin Li has noted that major breakthroughs more often result from widespread adoption than the other way around.
In the past year, limitations of large models have become widely accepted within the AI industry. OpenAI’s text-to-video generator Sora drew interest but struggled with practical results, emphasizing again that technology’s value is realized through application, not innovation alone. As Sora faltered, Li’s view gained traction: scaling applications is where true progress happens.
By 2024, AI had penetrated niche markets, but large models no longer felt revolutionary. While they drove incremental improvements, no single application dominated the landscape. Instead, competition shifted to heavy marketing and cash-intensive promotions, turning the focus from technical advancements to financial endurance. This approach has raised doubts about its sustainability and whether it can fuel long-term growth.
To counteract these challenges, companies like Baidu are shifting their focus to solving specific problems. For instance, Baidu’s incremental retrieval-augmented generation (iRAG) technology addresses hallucination rates in image generation, while its Miaoda coding assistant and AI-powered Wenku platform aim to enhance productivity. These efforts represent a move from scaling technology to refining its real-world applications.
In 2025, LLMs will likely evolve in tandem with the industries they serve, driven by commercialization efforts. Baidu has strategically positioned its platforms to meet user needs while broadening their appeal. Beyond Qianfan, Baidu’s Miaoda is tailored for everyday users, reflecting the philosophy of creating accessible tools for larger markets. Baidu’s history of trial-and-error has cultivated adaptability—a key factor Li believes will sustain its competitive edge.
Li anticipates that 2025 will be a pivotal year for AI applications, marked by exponential user growth and a vibrant market landscape. This prediction stems from 2024, often dubbed the dawn of AI apps, when industries began rapidly integrating AI into diverse scenarios. Thousands of applications were developed in just one year, setting the stage for 2025’s commercial surge.
But where will the standards be set for others to follow? AI-generated content (AIGC) is poised to lead, closely tied to intelligent search, productivity tools, and smart education. In China, the enterprise services market holds significant promise for AI apps. Despite underwhelming performance during the software-as-a-service era, enterprise services now offer scalability for embedding AI into industries.
The rapid adoption of Baidu’s large models reflects this potential. Over the past year, usage is said to have surged thirtyfold, with reported calls exceeding 1.5 billion. More than 60% of China’s central state-owned enterprises and national companies have integrated these models. Baidu has emerged as a key player, securing 40 contracts worth RMB 2.74 billion (USD 383.6 million) in 2024. Public data indicates that these projects are part of a broader market valued at RMB 17.1 billion (USD 2.4 billion). While partnerships between tech providers and enterprises hold immense promise, the full benefits will take time to unfold.
On the consumer side, business demand, rather than direct user needs, continues to drive AI applications. In 2024, Baidu’s Wenku attracted over 40 million paying users—a 60% increase—with most leveraging its AI-enabled features. Focused on productivity and knowledge management, Wenku has thrived, unlike entertainment apps that often struggle with monetization.
While general AI apps attract large audiences, they often lack clear monetization paths, limiting scalability. Wenku, with approximately 70 million monthly active users, highlights the potential of AI-powered productivity tools. Rather than investing heavily in consumer-facing tools like chatbots, Baidu has focused on enhancing existing products like Wenku, taking a pragmatic approach to AI commercialization.
Wenku’s rise underscores the challenge of building a superapp. Platforms like WeChat became essential by embedding themselves in daily routines—an achievement current AI apps have yet to replicate. Creating such an app requires sustained effort and an ability to meet broad, everyday needs.
Globally, AI apps are generating significant buzz, with platforms like Product Hunt spotlighting new tools almost daily. Yet, major tech companies are approaching the space with measured caution. OpenAI is concentrating on practical use cases, Apple is strengthening partnerships within its ecosystem, and Google is focusing on developing AI tools for specific applications. In China, the momentum has slowed—venture capital funding for AI in 2024 dropped to 80% of the previous year’s levels, according to ITjuzi. While the market is flooded with new AI tools, their low technical barriers to entry make it difficult to demonstrate sustainable, long-term value.
RELATED ARTICLE
This skepticism extends to large-scale advertising for AI models. Heavy spending on user acquisition is risky, often hinging on hopes of creating a superapp akin to WeChat, Taobao, or Douyin. However, current AI apps fall short—not due to technical limitations but because they lack the features to foster habitual engagement. Baidu’s dominance in search was not built on its search box alone but on its ability to deliver reliable answers and monetize through advertising—a formula yet to be replicated by AI applications.
Li acknowledged this challenge in a letter to employees, stating, “Competition is fiercer than ever.” Baidu’s strategy, he emphasized, is centered on making specialized capabilities accessible to broad audiences.
The race for the next superapp remains open. It will likely emerge in scenarios where users find indispensable value. Wenku’s expanding user base positions it as a contender, particularly as Baidu continues refining its AI tools. By mid-2025, Baidu plans to launch Ernie Bot’s version 5.0, laying a foundation for future applications. Whether this marks the next transformative platform is uncertain, but it signifies another step in AI’s growing role in everyday life.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This article was written by Xiao Xi for 36Kr.