The ‘Catfish’ Effect: DeepSeek’s Unintentional Disruption
DeepSeek’s emergence in the Chinese artificial intelligence (AI) sector has sent ripples throughout the industry, prompting a flurry of activity among its peers. This relatively new entrant, spun off from a hedge fund, has challenged conventional approaches to AI model development and pricing, forcing other startups to re-evaluate their strategies and seek new avenues for growth and funding. DeepSeek’s impact extends beyond China’s borders, making waves in both Wall Street and Silicon Valley. However, its influence is most acutely felt within the Chinese AI community, where it has overshadowed established players like Moonshot AI and MiniMax.
Liang Wenfeng, the driving force behind DeepSeek, admitted in a July 2024 interview that the company had inadvertently become the “catfish” in China’s AI market. This metaphorical term refers to a competitive element that stirs up activity and prevents stagnation within a particular industry. While DeepSeek’s initial intention may not have been to disrupt, the release of its V2 model in July 2024 triggered a price war, and subsequent releases (V3 in December and R1 in January) further cemented its disruptive role. These advancements posed existential questions for many players in China’s already crowded AI model market. The speed of DeepSeek’s advancements, coupled with its open-source approach and competitive pricing, created a sense of urgency among its competitors.
A Divergence in Approaches: China vs. the US
Paradoxically, DeepSeek’s disruption may have ultimately benefited China’s AI ecosystem. By pushing the boundaries of AI model capabilities and making the technology more accessible, DeepSeek has, according to some analysts, given China an “edge.” This ‘edge’ refers to the accelerated development and adoption of AI technologies within China, driven by the increased competition and innovation sparked by DeepSeek.
AI analyst Grace Shao, founder of the industry newsletter AI Proem, highlights a key difference in how AI is being approached in China versus the United States. Before DeepSeek’s R1, many Chinese AI startups were focusing on consumer-facing applications. The strategies are driven by the monetization strategy from the mobile internet era. This meant prioritizing applications that could quickly generate revenue through user engagement, similar to how mobile apps were monetized. In contrast, the US has largely embraced AI as a tool for enhancing enterprise and white-collar productivity. This difference in focus stems from the differing economic structures and priorities of the two countries.
Shao attributes this divergence to structural economic differences between the two markets. However, she emphasizes that strong model capabilities remain the cornerstone of the AI industry, regardless of the specific application. This means that even if a company is focused on consumer applications, the underlying AI model must be powerful and robust to deliver a compelling user experience. The foundational model serves as the bedrock upon which all applications are built.
The Race to Catch Up: China’s AI Startups Respond
Recognizing the importance of foundational AI advancements, other Chinese AI model developers are now striving to close the gap with DeepSeek. This involves not only improving their model capabilities but also adopting strategies that allow them to compete effectively in the new landscape shaped by DeepSeek.
Zhipu AI: Securing Funding and Embracing Open Source
Beijing-based Zhipu AI, a startup with roots in Tsinghua University, recently announced a significant funding round of 1 billion yuan (US$140 million). This investment included support from the municipal government of Hangzhou, where Zhipu AI has established a subsidiary. This substantial funding demonstrates the continued investor confidence in China’s AI sector and the recognition of Zhipu AI’s potential.
Beyond securing funding, Zhipu AI has also embraced the open-source movement. The company has made its AI models and agents available to developers, fostering collaboration and innovation within the broader community. This open-source strategy allows other developers to build upon Zhipu AI’s work, creating a more vibrant and collaborative ecosystem. A recent example of this commitment is the release of CogView-4, an open-source text-to-image model capable of generating Chinese characters. This demonstrates Zhipu AI’s commitment to contributing to the advancement of AI technology, even in areas that may not directly generate revenue.
The Open-Source Surge: A Cultural Shift
The trend towards open-source development in China’s AI sector reflects a broader cultural shift. This shift is driven by several factors, including a desire to showcase Chinese innovation and the allure of global recognition.
Why Open Source?
- A Desire to Prove Innovation: For a generation of entrepreneurs born in the 80s and 90s, there’s a strong desire to demonstrate that Chinese companies are capable of genuine innovation, moving beyond the perception of simply “copying” existing technologies. This generation is motivated to prove that China can be a leader in technological innovation, not just a follower.
- Global Recognition: The allure of international recognition is a powerful motivator. Being cited and utilized by developers and businesses outside of China is often seen as more prestigious than solely focusing on profit from individual projects. This desire for global recognition drives companies to contribute to the open-source community, where their work can be evaluated and appreciated by a wider audience.
Stepfun: Multimodal Models and Strategic Partnerships
Shanghai-based Stepfun, founded in 2023 by former Microsoft Research Asia chief scientist Jiang Daxin, is another startup making strides in the open-source arena. Stepfun is focusing on developing multimodal models, which can process and generate different types of data, such as text, images, audio, and video.
Stepfun’s Open-Source Contributions:
- Step-Video-T2V: A model that generates videos from text input. This allows users to create short videos simply by providing a text description.
- Step-Audio: Designed for voice interactions. This model focuses on processing and generating audio, enabling applications like voice assistants and speech synthesis.
- Upcoming Image-to-Video Model: Planned for release this month. This model will further expand Stepfun’s multimodal capabilities, allowing users to generate videos from images.
Stepfun’s strategic partnerships highlight the collaborative nature of China’s AI ecosystem. Backers include the Shanghai municipal government-owned Capital Investment Co, internet giant Tencent Holdings, Qiming Venture Partners, and 5Y Capital. These partnerships provide Stepfun with not only financial resources but also access to expertise and market reach.
MiniMax: A Belated Embrace of Open Source
MiniMax, known for its popular personalized AI apps Talkie and Xingye, initially pursued a more closed approach. However, the company shifted gears in January, shortly after DeepSeek’s V3 release. This shift demonstrates the significant impact of DeepSeek’s disruptive entry, forcing even established players to reconsider their strategies.
MiniMax’s Open-Source Offerings:
- MiniMax-Text-01: A large language model (LLM), the technology underpinning generative AI services like ChatGPT. This is MiniMax’s core offering in the open-source space, providing a powerful foundation for various text-based applications.
- MiniMax-VL-01: A multimodal model. This expands MiniMax’s capabilities beyond text, allowing it to process and generate other types of data.
Founder Yan Junjie candidly admitted in an interview with Chinese media outlet LatePost that, given a second chance, he would have chosen the open-source path from the very beginning. This admission underscores the growing recognition of the benefits of open-source development within China’s AI community.
Moonshot AI: Multimodal Reasoning and Innovation
Moonshot AI, recognized for its Kimi chatbot, has also been active in the open-source space. Moonshot AI is focusing on developing models with strong reasoning capabilities, allowing them to perform complex tasks that require understanding and inference.
Moonshot AI’s Contributions:
- K1.5: An o1-level multimodal reasoning model, released in January (coinciding with DeepSeek’s R1 launch). This model demonstrates Moonshot AI’s commitment to advancing multimodal reasoning capabilities.
- Open-Source Architecture and Optimizer Innovations: Introduced last month. These contributions aim to improve the efficiency and performance of AI models, benefiting the broader AI community.
Baichuan AI: Refocusing on the Medical Sector
Baichuan AI, founded by former Sogou CEO Wang Xiaochuan, has made a strategic shift to concentrate its efforts on the medical sector. This refocusing involved restructuring, including the disbanding of its financial services team. Baichuan AI confirmed this move, stating that it was “optimizing and adjusting the financial business to concentrate resources and focus on our core medical businesses.” This strategic shift demonstrates the growing trend of AI companies specializing in specific industries to deliver more targeted and impactful solutions.
01.AI: From Large-Scale Models to Industry-Specific Applications
01.AI, founded by former Google China president Lee Kai-fu, has also undergone a strategic pivot. The company has moved away from training large-scale AI models and is now sharpening its focus on developing industry-specific applications. This shift reflects a broader trend in the AI industry, where companies are moving beyond general-purpose models to create solutions tailored to the specific needs of particular sectors. A notable example of this shift is 01.AI’s partnership with Alibaba Group Holding’s cloud computing services unit to establish an “industrial large model joint lab.” This collaboration involved the transition of several 01.AI employees to Alibaba Cloud. This partnership demonstrates the increasing collaboration between AI startups and established tech companies in China.
The Evolving Landscape: Competition and Collaboration
The Chinese AI landscape is characterized by a dynamic interplay of competition and collaboration. DeepSeek’s disruptive entry has undoubtedly spurred a wave of innovation, forcing established players to adapt and newcomers to accelerate their efforts. The embrace of open-source principles is fostering a more collaborative environment, where knowledge sharing and collective progress are prioritized. The competition is driving companies to develop more powerful and efficient models, while the collaboration is accelerating the overall pace of innovation.
Key Trends:
- Increased Focus on Foundational Models: Startups are recognizing the importance of developing strong, foundational AI models as the basis for future innovation. This focus on foundational models is crucial for building robust and versatile AI applications.
- Open-Source Movement: The open-source approach is gaining traction, driven by a desire for global recognition and a belief in the power of collaborative development. Open-source initiatives are fostering a more vibrant and inclusive AI ecosystem.
- Strategic Partnerships: Collaborations between startups, established tech companies, and government entities are becoming increasingly common, pooling resources and expertise. These partnerships are accelerating the development and deployment of AI technologies.
- Industry-Specific Applications: Some companies are shifting their focus from general-purpose AI models to developing specialized applications tailored to specific industries, such as healthcare and finance. This specialization allows companies to create solutions that are more directly relevant and impactful.
The long-term impact of DeepSeek’s disruption remains to be seen. However, one thing is clear: the Chinese AI landscape is undergoing a period of rapid evolution, driven by a combination of competitive pressure, technological advancements, and a growing commitment to open-source principles. This dynamic environment promises to yield further breakthroughs and reshape the future of AI, not only in China but globally. The heightened competition is also forcing a greater emphasis on efficiency and cost-effectiveness. Companies are under pressure to deliver high-quality models at competitive prices, which ultimately benefits end-users and accelerates the adoption of AI across various sectors.
The shift towards industry-specific applications is also a noteworthy trend. By focusing on the unique needs and challenges of particular sectors, companies like Baichuan AI and 01.AI are aiming to create solutions that are more directly relevant and impactful. This approach could lead to faster adoption of AI in areas like healthcare, finance, and manufacturing, driving tangible benefits for businesses and consumers alike. The specialization allows for a deeper understanding of the specific requirements and challenges of each industry, leading to more effective and tailored AI solutions.
Furthermore, the involvement of government entities, such as the Shanghai municipal government’s investment in Stepfun and the Hangzhou government’s support for Zhipu AI, highlights the strategic importance of AI development in China. This government backing provides not only financial resources but also a degree of validation and stability, encouraging further investment and innovation in the sector. The government’s support signals a long-term commitment to fostering the growth of the AI industry in China.
The interplay between competition and collaboration is also a defining characteristic of the Chinese AI landscape. While companies are undoubtedly vying for market share and recognition, there’s also a growing recognition that collaboration, particularly through open-source initiatives, can accelerate progress for the entire industry. This collaborative spirit is evident in the sharing of models, code, and research findings, fostering a sense of collective advancement. The combination of competition and collaboration creates a dynamic and productive environment for AI innovation.
The ongoing evolution of China’s AI sector is a testament to the country’s ambition to become a global leader in artificial intelligence. The combination of entrepreneurial drive, government support, and a growing embrace of open-source principles is creating a fertile ground for innovation. As Chinese AI companies continue to push the boundaries of what’s possible, they are not only reshaping their domestic market but also contributing to the global advancement of artificial intelligence. The story of DeepSeek and its impact on the Chinese AI landscape is a compelling example of how a single disruptive force can catalyze widespread change and accelerate the pace of innovation across an entire industry. The ripple effects of DeepSeek’s entry are still being felt, and the long-term consequences will likely shape the future of AI in China and beyond for years to come. The increased focus on foundational models, the rise of open-source development, the formation of strategic partnerships, and the shift towards industry-specific applications are all key trends that are transforming the Chinese AI landscape and positioning it for continued growth and innovation.