Summary of Key Points
The article uses the tactical analogy of the Cape Verde football team drawing with the World Cup champions to describe China's AI industry's long-standing strategy of "following and imitating" the United States. China has been doing whatever the US does, relying on its underlying technologies (frameworks, models, chips), and focusing on making incremental improvements in order to catch up. However, this approach carries several risks: it may lead to a widening gap in technology, the risk of supply disruptions, and a stall in the innovation momentum. The article also mentions that China's AI efforts, such as the PaddlePaddle framework and the Panggu large model, failed to sustain due to various reasons. It concludes by calling on China's AI community to shift from a "weakling's mindset" to a "stronger mindset," leveraging its own strengths (a new national governance system, industry intelligence, hardware capabilities, and engineering breakthroughs) to achieve genuine innovation rather than merely aiming for parity.
I. China's AI Strategy: The Current Situation of "Following the US"
The core approach of China's AI development in recent years has been to "follow trends and make incremental improvements":
- "Making incremental improvements" means relying on existing technologies and focusing on superficial enhancements: While the US is responsible for groundbreaking innovations from scratch (such as the PyTorch framework, LLaMA models, and NVIDIA chips), China focuses on localized adaptations (e.g., fine-tuning industry-specific models using LLaMA or developing applications within the CUDA ecosystem). It's like building a house on someone else's foundation, only adding finishing touches.
- "Following trends" means reacting quickly to US initiatives: When ChatGPT became popular, China immediately saw a surge in AI model development; similarly, when OpenAI's Agent technology gained attention, many companies rushed to adopt it. Even business models (API services, SaaS tools) are directly copied from US examples. Many believe that since US products cannot enter the Chinese market, replicating local versions is sufficient, considering it a victory.
II. The Hidden Dangers of Following the US
This defensive strategy may seem secure, but it actually poses several threats:
1. Widening technology gap: In 2024, the gap between Chinese and US AI models was reduced to three months, but in 2025, as the US increased its R&D efforts, China fell into a period of focusing more on marketing rather than innovation, leading to a renewed widening of the gap.
2. Risk of supply disruptions: The US has begun to restrict the export of AI technologies (for example, Anthropic's models are only available to Americans), which could leave China vulnerable if its reliance on these technologies is severed.
3. Stall in innovation momentum: Companies often prioritize profit-making over R&D, leading to a stagnation in core technologies. For instance, there were few new breakthroughs in the AI field this year, with most efforts being imitative.
4. Trapped as a follower: Historically, Japan and Europe failed to surpass the US during industrial revolutions, staying at around 80% of the leading level, perpetuating a sense of anxiety and constant fear of falling behind.
III. China's Attempts at Original Innovation
China has made attempts at independent innovation, but they often didn't succeed:
- PaddlePaddle framework: Baidu aimed to create an AI "operating system" that was more comprehensive than US frameworks, integrating industry-specific solutions. However, with the rise of large models, resources were diverted, and its impact remains limited, especially in government and enterprise contexts.
- Panggu large model: Huawei was among the first to integrate AI with industries (mining, transportation), but the project lost momentum due to controversy, casting doubt on the feasibility of combining AI with various sectors.
- DeepSeek's MoE optimization: This innovation promised to double computing power without increasing the number of parameters. However, it proved to be inefficient in practical applications and became just one of many competing solutions.
The reason for these failures is simple: fundamental innovation requires significant investment and a long time frame, while companies prefer to wait for others to develop the technologies before investing themselves.
IV. From "Weakling" to "Stronger": The Path Forward for China's AI
To succeed, China needs to adopt a more proactive approach:
1. A new national governance system: Similar to how China overcame chip shortages by focusing on domestic technology development, it should concentrate resources on building its own AI infrastructure.
2. Industry intelligence: Instead of competing with the US in consumer markets (e.g., chatbots), China should leverage its strengths in industrial sectors (manufacturing, transportation, agriculture) where the potential for AI integration is vast.
3. Hardware capabilities: China has the capability to produce hardware (servers, sensors). It should use hardware to enhance AI performance rather than relying on AI to improve existing systems.
4. Engineering breakthroughs: With a large pool of software engineers, China can use engineering expertise to overcome theoretical shortcomings and deliver better practical solutions.
Conclusion
China's AI industry does not need a "Cape Verde miracle" — achieving parity through defense is not a true victory. It must shift from a follower to a leader, utilizing its unique advantages to prove that winning the global AI competition is not a stroke of luck but a matter of strategy and effort.