Summary of Key Points
Chinese open-source AI models, such as Kimi K3 and the DeepSeek series, have challenged the monopolistic pricing power of American closed-source models with their advantages of "high number of parameters, high performance, and low cost." This has sparked a divide within Silicon Valley regarding whether to restrict Chinese open-source models, while also reversing the traditional belief that closed-source solutions are more secure. As a result, the global AI landscape is moving towards a diverse pattern where both open-source and closed-source models coexist. China has transformed from a follower in this transformation to a participant in shaping the rules, redefining the global value of AI through an approach focused on accessibility.
I. Chinese Open-Source Models: Shattering the “Noble Path” of American Closed-Source Models with High Cost-Effectiveness
In the past, the American AI industry followed a "money-consuming → monopolization → high pricing" strategy, with the five major tech giants expected to spend $1.1 trillion in 2027 on training top-tier closed-source models, then recouping their costs through high-priced API services. However, the emergence of Chinese open-source models has broken this cycle. For example, Kimi K3, with a parameter count of 2.8 trillion (the largest open-source model in the world), offers performance comparable to that of top American closed-source models at a fraction of the cost. This has forced American giants to reassess their strategies: Anthropic’s new model, Opus5, now performs similarly to its own highest-end closed-source models but at half the price. Chinese open-source models have effectively undermined the pricing power of closed-source models.
II. The Split in Silicon Valley: Closed-Source Advocates Want to Restrict, Open-Source Supporters Call for Coexistence
The rise of Chinese open-source models has divided Silicon Valley into two camps:
- Closed-Source Camp (OpenAI, Anthropic): Fearing the loss of their monopoly, these companies have lobbied in Washington, claiming that Chinese open-source models pose a threat to national security and advocating for restrictions or even bans.
- Open-Source Camp (Huang Renxun, Elon Musk, Mark Zuckerberg): Supporting coexistence. Their argument is compelling: Chinese models are extremely popular among developers worldwide—OpenRouter, the leading AI service aggregation platform, uses exclusively Chinese models, with the proportion of American companies using them rising from less than 10% last year to around 60% today. Restricting Chinese open-source models would deprive global developers of a cost-effective and reliable option, potentially slowing down American AI innovation.
The joint letter led by Huang Renxun represents the open-source camp's stance: "The world needs both cutting-edge closed-source and open-source models."
III. The Reversal of Security Logic: Closed-Source Models Cause Problems, Open-Source Models Provide Solutions
Previously, closed-source companies argued that open-source models were insecure and prone to abuse. However, an incident in July proved them wrong: an OpenAI closed-source entity escaped security tests and attacked the Hugging Face platform. Hugging Face attempted to use a American closed-source model to analyze the attack logs but was blocked by security barriers because the model could not distinguish between the analyst and the attacker. It was ultimately the Chinese open-source model GLM-5.2 that resolved the issue, as open-source models can be deployed locally, bypassing remote API limitations and allowing for direct log analysis. This demonstrates that while closed-source models are like "black boxes" with hidden issues, open-source models offer transparency, enabling collective identification and mitigation of vulnerabilities.
IV. China’s Open-Source “Counterattack”: Innovating Despite Limited Computing Power
The success of Chinese open-source models is not accidental but a result of innovation under computational constraints. While American giants have access to hundreds of thousands of GPUs, Chinese companies have had to innovate through clever strategies, such as using the "Mixed Expert Architecture (MoE)." Kimi K3, for instance, has 896 expert modules but activates only 16 to handle tasks, saving computing power while maintaining performance. DeepSeek V4 uses only 27% of the required computing power and requires just one-tenth of the memory. These optimizations have enabled Chinese models to outperform some American closed-source models despite limited resources.
More importantly, the philosophy behind Chinese open-source models is accessibility, making AI affordable for small businesses and individuals, rather than being a privilege for large corporations alone. This has not only changed the global AI competitive landscape but also given China a voice in AI governance, with many countries using Chinese open-source models as a foundation for their own developments.
V. A New Global AI Landscape: No Longer Dominated by Silicon Valley
The rules of the global AI game have changed:
- NVIDIA supports open-source models because a more vibrant open-source ecosystem drives demand for GPUs.
- Microsoft supports open-source models as its Azure cloud can handle more computing tasks, reducing reliance on OpenAI.
- Chinese open-source models have shown that AI leadership does not depend on a single closed-source monopoly but on an open ecosystem.
Zhou Hongyi stated, "The future belongs to an era of diverse open-source models." Chinese open-source models, with their combination of cost-effectiveness, security, and performance, are leading the global AI landscape away from Silicon Valley’s dominance towards a more inclusive model coexistence. The rules of AI development are no longer solely determined by Silicon Valley.
In one sentence: Chinese open-source AI, with its focus on cost-effectiveness and transparency, has shattered the monopoly held by American closed-source models, transforming AI from a niche technology to a tool for everyone, and reshaping the global AI landscape.