Summary of Key Points
This article focuses on an open letter co-signed by 70 American technology companies, led by NVIDIA, discussing the intense debate within the U.S. industry and government regarding the issue of whether to ban Chinese open-source large-scale AI models. Chinese open-source models (such as Kimi K3, Alibaba Qwen, and Zhipu GLM) have rapidly gained momentum in the U.S. market due to their advantages of performance close to the forefront, extremely low prices, and the ability for local deployment, challenging the monopolistic interests of closed-source giants like OpenAI and Anthropic. The closed-source camp attempts to suppress these Chinese models through government regulation, but the U.S. industry (especially small and medium-sized enterprises that rely on open-source technology and developers) as well as some government officials (such as those from the Department of Defense and the Technology Advisory Committee) strongly oppose this approach. As a result, 70 companies jointly called for not banning open-source models. This debate essentially reflects the rise of China's AI competitiveness, putting the United States in a dilemma where it wants to use these models but is afraid to do so, and at the same time, it cannot completely prevent their adoption.
1. The Closed-Source Giants Are Panicking: Trying to Kill Chinese Open-Source Models Through Regulation
The panic among the closed-source camp (OpenAI, Anthropic) stems from the fact that Chinese open-source models are directly threatening their business models.
- OpenAI's Counterattack: Paul, OpenAI's strategic director and former AI policy advisor to the White House, wrote an article claiming that the open-source world is a dystopian nightmare and suggested that the government create regulatory uncertainty (by making it seem possible at any time that Chinese models could be banned) to deter American companies from using them. His motivation is straightforward: OpenAI earns revenue by charging for its closed-source model APIs, while Chinese open-source models like K3, which are among the top three in performance globally and are inexpensive, are directly competing with their services.
- Anthropic's Underhanded Tactics: Anthropic accused the Chinese K3 model of "distilling" its own Fable 5 technology and even claimed to have provided "classified information" to the White House. Essentially, they are trying to use the government to eliminate their competitors, as Anthropic relies entirely on the scarcity of its closed-source models for its nearly trillion-dollar business.
- The Rotation of Power Between Government and Business: Paul's rapid transition from a White House policy advisor to an OpenAI executive exposes the closed-source camp's attempt to "hijack regulation" to suppress their rivals, a move that was directly criticized by the chairman of the Technology Advisory Committee as "regulatory capture."
2. Chinese Models in the U.S.: Both a Threat and a Lifeline
The joint statement from 70 companies highlights the dual impact of Chinese open-source models on the U.S. market:
- Threat: Chinese models have become dominant in the U.S.: On the world's largest model hosting platform, Hugging Face, Chinese models account for over 40% of downloads (for the first time surpassing those from the United States); the Alibaba Qwen family of models accounts for more than half of the global market share, with over 200,000 derived models; and in June, the top five most used models on OpenRouter were all Chinese models. Chinese models have essentially established themselves as a dominant force in the U.S. market.
- Benefit: These models are a lifeline for small and medium-sized enterprises: Many American companies and developers cannot afford closed-source models (for example, a five-person law firm might spend thousands of dollars per month on closed-source APIs, while Chinese open-source models cost only a few dozen dollars). More importantly, Chinese models can be deployed locally, addressing the issue of data security that is a concern with closed-source models. In fact, when Hugging Face was attacked by an AI, it was the Chinese Zhipu GLM-5.2 that helped defend the platform by analyzing 17,000 logs. Chinese models have become a necessity for American businesses.
3. The Real Motives Behind Silicon Valley Companies' Signatures
The decision to sign the letter is not based on sentiment alone but on practical self-interest:
- NVIDIA: As a company that sells GPUs, the more people use open-source models, the fewer GPUs it can sell. Jensen Huang explicitly stated that American companies should use Chinese models to protect its $26 billion investment in open-source technology.
- Microsoft: As a cloud service provider, Microsoft is both a shareholder in OpenAI (receiving profits from closed-source models) and hosts all models (earning rent from open-source models). Signing the letter is a way to ensure profitability regardless of whether closed-source or open-source models prevail.
- Meta: Having just made its flagship model closed-source, Meta signed the letter to rehabilitate the reputation of distillation technology and to clear legal risks.
- OpenAI: By signing, OpenAI avoids being isolated; the joint statement represents a majority opinion, and not signing would mean it would be seen as part of the "regulatory capture" movement. Since it already has a low-performance open-source product line, signing does not incur any loss for them.
- Anthropic: The only closed-source giant to refuse to sign, Anthropic relies entirely on its closed-source models for revenue and is releasing new models in an attempt to extend its lead by improving their performance enough that users cannot easily switch.
4. The U.S.'s Dilemma: Paying Taxes to Use Chinese Models While Fearing Being Held Back
The U.S. has a contradictory attitude towards Chinese open-source models: it wants to use them but also fears the consequences:
- Want to Use Them: The 70 companies are defending the right to use these models, as they provide a low-cost alternative for American small and medium-sized enterprises. Cutting off access to these models would be equivalent to imposing an "intelligence tax" on domestic businesses.
- Fear of Consequences: The Treasury Department is investigating, the White House is issuing accusations, and there are threats of adding Chinese companies to sanctions lists, all out of concerns about the potential impact on national security if these models penetrate the U.S. ecosystem.
- Concerns About Rapid Development: There are allegations of "industrial-scale model distillation," but even Paul acknowledges that the capabilities of K3 cannot be replicated through such methods, indicating a reluctance to accept China's innovative AI capabilities.
The root of this contradiction is that Chinese models are no longer just cheap alternatives; they have reached a level of performance that matches or surpasses that of U.S.-developed open-source models, and the U.S. itself has lost its ability to rely on its own open-source ecosystem. Without using Chinese models, it faces difficulties, but using them exposes it to risks.
5. The Rise of Chinese AI: Not Copycatting, but Genuine Strength
The success of Chinese open-source models is due to a combination of innovation and cost-effectiveness:
- Technological Innovation: Features like K3's KDA hybrid linear attention mechanism (published eight months ahead of schedule) and AttnRes technology (which increases efficiency by 25% at a 2% cost increase) demonstrate original research, not just imitation.
- Cost-Effectiveness: Similar to how China transformed the home appliance, smartphone, and electric vehicle industries, Chinese AI combines cutting-edge performance with low prices, reshaping the landscape of the intelligent technology sector.
- Policy Support: Four Chinese departments are encouraging the adoption of "Token as a Service" policies, while U.S. policies are often volatile (for example, the Reagan administration dismantled Carter's solar energy initiatives). The continuity of Chinese policy support has also been crucial to the rise of AI.
This debate ultimately confirms that China's AI has risen to a point where its competitors must publicly struggle with the decision of whether to adopt or prevent it. The 70 companies' joint statement represents a late acknowledgment of China's AI achievements. This time, it is the Chinese who are knocking on the door.
Conclusion
The U.S.'s hesitation reflects the reality of China's AI competitiveness. As Chinese models have evolved from being cheap alternatives to becoming high-performance, essential tools, the U.S. finds itself in a difficult position. This debate is not the end but the beginning of China's global dominance in AI. Over the past three decades, China has redefined multiple industries with its combination of quality and affordability; today, this same pattern is playing out in the AI sector. Only this time, the stakes are higher, and the competition is more intense, but China is ready for the challenge.