- Published on
OpenAI executive criticizes Kimi K3 open-source, Silicon Valley refutes from multiple sides
- Authors
- Name
- DeepHorus
On July 19, local time, Chinese AI company Moonshot AI released the Kimi K3 model, which approaches the level of top US models in multiple benchmarks but at significantly lower costs. This progress has once again sparked anxiety in the US about China's rapid AI catch-up and intensified the debate in Silicon Valley over open-source versus closed-source approaches. OpenAI's Director of Strategy, Dean Ball, publicly questioned the Chinese government's permission for such a powerful model to be open-sourced, but faced rebuttals from several Silicon Valley investors and technical experts.
Dean Ball, who previously served as a senior AI advisor to US President Trump, stated on social media that given the potential risks, the Chinese government's open-source policy for a powerful model like Kimi K3 is surprising. He further predicted that the Trump administration might eventually create regulatory uncertainty around the use of China's open-weight models, leading most US companies to avoid adopting them. Ball emphasized that this is only a prediction, not a policy recommendation, and claimed he supports open source until AI develops to a point where it becomes too dangerous.
Ball's remarks quickly drew widespread criticism. Venture capitalist David Sacks accused OpenAI and Anthropic of attempting to use regulatory means to eliminate open-source competitors, calling it "completely unacceptable." Sacks, who served as the first head of AI and cryptocurrency under the Trump administration, argued that leading closed-source labs have formed a duopoly in AI model revenue and now want government intervention. He called on other forces in Silicon Valley to express a stance in favor of maintaining open competition.
Well-known investor Chamath Palihapitiya also publicly supported open source, stating, "The future belongs to open source, and we need to embrace it." Software engineer Suhail Doshi pointed out that US AI labs train models using all of humanity's data without paying for it, and any legislation banning open-weight models under the pretext of "model distillation" would hinder US future innovation. He emphasized that open source does not automatically confer a market advantage; the key lies in operational efficiency.
However, some analysts offered different perspectives. Citrini Research analyst Jukan argued that the competitiveness of Chinese AI companies does not stem from open source itself but from intrinsic operational efficiency. Citing DeepSeek as an example, he noted that the company maintains low token costs due to efficient internal operations, not merely relying on open source. Jukan stated that while Chinese companies may lack computing power, they are not selling at a loss; their business model is sustainable.
Currently, the core conflict in the US-China AI competition focuses on the choice of development path. Chinese companies actively embrace open source or open-weight models, while leading US AI labs like OpenAI and Anthropic adhere to the closed-source approach. The latter argue that if powerful models are fully open, anyone can use them without effective regulation, posing security risks. In contrast, closed systems allow developers to have stronger control over safety mechanisms and commercial pricing.
This debate reflects the divergence in the future direction of the AI industry. On one hand, the open-source model helps lower barriers, promote innovation, and increase accessibility; on the other hand, the closed-source model ensures security, controllability, and commercial benefits. As Chinese model capabilities continue to approach those of the US, the internal struggle within Silicon Valley over how to respond will intensify. Regardless of the eventual outcome, the balance between regulatory policy, technical routes, and commercial interests will become a key issue.
