(By Chen Jishen, Editor: Zhang Guangkai)

Google's former CEO Eric Schmidt said yesterday that he is concerned that most countries may end up using China's AI models due to cost issues.

This is another warning from a tech industry figure regarding the development of U.S. AI, especially its closed-source strategy, following last week's statement by NVIDIA CEO Jensen Huang that "China will win the AI competition."

Last Tuesday, in an interview on a podcast called "Moonshot," Schmidt candidly stated, "There is a 'strange paradox' in the current global AI landscape. The largest AI model in the U.S. is closed-source and paid, while the largest AI model in China is open-source and free."

According to Schmidt, "Open-source AI models allow anyone to use and share the software freely and publicly for any purpose. Therefore, most governments and countries without the financial resources like the West will ultimately adopt China's models, not necessarily because they are better, but because they are free."

From 2001 to 2015, Schmidt served as Google's CEO, leading the company through its initial public offering (IPO) in 2004. He is currently a founding partner at the venture capital firm Innovation Endeavours and runs a space startup called Relativity Space. According to Bloomberg, his net worth is close to $5 billion.

Notably, Google's core large model products, Gemini, and OpenAI's GPT series are both closed-source large models. Schmidt's direct "criticism" of his former employer has drawn attention from the industry.

In the early development of large models, open-sourcing was the mainstream choice in the industry. Even OpenAI initially released GPT1 and GPT2 models as open-source.

After the rise of ChatGPT, OpenAI took the lead in closing its source code, prompting many American AI companies to follow suit. Although Elon Musk's Grok and Meta's Llama are open-source large models, they have also closed their latest models, only open-sourcing previous generations, which has led to a significant performance gap between their open-source models and the latest closed-source models.

In China, there were initially both open-source and closed-source factions, but with the development of the industry, especially the emergence of DeepSeek, the Chinese industry has formed a general consensus to embrace the open-source ecosystem.

Regarding this, overseas open-source supporters believe that open-sourcing can promote technological development in a democratic way and offers significant cost advantages, as anyone can modify and redistribute the code. On the other hand, advocates of closed-source models argue that since the code is not public, it is more secure, and closed-source models are relatively more advanced than open-source models.

However, with the rise of Chinese open-source models, the security advantage that the closed-source faction once boasted no longer holds water after open-source models can be locally deployed. As for performance advantages, closed-source models are increasingly being challenged by open-source large models, and their high prices are powerless against the open-source army. Jensen Huang's recent statement that "China's strength in AI is only a few nanoseconds behind the U.S." is not an exaggeration.

This year, Chinese models such as DeepSeek, Alibaba's Tongyi Qwen, Zhipu, Kimi, Minimax, and others have embraced open-source and continuously updated their large models, maintaining a top-tier position in performance and enjoying widespread user popularity. This has raised concerns about the U.S. AI competitive advantage.

Observatory Network reported earlier that Silicon Valley giants such as Airbnb have openly stated that they are using Chinese open-source large models and have praised them for their quality and affordability.

The leading domestic open-source models that have amazed Silicon Valley figures and global developers continue to iterate and evolve at the speed of China.

On October 27, MiniMax released and open-sourced its new text large model, MiniMax-M2. The model ranked in the top five globally on the authoritative evaluation list Artificial Analysis and was the first among open-source models. It competed with American closed-source models such as OpenAI, Anthropic, and Google without falling short.

On November 6, the release of the Kimi K2 Thinking model sparked attention both domestically and internationally. This open-source large model launched by Moonshot achieved a score surpassing GPT-5 and topped the large model ranking. It is rumored to have a development cost of only $4.6 million and extremely low API costs, causing real pressure on U.S. closed-source model companies that require training costs in the hundreds of billions of dollars.

On November 12, in the latest Code Arena rankings focused on AI programming performance, the Glm-4.6 model released by Zhipu in September was tied for first place with GPT-5 and Claude models.

Besides benchmarks, Chinese models have also captured the minds of global developers, including those in the U.S., in terms of application.

In May, the Japan Economic News reported that many Japanese companies have adopted the open-source Qwen model to build their own models and applications. Qwen has become one of the most important foundations for AI technology in Japan.

According to Hugging Face statistics, in terms of historical download volume, Alibaba's Qwen surpassed Meta's Llama in October, becoming the most popular open-source large model globally.

Aside from Qwen, the latest model SWE-1.5 of the foreign AI programming product Windsurf was found to be a customized model based on Zhipu's GLM-4.6.

Notably, Windsurf was acquired by OpenAI for $3 billion this May. A U.S. company does not use its parent company OpenAI's large model but instead chooses a Chinese open-source model as the foundation of its product, indicating that Chinese models are no longer just a cost-effective option, but one of the preferred choices in the industry for their quality, affordability, and excellence.

Regarding the current situation where Chinese models not only dominate the open-source first tier but even exceed some top closed-source models, Matt Stoller, a researcher at the U.S. economic freedom organization, expressed concern about the closed-source strategy of U.S. AI companies on social media. He believes that the core reason why China is currently leading in artificial intelligence is due to the excessive reliance of the U.S. AI industry on closed-source monopolies. This approach carries high risks, and if it fails, the consequences would be catastrophic.

With the continuous advancement of open-source models led by China, like Schmidt, NVIDIA CEO Jensen Huang and Mistral CEO Arthur Mensch have also publicly expressed support for open-source models.

At the Chain Blockchain Conference in July this year, Huang mentioned DeepSeek, Alibaba, Tencent, MiniMax, and Baidu's open-source models, which are driving global AI development. He called China's open-source initiative a catalyst for global progress, allowing every country and industry to join the AI revolution. At last month's NVIDIA GTC conference, Huang reiterated that open-source models and open collaboration are the cornerstone of global innovation.

Analytical experts believe that the vigorous development of China in the field of open-source models is reshaping the global AI competition landscape. This trend has already caused concerns among Americans: under the current complex geopolitical context, more and more countries may turn to Chinese open-source models due to their advantages of being open-source, safe, and low-cost, further evolving the global AI landscape.

Original article: https://www.toutiao.com/article/7572191871145902626/

Statement: This article represents the views of the author. Please express your opinion below using the [Up/Down] buttons.