(Written by/Editor by Chen Jishen/Zhang Guangkai)
Recently, more and more U.S. AI companies and even AI factories have started publicly claiming to use the Chinese AI model.
On October 22, Airbnb CEO Brian Chesky said in an interview that Airbnb's customer service AI consists of 13 models, and that the company "relies heavily on Alibaba's Qwen, which is better and cheaper than OpenAI's products."
At the same time, he also said that Airbnb has not yet been connected to ChatGPT."We will also use the latest models of OpenAI, but we usually don't use them extensively in production environments because there are faster and cheaper models to choose from. OpenAI's connectivity capabilities are not yet fully ready."
Windsurf, the overseas-headed AI programming product, recently introduced a mysterious model, which is said to be a model designed specifically for speed and Agentic.
Coincidentally, the US cloud service platform Together AI also officially announced the deployment of Qwen-3-Coder in July this year.
And just this month, Chamath Palihapitiya, a well-known investor in Silicon Valley and founder of Social Capital, said bluntly: "We have started using Kimi-K2 on Groq. OpenAI and Anthropic's models are good, but they are too expensive."
Faced with the current situation of major U.S. manufacturers defecting and adopting the China model, a well-known AI researcher sharply asked: "Is Silicon Valley built on Tongyi Thousand Questions?" Behind this is the fact that more and more American companies are no longer concealing their use of China AI.
China's AI accelerates occupation of the United States
Airbnb CEO's statement is the most representative. He directly pointed out Qwen's core advantages of "better and cheaper". This recognition has long spread among Silicon Valley executives.
In May this year, Nvidia’s CEO, Yuan Yong-young, spoke at the earnings call and said that Alibaba’s generic Qi is the best of the open-source AI model. Twitter founder, Jack Dorsey, also sent a letter praising the Qwen 3-Coder code model for the Qwen series. Even Musk said the effect was “difficult to discern” after seeing the images generated based on the Qi Wan 2.2 training.
This influence is rapidly transforming into practical commercial applications. In addition to the Together AI mentioned at the beginning, e-commerce giant Amazon has been exposed to be adopting the Alitong Yiqian Model in its robot control system. It has been repeatedly reported that Apple plans to introduce the Tongyi Thousand Questions model to iPhones and other devices in the China market to support their AI functions.
Expanding its perspective to the world, Japan Economic News also said that General Intelligence has become the foundation for Japanese AI development; the UAE Muhammad bin Zayed University of Artificial Intelligence (MBZUAI) has also built its presidential-backed K2 Think model based on General Intelligence.
Behind this is Tongyi Qianwen's terrible strength in open source: at present, more than 300 models have been open source, and the number of derivative models has exceeded 170,000, forming the world's first open source model matrix. On the latest list of Hugging Face, the world's largest AI open source community, the Ali Tongyi series also achieved the "slaughter list" with 7 models selected into the top ten.
In the developer tool and platform layer, the intelligence GLM penetrates equally quickly. The AI programming product Windsurf, mentioned at the beginning, "hidden" using GLM 4.6, is just the tip of the iceberg.
What’s more, Vercel, a US company valued at $3.3 billion, recently announced that it has partnered with SmartSpot to provide GLM-4.6 API services on its platform. Its owner, Guillermo Rauch, also praised GLM 4.6 as “good, ranked third on http://nextjs.org/evals, or the only open source model in the top five.”
In addition, U.S. AI reasoning service provider Cerebras – the company that released the world’s largest AI chip – has also put GLM-4.6 on its platform for global developers to call, suggesting that China’s big model is being integrated into the U.S. AI development ecosystem as an infrastructure.
If Qwen and GLM demonstrate the breadth of the platform, Kimi directly ignites the fuse of "cost performance" and is even regarded as a "defection" signal from the core of Silicon Valley.
The reason why Chamath Palihapitiya's "defection" is so important is due to his undisputed status in the industry. He is not an ordinary commentator, but a top player worth tens of billions. He led the growth of Facebook users from 45 million to 700 million, and the viral growth feature of "People You May Know" was the work of his team. The Social Capital he founded also made precise bets on Slack, Box, etc. Therefore, when such a top catcher who pursues extreme business rationality publicly replaces OpenAI and Anthropic with Kimi K2, this is already a strong market-leading indicator.
Chamas’s choice is rapidly evolving into a group trend. A series of platforms that have played a major role in the U.S. developer ecosystem have been acting quietly: Vercel: the head of the cloud development platform, has integrated Kimi K2’s API. Cursor: As the star product of AI native code editors, it has made the Kimi set a core option. Perplexity, Genspark, Youware: these widely acclaimed AI native applications have also announced access to Kimi K2.
The "voting" by developers with code is a stronger market proof than any ranking.It marks Kimi's smooth transition from a "external model worthy of attention" to a tool with real production value that is quietly incorporated into the global workflow of developers.
Say goodbye to catch-up, a dual-track parallel AI world
Seeing this, the motivation for American companies to turn around is very clear. This is not just a "price war", but also a profound paradigm shift, which has been attested by authoritative reports.
The State of AI Report 2025, known as the global AI industry’s “wind of the year”, for the first time elevated the “Chinese AI system” from the “outside chaser” to the “parallel competitor”, and made it clear that “in 2025 China will no longer be the chaser – it is setting a pace in open-source AI and commercial deployment.”
This is not an empty praise. When taking stock of the most important technological developments of the year, the report listed only three representative models: OpenAI's o1, and China's DeepSeek-v3 series and Kimi-K2 series. In the world's highest AI research hall, China models already account for two-thirds of the seats.
Behind this are two development paradigms that have evolved in global AI: the United States '"technological summit" paradigm: represented by OpenAI and Anthropic, and with academic and research laboratories as the core, it has devoted all resources to impact AGI and pursued absolute technological heights. China's "application co-prosperity" paradigm: represented by the dark side of the moon, in-depth exploration, and universal questions, it is committed to building an open and prosperous application ecosystem, and adopts cost-effective, fast-iterative open source/open model Empowers thousands of developers and allows innovation to grow from the bottom up. Therefore, the "defection" of Vercel, Airbnb and Chamas burst the bubble of "strongest model theory". They are voting with their feet to choose a cost-effective solution under the "application co-prosperity" paradigm.
This marks that China's development of AI has passed the sequence of "tracking" and started "ecological outcry". a "technology top" and "application communion" dual parallel AI world pattern is gradually unfolding in front of our eyes.