Last week, Nvidia CEO Jensen Huang stated, “China is about to win the AI war,” emphasizing that “the United States must speed up to win over developers worldwide first,” because “the battle to attract developers may already be slipping out of America's hands.”
This remark further clarifies the current market situation, as recently the U.S. AI market has begun shifting focus toward using Chinese open-source models due to their lower costs. Even within Silicon Valley, many large and small tech companies are adopting these models.
Open-source refers to software, programs, or code that anyone can access, study, modify, distribute, or use freely as they wish.
Previously, Airbnb CEO Brian Chesky acknowledged that Airbnb no longer connects its travel app system to OpenAI’s ChatGPT because the API integration was not sufficiently ready. He revealed that Airbnb’s AI Agent uses dozens of AI models, heavily relying on Alibaba’s Qwen model. Chesky praised it as “very good, fast, and affordable.”
More Silicon Valley tech companies are openly admitting this trend. For example, Thinking Machines Lab, a startup founded by Mira Murati, former CTO of OpenAI, wrote on its company blog that recent research was inspired by the Qwen team and that products are being developed on Alibaba’s Qwen3.
Famous investor Chamath Palihapitiya revealed on the All-In podcast by David Sacks—currently an AI and crypto advisor to the White House—that a company he works with shifted its main workload to the Kimi K2 model from China’s Moonshot AI, confirming that this model is “much cheaper than OpenAI and Anthropic.”
Related news:
Other subtle signs are also increasingly evident. According to a Bloomberg report, some have noticed that startups like Cursor—a $10 billion AI coding startup—released a new AI version that users observed appeared to use China’s DeepSeek model because the system momentarily switched its processing language to Mandarin Chinese.
Similarly, Cognition AI Inc., another $10 billion startup, is widely believed to be using China’s Zhipu AI, also known as Z.ai, especially after Z.ai tweeted, “Open-source models will benefit the global AI ecosystem.”
Data from Hugging Face, a global AI model platform popular among developers, confirms this trend: in early 2024, Meta’s Llama model was downloaded around 10.6 million times, while Alibaba’s Qwen model had only 500,000 downloads.
However, in less than a year, this reversed. Last month, Qwen’s cumulative downloads exceeded 385.3 million, surpassing Llama’s 346.2 million. New models built upon Qwen account for over 40% of all models on Hugging Face, while Meta’s share has dropped to just 15%.
Despite ongoing geopolitical concerns between the U.S. and China, including fears that foreign users might be “indoctrinated by the Chinese Communist Party” through these AI models,
developers racing against time to launch products—especially in coding and software development—care more about cost and model performance. Moreover, open-source models can be downloaded, customized, and run locally, which reduces data privacy and security risks.
Source:Bloomberg
Follow the Facebook page Thairath Money at the following link -https://www.facebook.com/ThairathMoney