China’s low-cost open-source AI sweeps Silicon Valley, outweighing US’s security concerns

A growing body of evidence suggests that China’s open-source artificial intelligence (AI) models and developer ecosystem are gaining traction worldwide, raising new questions about the future competitive landscape.

According to Bloomberg, downloads of Chinese models on the developer platform Hugging Face have surged past those of US alternatives, with China-based Alibaba Group Holding Ltd.’s “Qwen” models accumulating roughly 385.3 million downloads compared with the 346.2 million for Llama from US-based Meta Platforms Inc.. Chinese-origin derivatives now account for more than 40 per cent of new language-model releases on Hugging Face, while Meta’s share has fallen to about 15 per cent.

Industry participants say the shift is already influencing US-based companies. Venture capitalist Chamath Palihapitiya recently said on his influential All-In podcast that the company he works for has moved major workloads to a Chinese open-source model developed by Moonshot AI (its “Kimi K2” model), citing cost advantages.

Meanwhile, the Airbnb Inc. CEO admitted his company did not integrate with OpenAI’s ChatGPT because the connectivity tools “weren’t quite ready” and instead relies heavily on Qwen models from Alibaba, describing them as “very good. It’s also fast and cheap.”

Hardware advantages remain, but momentum is shifting

The US retains advantages in high-end AI hardware, particularly in access to cutting-edge chips and computing infrastructure. US firms such as Nvidia Corp. continue to dominate that domain. For example, Nvidia CEO Jensen Huang has emphasised the company’s belief that America must “race ahead and win developers worldwide” even as he acknowledged that China is “just nanoseconds behind.”

However, experts say that for many real-world AI applications, especially those built by smaller players or startups, the cost, licensing flexibility and accessibility of open-source Chinese models can be more appealing than exclusive access to proprietary US systems. One coalition supporting open-source AI found that, globally, developers are increasingly choosing Chinese models for the ease of download, fine‐tuning and local deployment.

The trend raises strategic concerns beyond cost. Some analysts worry about dependency on foreign models and implications for data governance, but companies racing to market appear increasingly willing to trade those concerns for performance and affordability.

Although it is too early to declare a decisive outcome in the global AI race, the emergent pattern signals a potential shift. US policymakers and industry executives may need to address why Silicon Valley and global developers are increasingly turning to alternatives. As one expert put it, the question is no longer just “who has the best hardware?” but “which ecosystem do developers build on?”