Chinese AI models surge in Silicon Valley, undercutting US rivals and exposing export control limits

Chinese artificial intelligence models are rapidly gaining traction in Silicon Valley, quietly powering American startups and even major platforms, and in the process, revealing the limits of Washington’s effort to choke China’s tech rise.

Developers such as Alibaba’s Qwen, Z.ai, Moonshot, and MiniMax are reshaping the AI ecosystem by offering so-called “open” language models at a fraction of the cost of their American counterparts.

Airbnb CEO Brian Chesky raised eyebrows in October when he revealed that his company had opted for Alibaba’s Qwen over OpenAI’s ChatGPT, calling it “fast and cheap.” Around the same time, venture capitalist Chamath Palihapitiya said his firm, Social Capital, had moved much of its workflow to Moonshot’s Kimi K2 model, praising it as “way more performant” and “a ton cheaper” than OpenAI or Anthropic.

Data from OpenRouter, a platform linking developers with AI models, shows Chinese tools now occupy seven of the top 20 most-used models, including DeepSeek’s V3.2, Z.ai’s GLM 4.6, and MiniMax’s M2. Four of the top 10 models for programming were also Chinese.

China’s open-weight models, whose parameters are publicly available, have been downloaded over 540 million times, according to Atom Project’s analysis of Hugging Face data.

The appeal is partly economic. Chinese firms have mastered the art of doing more with less, training advanced models on older-generation chips that aren’t subject to US export controls, dramatically lowering their costs. An analysis by AllianceBernstein found that DeepSeek’s pricing was up to 40 times cheaper than OpenAI’s earlier this year.

Analysts say the divide between “premium” US AI models and China’s cheaper open alternatives is becoming a class split in the industry: wealthy corporations can afford closed, high-end systems, while startups are turning eastward.