Even as the United States is locked in a fierce competition with China over artificial intelligence, Chinese technology is increasingly entering the United States market, especially through open source models that allow extensive customization. Unlike closed generative artificial intelligence systems such as OpenAI’s ChatGPT or Google’s Gemini, Chinese offerings from companies including Alibaba, DeepSeek, MiniMax, Z.ai and Moonshot AI expose configurable components that appeal to developers and smaller companies. These models are gaining traction largely out of public view, as firms look for cheaper, flexible alternatives that can support many business applications without requiring the most advanced capabilities.
According to a report published in December by developers’ platform OpenRouter and United States venture capital firm Andreessen Horowitz, global use of Chinese-developed open models has risen sharply, increasing from just 1.2% in late 2024 to nearly 30% in August. China’s open source models “are cheap – in some cases free – and they work well”, said Wang Wen of Renmin University of China. One American entrepreneur said their business saves US400,000 (S517,000) annually by using Alibaba’s Qwen artificial intelligence models instead of proprietary systems, while still turning to OpenAI, Anthropic or Google for the most cutting-edge tasks. Major names such as Nvidia, Perplexity and Stanford University are also using Qwen in some of their work, underlining how integrated these Chinese tools have become.
The January launch of DeepSeek’s high-performance, low-cost and open source “R1” large language model challenged assumptions that only United States giants like OpenAI, Anthropic or Google could deliver the best artificial intelligence technology and highlighted how far China has advanced. Chinese models from MiniMax and Z.ai are attracting users overseas, and Chinese firms are joining the race to build artificial intelligence agents that can act on users’ behalf online, helped by “agent-friendly” open source systems like the Kimi K2 model from start-up Moonshot AI released in November. While a Trump administration “AI Action Plan” in July called for “leading open models founded on American values”, many United States companies are moving toward closed systems: Meta has shifted focus away from open source, and OpenAI has only released limited “open-weight” models under pressure to revisit its non-profit roots. In contrast, France’s Mistral is one of the few Western firms still committed to open source, yet it trails DeepSeek and Qwen in usage rankings, and some users say Western open source options are “just not as interesting”.
The Chinese government has encouraged open source artificial intelligence despite questions about profitability, and executives and analysts say technical and security concerns may be overstated. OMNIUX chief technology officer Mark Barton is weighing Qwen but notes some clients are uneasy about using Chinese-made artificial intelligence and worry about political risk, including the possibility of sanctions that could suddenly cut off access. However, DGA-Albright Stonebridge Group partner Paul Triolo said there were no “salient issues” with data security and argued that companies can adopt these models “without any connection to China”. A Stanford study cited in the article concluded that “the very nature of open-model releases enables better scrutiny”, an assessment echoed by Gao Fei, chief technology officer of Chinese artificial intelligence wellness platform BOK Health, who said that the transparency and sharing inherent in open source are among the best ways to build trust in the technology.
