American startups are heavily relying on free Chinese AI models

Technical experts are raising alarms that the US artificial intelligence landscape faces a growing challenge as American startups increasingly adopt free and customizable open-source AI models from China.

Experts, including Misha Laskin, a former Google machine learning engineer, note that Chinese systems are rapidly closing the capability gap with proprietary US models.

While American companies like OpenAI and Anthropic still lead in AI development with their closed-source systems—GPT-5 and Claude—many Chinese open-source models, such as DeepSeek’s R1 and Alibaba’s Qwen, are being used for numerous commercial applications.

Airbnb, for example, heavily relies on Qwen, according to CEO Brian Chesky. Startup founders report that running these Chinese models on their own hardware is often faster and cheaper than using their closed and costly American counterparts.

Productivity app Dayflow’s founder, Jerry Liu, notes that using a closed-AI model can cost his company up to $1,000 per person, while Qwen is free.

The appeal of Chinese open-source models extends beyond performance and cost. They have abundant online training resources and community support that further encourage developer adoption.

Risks and political barriers persist
Critics point to the possibility of Chinese models heavily borrowing from American groundwork, raising questions about their sustained long-term innovation.