Startups and enterprises are increasingly adopting open-source AI models from Chinese developers such as DeepSeek, Moonshot AI, and Alibaba. This shift is driven by rising costs associated with US-based AI providers, where some enterprise bills have reportedly increased by up to 100 times.
Data from the OpenRouter platform shows that Chinese models accounted for over 30% of weekly tokens processed since February 8, 2026. By mid-September 2026, this share reached a peak of 67%. Chinese models are reported to be 10 to 50 times cheaper per token than US alternatives, with some prices being 90% lower.
The startup Lindy transitioned from Anthropic’s Claude to DeepSeek’s V4 model in June 2026, resulting in an approximate 90% reduction in inference costs. Other major companies are diversifying their model usage to avoid vendor lock-in: DoorDash utilizes Moonshot AI’s Kimi, Airbnb works with Alibaba’s Qwen, and Coinbase uses models including GLM-5.2.
Chinese labs have focused on creating efficient, open-weight models despite US export controls on advanced chips. While these models offer significant cost savings, businesses are also evaluating potential geopolitical and regulatory risks associated with relying on technology from Chinese developers.
