Startups Turn to Cheaper Chinese AI Models Amid US Pricing Pressure

The AI industry is seeing a major shift as startups and enterprises move workloads to cheaper Chinese AI models due to pricing pressure from US providers. This trend is evident on OpenRouter, a platform for accessing AI models, where Chinese models have accounted for over 30% of weekly tokens since February 8, 2026. Their share peaked at 67% in mid-September 2026, indicating a substantial shift in AI workload spending.

Chinese models are significantly cheaper, costing 10 to 50 times less per token than US offerings, with some cases showing a 90% price reduction. In contrast, US providers like OpenAI and Anthropic have seen enterprise AI costs surge, with some bills reportedly increasing up to 100 times.

Companies like Lindy have fully transitioned from Anthropic’s Claude models to DeepSeek’s V4, cutting inference costs by approximately 90%. Larger firms such as DoorDash, Airbnb, and Coinbase are diversifying their AI model usage, incorporating options like Moonshot AI’s Kimi, Alibaba’s Qwen, and GLM-5.2.

The shift is partly driven by US export controls on advanced chips, which pushed Chinese labs to develop more efficient and affordably priced models. The rise of open-weight models allows companies to download and run them locally, reducing reliance on API-based access.

For enterprises, the key takeaway is to avoid vendor lock-in and maintain flexibility to switch between models. However, relying on Chinese models also introduces geopolitical and regulatory risks, particularly amid ongoing US-China tech tensions. The figure to watch remains OpenRouter’s token share, which has stayed above 30% since February 2026.