Chinese AI models seize about 45% of US enterprise token usage

Data from the OpenRouter API marketplace show that Chinese‑origin AI models accounted for roughly 45 % of the token volume used by US‑based enterprises by mid‑2026, up from just over 11 % at the start of the year. Models such as DeepSeek V4 Flash, Alibaba’s Qwen and Tencent’s Hy3 are priced 60–90 % lower than leading US offerings, delivering comparable performance at a fraction of the cost. The price gap has prompted firms including Shopify, Airbnb and a startup named Lindy to switch large workloads to Chinese models, saving millions of dollars.

The rapid shift has drawn attention from US policymakers. A State Department spokesperson warned that reliance on Chinese AI could pose national‑security risks, and two House committees (Homeland Security and the Select Committee on China) have opened investigations, sending letters to companies like Airbnb and Cursor about potential exposure. At the same time, OpenAI released a set of partnership principles restricting the use of its models for mass surveillance, autonomous weapons and high‑impact decisions without human oversight.

Analysts note that the trend reflects both cost‑driven market dynamics and growing geopolitical tension over AI supply chains, suggesting that US firms may face tighter regulatory scrutiny while Chinese providers continue to expand their foothold in the global enterprise AI market.