U.S. Companies Accelerate Adoption of Chinese AI Models as OpenAI and Anthropic Costs Surge

My take: Data circulating this week confirms what many builders already suspected: the price gap between Chinese and U.S. AI models is too wide to ignore in production. Leading Chinese models now charge around 18 cents per million input tokens, compared to roughly $4 for U.S. frontier models, a difference of more than 20 times. GLM 5.2, Z.ai's model, grew its daily token volume 27x and its customer count 80x in a single first week. Startups like Lindy AI have already moved 100% of their traffic from Claude to DeepSeek, citing savings equivalent to millions of dollars.

The logic behind these moves is not ideological: it is economic. When a task does not require the best available model, the team with the most AI maturity routes it to the cheapest model that solves the problem. That is already standard practice in the teams that use their AI budgets most efficiently. Maturity is not measured by using the most powerful models for everything, but by knowing when not to.

The real complicating factor is different in nature. The U.S. Congress is already investigating Airbnb and Anysphere for disclosing the use of Chinese open models like Qwen and Kimi. Data sovereignty, regulatory, and geopolitical risk questions are not hypothetical: they are part of the analysis that any model routing decision requires today. Price and performance matter, but regulatory and reputational exposure matter too.

Does your company have a defined policy on which data or tasks can be processed by models from providers outside the U.S.?