Updated
Updated · CNBC · Jul 7
U.S. Companies Shift Up to 46% of AI Tokens to Chinese Models as Costs Surge
Updated
Updated · CNBC · Jul 7

U.S. Companies Shift Up to 46% of AI Tokens to Chinese Models as Costs Surge

3 articles · Updated · CNBC · Jul 7

Summary

  • OpenRouter data show Chinese models have accounted for more than 30% of U.S. companies' weekly token usage since Feb. 8, peaking at 46%, versus a 12-month average of 11%.
  • Prices are driving the shift: Chinese open-source and open-weight models can be 60% to 90% cheaper than top OpenAI and Anthropic offerings while closing the performance gap to roughly six to nine months.
  • Lindy moved 100% of its traffic from Anthropic's Claude to DeepSeek in June and said the switch will save millions of dollars within months while improving many core use cases.
  • Z.ai's GLM 5.2 is gaining fast traction too, with Vercel saying first-week daily token volume rose about 27-fold and customer count about 80-fold; LaunchLemonade now ranks it among its top five models.
  • The shift comes as Washington weighs tighter controls on advanced AI: OpenAI delayed a late-June model rollout at the government's request, underscoring the appeal of cheaper overseas alternatives.

Insights

With AI models becoming cheap commodities, what are the hidden risks for companies relying on foreign tech?
As AI model costs plummet, where will the real profits be made in the new AI economy?
Did US chip bans inadvertently create a more efficient and competitive Chinese AI industry?