Chinese AI models just captured 46% of US enterprise token usage. Your AI vendor is losing the cost war.

46% of US enterprise API tokens are now flowing through Chinese models.

Not in some niche developer community. On OpenRouter — the platform powering production workloads for thousands of US companies.

Up from 11% average over the prior 12 months. Up from 4.5% in early 2025.

The math is brutal: DeepSeek V4 Flash costs $0.28 per million output tokens. GPT-5.5 costs $30.

That's a 100x gap. And when a single agentic task calls the model 50 to 200 times, the gap compounds with every loop.

GLM-5.2 from Z.ai scores 62.1% on SWE-bench Pro — beating GPT-5.5's 58.6% — at one-sixth the cost.

Lindy moved 100% of its traffic from Anthropic's Claude to DeepSeek. CEO said the cost curve "crashed to the ground."

The House Select Committee on China and House Homeland Security launched a joint investigation into companies using Chinese AI — Airbnb and Cursor's parent Anysphere are both under scrutiny.

Meanwhile, China's Ministry of Commerce is in talks with Alibaba, ByteDance, and Z.ai about restricting overseas access to their most advanced models.

Your AI stack is now a geopolitical risk surface. Audit every model in your production pipeline today.

SOURCE: https://www.cnbc.com/2026/07/07/chinese-ai-models-costs-us-openai-anthropic.html
VERIFIED: CNBC (July 7, 2026), ByteIota (July 9, 2026), IBTimes Singapore (July 9, 2026)
SIGNAL: US frontier AI labs are pricing themselves out of the middle tier of enterprise workflows. The cost math forces rational engineers to route to Chinese models — and both governments are now moving to control which models you can use.

---