Chinese AI models now carry up to 46 percent of US enterprise AI traffic

Chinese AI models now handle a startling share of American enterprise AI workloads.

On July 7, 2026, CNBC reported that the share of tokens used by US companies on Chinese AI models via OpenRouter, a platform developers use to access a range of AI models, has sat above 30 percent every week since February 8, 2026, rising as high as 46 percent.

How fast the shift happened

The trendline is the story. The average share across the previous 12 months was just 11 percent. In the first half of 2025, it was only 4.5 percent. In roughly a year, Chinese models went from a rounding error in US enterprise traffic to, at peak, nearly half of it on this platform.

That is not a gradual migration. Something flipped in early 2026, and the share has held above 30 percent every single week since.

Price is the driver

The reason is not mysterious. Open-weight Chinese models like DeepSeek and GLM run 60 to 90 percent cheaper than top OpenAI and Anthropic models. For many production workloads, the quality gap has narrowed enough that the switch pencils out.

This is the quiet logic of enterprise AI spending: benchmarks decide the headlines, but unit economics decide the traffic. When a model that costs a tenth as much is good enough for the job, the job moves.

Why it matters

The AI race is usually framed as a contest over who has the smartest model. This data suggests a second contest that matters just as much: who owns the traffic. US companies are quietly routing more of their token volume east, one cost-optimization decision at a time.

The open questions are real. Enterprise buyers weigh more than price: data governance, compliance and geopolitics all sit on the other side of the ledger. But the OpenRouter numbers show that for a large slice of workloads, those concerns are not stopping the shift. Would you run a Chinese open-weight model in production? For a growing share of US companies, the answer is already yes.