Member-only story
Chinese AI models peaked at 46% of US enterprise token usage in a single week this summer, according to a CNBC investigation of OpenRouter traffic published July 7. Eighteen months ago that number was 4.5%. The twelve-month average is 11%, and for every single week since February 8, 2026, Chinese-origin models have carried at least 30% of enterprise token volume on the largest neutral LLM router in the world.
I’ve spent the past week digging into why — and the answer is not patriotism fatigue or benchmark worship. It’s arithmetic. GPT-5.6 Sol costs $5 per million input tokens and $30 per million output. DeepSeek V4 Flash costs $0.14 and $0.28. That is a 35x gap on input and a 107x gap on output, for a model that handles a 1-million-token context window and lands within a few points of the frontier on most workloads that actually run in production.
When the price gap between “good enough” and “flagship” hits two orders of magnitude, the market stops being sentimental. Here’s what’s actually happening, who’s doing it, where the real risks are — and how to test the economics on your own workload in five minutes.