How much of the world's AI runs on Chinese models?
The share of routed LLM tokens served by Chinese-lab models — DeepSeek, Alibaba/Qwen, MiniMax, Moonshot, Zhipu, Tencent. Up from 18% to ~58% since September 2025.
The share of routed LLM tokens served by Chinese-lab models — DeepSeek, Alibaba/Qwen, MiniMax, Moonshot, Zhipu, Tencent. Up from 18% to ~58% since September 2025.
The DeepSeek-moment as a settleable number — each month the arithmetic mean of the daily Chinese-lab share. Each month settles to the arithmetic mean of the daily index (Asian-style, like the GPU-compute futures), finalized on the 1st.
Live, settleable, free — the Chinese-lab share in one number.