Enterprise AI in Numbers · Capability and cost
Share of open AI model downloads from Chinese labs
61.5%
61.5% as of October 25, 2026, according to Hugging Face Hub download counts, our aggregation. Share of those models' downloads that belong to models developed by Chinese labs; a fine-tune or compressed copy counts for the lab of its base model. Parts give every region.
What it means for leaders
Chinese models such as Qwen and DeepSeek have become the default base for open-weight AI. If your teams fine-tune or self-host open models, set a clear policy on model provenance, licensing and security review.
What would change the reading: A US or European family regaining the lead in downloads.
The breakdown
Every reading
| Date | Value | Change | Source |
|---|---|---|---|
| 61.5% | — | Source |
Definition and method
- What it measures
- Share of those models' downloads that belong to models developed by Chinese labs; a fine-tune or compressed copy counts for the lab of its base model. Parts give every region.
- Population
- The 100 most-downloaded text-generation models on the Hugging Face Hub, over the last 30 days
- Updated
- Each update (rolling 30 days)
- Source
- Hugging Face Hub download counts, our aggregation · CC BY 4.0
Caveats
- Downloads include automated pipelines and are not deployments.
- Models hosted only elsewhere, or behind APIs, are not counted.
Cite this
Enterprise AI in Numbers, "Share of open AI model downloads from Chinese labs", https://lucianoayres.com/ai-brief/data/open-model-downloads-by-origin/, retrieved October 25, 2026. Data: Hugging Face Hub download counts, our aggregation.