Chinese-developed AI models accounted for 41% of downloads on Hugging Face over the past year, overtaking models from the United States in both monthly and overall downloads, according to the platform’s Spring 2026 report on the open-source AI ecosystem.
The result makes China the largest identified national source of model downloads on Hugging Face, a major global repository for AI models, datasets and developer tools. It is a meaningful shift in the distribution of open models, though it should not be read as a measure of all global AI usage, commercial deployment or frontier-model capability.
Hugging Face said Chinese models had quickly reached a plurality of downloads as Chinese companies accelerated open releases in 2025. Baidu, which had no model releases on the Hub in 2024, published more than 100 in 2025, while ByteDance and Tencent increased their releases by eight to nine times, according to the report. It also found that a majority of newly created trending models in 2025 were either developed in China or derived from Chinese models.
The rise is not confined to a single company. Alibaba’s Qwen family has become a major base layer for downstream development: Hugging Face counted more than 113,000 derivative models from Alibaba alone, more than Google and Meta combined. When all models carrying a Qwen tag are included, the total exceeds 200,000.
That matters because open-weight models can be downloaded, fine-tuned and deployed locally, subject to their individual licences. For developers and public institutions, the model layer can therefore be adapted to local languages, data and compliance requirements without relying exclusively on a foreign cloud API. Hugging Face said the trend has made open models increasingly relevant to debates over technological sovereignty.
The data, however, has clear limits. A download does not prove that a model is used in production, nor does it establish superiority on capability or safety benchmarks. The platform itself notes that usage is highly concentrated: its 200 most-downloaded models account for 49.6% of all downloads, while smaller models and automated systems can materially inflate download counts.
The more durable signal is ecosystem reach. China’s AI companies are releasing models often enough, and at enough scales and specializations, to shape what global developers can experiment with and build upon. The next test will be whether that distribution lead translates into sustained international use through better documentation, multilingual support, reliable deployment tools and licences that developers can confidently use in commercial settings.
For now, the result marks a redistribution of influence in open AI. American companies remain central to the ecosystem, particularly through firms such as Meta and Google, but Chinese model developers have become its most active source of downloadable model weights on one of the world’s most important AI hubs.
