Alibaba's ModelScope and OSChina's MoArk are vying to become Chinese alternatives to Hugging Face; ModelScope hosts 170K+ open models, while MoArk hosts ~20K

If you use AI models, Chinese alternatives to Hugging Face now offer local hosting and support for Chinese hardware.

Chinese tech companies Alibaba and OSChina are developing domestic alternatives to the popular AI model platform Hugging Face. Alibaba's ModelScope, launched in 2022, currently hosts over 170,000 open models and boasts 250 million users. OSChina's MoArk, established in 2023, offers approximately 20,000 models. This push is driven by a desire for a self-reliant AI ecosystem and fears of potential disruptions to access to U.S.-based platforms like Hugging Face due to geopolitical tensions and regulatory actions. While Chinese developers often use VPNs to access global platforms like Hugging Face and GitHub for their extensive model libraries, the domestic platforms aim to provide faster download speeds and ensure AI models are compatible with China's indigenous chip technology. Regulators in China have previously blocked foreign websites, creating an environment where domestic alternatives are encouraged, even as the government tolerates workarounds for developers needing global access.

The emergence of ModelScope and MoArk signifies China's accelerated efforts to build a sovereign AI ecosystem, mirroring its strategy during the internet era. This move is partly a response to geopolitical concerns and the potential for U.S. restrictions on Chinese AI models, pushing for greater technological independence. The platforms aim to serve Chinese developers, offering models optimized for local hardware, a critical factor given U.S. chip export bans.

While these domestic platforms are growing, they still face the challenge of competing with the vast model repositories and established global developer communities of platforms like Hugging Face. Their success will depend on convincing developers of the benefits beyond just faster downloads, potentially through specialized features, easier integration with local infrastructure, and robust support for Chinese AI chip architectures.