Chinese open-weight AI models are spreading quickly worldwide

In short: Developers and companies around the world are increasingly using Chinese “open-weight” AI models because they can download them, run them themselves, and modify them.

What's going on

Several Chinese AI labs, including Alibaba (Qwen), DeepSeek, and Zhipu (GLM), have released powerful models with “open weights.” That means the core model files are available for anyone to download, like getting a recipe you can cook in your own kitchen instead of ordering takeout. In many cases, the full training data and all the code are not shared, so these models are not fully open source in the strictest sense.

This approach is gaining traction fast. Reports cited by The New York Times say Alibaba’s Qwen family has been downloaded hundreds of millions of times and has spawned well over 100,000 modified versions made by other people. DeepSeek’s models have also climbed popular rankings used by developers, and companies like Nvidia and Amazon now offer DeepSeek models in their own services.

The adoption is not limited to small teams. The reporting describes large companies using these models in real products, often because they are cheaper to run and easier to control. For example, Shopify reportedly moved to Qwen and saved about $5 million per year, and Airbnb’s CEO has said the company uses Qwen because it is “very good, fast, and cheap.”

What to watch

A key issue is where the model is hosted. Surveys cited in the reporting suggest many users are open to using Chinese models if they are run outside China. Another thing to watch is licensing, since “open-weight” does not always mean you can do anything you want, and companies still need to read each model’s rules.

Source: NYTimes