For weeks this summer, the AI world was transfixed by Anthropic's latest release and Washington's struggles to control access. But while everyone watched the frontier, developers kept building—without waiting for permission.

Chinese open-source models now dominate on platforms like Hugging Face, with major firms such as Tencent and Xiaomi leading the charge. These open models are not only more accessible but also cheaper, making them a popular choice among businesses. As this trend grows, some wonder if the cutting-edge models will be reserved for high-value tasks while others run on customizable, cost-effective alternatives.

For Hugging Face CEO Clem Delangue, this shift represents a significant change in how companies view AI. He believes that owning and customising their own models gives businesses more control and reduces dependency on black box APIs. This, he argues, is reflected in the rapid growth of model repositories on his platform.

The rise of these open-source alternatives coincides with steady improvements from Chinese labs, releasing models like Z.ai's GLM-5.2 that are not only powerful but also cheaper to deploy. These developments raise questions about whether the most capable models should be broadly available at all and who has the right to control them.

Delangue sees transparency as key to making AI safer by levelling up the playing field, arguing that keeping the technology behind closed doors doesn’t eliminate risks but can create an imbalance of power. As these debates continue, it’s clear that the real race in AI is no longer about who controls the frontier but how we democratise its use.