China has overtaken the United States in the global market for open artificial-intelligence models, marking a pivotal moment in the race to shape the future of the technology. A joint study by the Massachusetts Institute of Technology and Hugging Face reveals that Chinese-made open models accounted for 17 per cent of global downloads over the past year, surpassing the 15.8 per cent share from American giants such as Google, Meta and OpenAI. According to the Financial Times, this is the first time China has edged ahead of the US in this critical segment of AI development.
Open models — freely downloadable and customizable by developers — carry enormous strategic weight. They accelerate innovation for start-ups, democratize research, and influence the norms and values embedded in future AI systems. Beijing’s growing dominance in this space highlights a sharply diverging strategy from the US, where leading companies like OpenAI, Google and Anthropic have kept their most advanced models tightly controlled through subscription-based or enterprise offerings.
Cut off from Nvidia’s high-end chips due to US export restrictions, China has instead embraced aggressive openness. Beijing officials have encouraged domestic labs to release broad-access models, betting that community-driven iteration can compensate for hardware limitations. Analysts, including those cited by the Financial Times, say this policy has cultivated a thriving ecosystem of developers who are rapidly experimenting with new architectures and training techniques.
Much of China’s surge is driven by DeepSeek and Alibaba’s Qwen models, which together represent the bulk of the country’s downloads. DeepSeek stunned Silicon Valley when it introduced its R1 reasoning model — comparable in performance to leading US systems but built with far less compute and cost. The breakthrough raised uncomfortable questions for American labs investing billions in hyperscale data centers and frontier-model development.
In Washington, the Trump administration has urged US groups to commit more openly to “American values” in AI, making open-source development a key plank of its AI Action Plan. OpenAI responded by releasing its first “open-weight” models in August, though these fall short of fully open-source releases that include training data and code. Meta, once the poster child of open-weight AI with its Llama series, has meanwhile shifted toward closed, high-performance systems as it races toward so-called superintelligence.
MIT researcher Shayne Longpre said Chinese labs are redefining the release cycle by shipping new models weekly or biweekly, offering developers a menu of rapid-fire variations — in stark contrast to the slower cadence of US counterparts. Other experts note that China’s abundance of skilled researchers has allowed its labs to innovate under constraints, using distillation and other efficiency techniques to produce smaller but powerful models and investing heavily in AI video generation.
The rise of Chinese open models is already shaping the information landscape. Studies show these systems frequently reflect Chinese Communist Party narratives and avoid politically sensitive topics such as Taiwan or the Tiananmen Square crackdown. While US labs focus on pushing the boundaries of frontier AI and the pursuit of artificial general intelligence, the US has relatively few large independent open-source institutions — one notable exception being the Allen Institute for AI, which launched its fully open Olmo 3 model in November.
Janet Egan of the Center for a New American Security warned that China’s momentum should alarm US policymakers. “It should be of concern to the US that China is making great strides in the open … model domain,” she said — a sentiment echoed across industry and underlined in recent Financial Times coverage of the shifting AI race.
