3 min read
China has overtaken the United States in the global market for open artificial intelligence models, marking a major shift in technological influence that could reshape how AI is built and used around the world.
According to a new study by the Massachusetts Institute of Technology and open source AI platform Hugging Face, Chinese-developed open models captured 17 percent of all global downloads over the past year, surpassing the 15.8 percent share attributed to American developers for the first time.
The finding underscores a sharp divergence in strategy between the world’s two biggest AI powers. While the United States has focused on closed, proprietary models controlled tightly by companies such as OpenAI, Google and Anthropic, Chinese tech groups have accelerated their release of open models that developers can freely download, modify and integrate. The widespread adoption of these open systems could give China disproportionate influence over the next phase of AI development.
Chinese companies have leaned into open source partly out of necessity. With export controls cutting off access to advanced Nvidia chips, Beijing has encouraged companies to push development into smaller, more efficient models that gain reach by being open. Wendy Chang, senior analyst at the Mercator Institute for China Studies, noted that open source culture has become “a mainstream trend” in China, in stark contrast to the US where companies remain reluctant to reveal their advanced model architectures.
Chinese models such as DeepSeek and Alibaba’s Qwen dominate the country’s contribution to the open ecosystem. DeepSeek stunned the industry when it released its high-performance reasoning model R1, which rivaled leading US models despite being trained at a fraction of the cost and computing power. The achievement raised uncomfortable questions within Silicon Valley about whether vast spending on data centers and cutting edge hardware guarantees leadership.
MIT researcher Shayne Longpre described China’s pace of model releases as “paradigm shifting.” Instead of unveiling major versions every six months or year, Chinese labs ship weekly or biweekly iterations with multiple variants optimised for different tasks. This rapid-cycle development allows them to respond quickly to user needs and experiment more aggressively.
Despite constraints on computing hardware, China benefits from a deep pool of AI researchers who have become adept at techniques such as model distillation, enabling them to create small yet powerful models. Chinese labs have also invested heavily in AI video generation, an area expected to become central to next-generation computing.
The rise of Chinese open models has implications beyond market share. Researchers have found that these models often reflect ideological positions aligned with the Chinese Communist Party. They tend to avoid generating content on politically sensitive topics including Taiwan and the Tiananmen Square massacre. As adoption grows globally, these biases could shape the information people receive.
Meanwhile, US labs remain focused on building frontier models and pursuing artificial general intelligence. Meta, once a leader in open-weight models through its Llama series, has recently shifted emphasis toward more closed, high-performance AI systems. Only a handful of US institutions, such as the Allen Institute for AI with its Olmo 3 model, continue to champion fully open development.
Janet Egan, senior fellow at the Center for a New American Security, warned that China’s momentum should be a wake-up call.
“It should be of concern to the US that China is making great strides in the open model domain,” she said.
