Chinese open-source AI models have surpassed American open-source models in global downloads for the first time, shifting the balance of power in the global AI market. According to a joint study by MIT and Hugging Face, Chinese models account for a 17 percent share of the open-source market, while US models hold a 15.8 percent share.
China is revolutionizing AI
This remarkable shift has been accelerated by the affordability and frequent updates of Chinese models, particularly DeepSeek and Alibaba’s Qwen. This open-source approach by China is a clear departure from the strategy pursued by US tech giants.
In the US, companies like OpenAI, Google, and Anthropic are keeping their most advanced AI models private. These giants base their revenue models on subscriptions and focus on their ultimate goal of developing Artificial General Intelligence (AGI). Even Meta, previously known for its open-source Llama models, is turning to closed-source models as it moves toward its goal of “superintelligence.”
Chinese companies, however, have largely turned to these open-source versions due to US export restrictions on Nvidia’s advanced chips. These restrictions have pushed Chinese developers to adopt more innovative approaches, leading them to release versions of their models weekly or biweekly, compared to the typical six-month or annual cycles of US companies.
One of the few major US initiatives in the open-source space is Olmo 3, published by the Allen Institute for AI in November. Despite these developments, it is reported that up to 80% of AI startups submitting proposals to the venture capital firm Andreessen Horowitz now use Chinese open-source models. This rate, as reported by partner Martin Casado, clearly demonstrates the practical impact of China’s rise in this field.
This rapid increase in the use of Chinese models is also triggering security concerns in the West. Research published by cybersecurity firm CrowdStrike on November 20th showed that DeepSeek models generated significantly more insecure code when given commands related to Tibet, the Uyghurs, or other political issues deemed sensitive by Beijing.
The studies also reveal a clear ideological alignment with the Chinese Communist Party, with the models refusing to generate content on sensitive topics like Taiwan or the Tiananmen Square incident. Experts say China’s rapid advancement in open models serves as a warning to the US.
