“Chinese open-source large models have become the global AI infrastructure foundation, and many countries are avoiding export controls.”

In September 2026, Chinese large models were established as the global foundation for artificial intelligence infrastructure. Although the M3 model released by Saudi HUMAIN claims to be a “100% Saudi Model,” it is actually developed based on MiniMax M3 and trained using Saudi-language data; companies such as Rakuten in Japan, Singapore’s National Artificial Intelligence Program, Malaysia’s Communications Ministry, and South Korea’s sovereign AI project have successively adopted DeepSeek, Qwen, or Zhipu GLM as their base models. Chinese open-source models, due to their MIT or Apache 2.0 licensing, allow for commercial use and local deployment. They also offer advantages such as multilingual adaptability and low inference costs, making them a solid choice for countries to avoid US export controls on closed-source manufacturers like Anthropic and to meet their sovereign AI needs.