China Orders Data Centers to Use 50% Domestic Chips

China has expanded a new mandate requiring publicly owned data centers to source at least 50 percent of their chips from domestic manufacturers, marking a major step in reducing dependence on foreign technology amid U.S. export restrictions.

Initially piloted in Shanghai, the policy is now nationwide, reflecting Beijing’s strategy to secure semiconductor self-sufficiency and strengthen AI infrastructure. Over 500 new data center projects were launched in China during 2023–2024, creating strong demand for chips. While Chinese-made semiconductors are capable of running pre-trained AI models, U.S. chips like Nvidia’s H100 and H800 remain dominant for training advanced models.

Key Highlights

• China mandates 50 percent domestic chips in public data centers
• Policy aligns with $305B semiconductor goal by 2030
• Technical hurdles remain as ecosystems like CUDA and CANN clash

This mandate ensures guaranteed demand for local suppliers while aligning with the country’s long-term “Made in China 2025” initiative, which targets $305 billion in semiconductor output by 2030—up from $65 billion in 2016.

China has already invested more than $40 billion since 2014 toward its goal of achieving 70 percent chip sufficiency by 2030. Currently, domestic production excels in mature 22/28nm processes, with ambitions to capture 40 percent of this market by the end of the decade. However, the policy presents significant technical challenges: different chipmakers use incompatible ecosystems—such as Nvidia’s CUDA versus Huawei’s CANN—making model migration difficult.

Also Read: Trump May Allow Nvidia's Blackwell AI Chip Sale to China

Despite this, progress is being made. iFlytek has adopted Huawei chips for AI training, and Beijing-based SiliconFlow has collaborated with Huawei to optimize AI models that can outperform Nvidia’s H800 in select use cases. These developments signal the beginning of a more resilient domestic chip ecosystem in China.