중국 칩 제조업체 Cambricon, 베이징의 데이터센터 국내 AI 칩 의무화에 따라 5억 6천만 달러 조달 중

China's AI industry is undergoing rapid transformation as the government intensifies efforts to reduce reliance on foreign AI chips amid escalating trade tensions with the United States. The geopolitical pressures, particularly between Washington and Beijing over AI technologies, have prompted both nations to strengthen control over their respective AI capabilities. Beijing has reportedly advised local tech giants against purchasing foreign chips due to security concerns, while simultaneously pushing domestic alternatives. The Chinese government is now imposing requirements on data center buildouts to use homegrown solutions, directly benefiting companies like Huawei and Cambricon.

The government is demanding that data centers be equipped with more than fifty percent of AI chips coming from domestic AI companies as part of its strategy to reduce reliance on NVIDIA. This shift has been accelerated by the US government's plans to impose security backdoors into chips flowing into China, making the transition toward chips like Huawei's Ascend more appealing. However, a significant challenge remains: domestic AI chips cannot deliver the performance required to train top-tier AI models. This limitation has reportedly delayed DeepSeek's next R2 model, highlighting the gap between current domestic capabilities and industry demands.

Capitalizing on growing demand for Chinese AI chips, Cambricon is raising around 4 billion yuan to fund its expansion efforts and push its AI chips as replacements for AMD and NVIDIA solutions. The company offers the Siyuan Series for data centers and cloud computing, and is currently developing advanced options for large language models to enable domestic AI firms to train next-generation AI models. Despite these ambitious efforts, Cambricon has yet to achieve a solid breakthrough in the market.

Huawei provides the primary alternatives for Chinese AI customers, offering its Ascend AI chip lineup that includes models such as the Ascend 910B and 910C. The Ascend 910C is rumored to beat NVIDIA's H100 chip in training performance. Additionally, Huawei has developed a rack-scale solution called the CloudMatrix 384, which is said to rival NVIDIA's Blackwell NVL72 system. The transition to domestic AI solutions remains difficult for Chinese firms, however, particularly because no software equivalent to NVIDIA's CUDA currently exists.

In the near term, China must continue relying on American technology while its domestic semiconductor industry develops viable alternatives to NVIDIA's dominance in AI infrastructure.