Login to Continue Learning
Beijing recently advised local tech giants to halt purchases of NVIDIA’s H20 AI chip amid concerns about potential security flaws. In response, China is now mandating that data centers use homegrown AI solutions, benefiting companies such as Huawei and Cambricon. According to reports, the government requires that more than 50% of AI chips used in data centers be sourced from domestic manufacturers. This move aims to reduce reliance on NVIDIA, especially considering the US’s plan to insert security backdoors into chips flowing into China.
Cambricon, a Chinese competitor to NVIDIA, has announced plans to raise $560 million to accelerate its efforts in this market. The company is developing options like the Siyuan Series for data centers and cloud computing, with advanced models being designed specifically for large language models (LLMs). However, domestic AI chips currently struggle to match the performance of top-tier models, leading to delays such as that reported for DeepSeek’s next R2 model.
With growing demand for Chinese AI chips, companies like Cambricon are capitalizing on this opportunity. Huawei, another major player in the sector, offers its Ascend AI chip lineup, including the 910B and 910C models, which some speculate outperform NVIDIA’s H100. Huawei also unveiled a rack-scale solution called CloudMatrix 384, aiming to rival NVIDIA’s Blackwell NVL72 system.
Despite these advancements, Chinese firms still face challenges in fully transitioning from foreign AI solutions due to the lack of equivalent software support, particularly for frameworks like NVIDIA’s CUDA. Nonetheless, China continues its push towards self-sufficiency in AI technology, with plans to increase domestic chip usage in data centers.
