Huawei takes on Nvidia in China’s AI chip market with new Ascend 910C processor
Huawei is rolling out its new Ascend 910C processor to compete with Nvidia in China’s AI chip market.
The company has begun sending samples to major Chinese server makers for testing, according to sources who spoke to the South China Morning Post.
The Ascend 910C builds on Huawei’s previous 910B chip. Huawei plans to offer it to large Chinese internet companies that have been key Nvidia customers.
The move comes as US export restrictions limit Nvidia’s ability to sell its most advanced GPUs in China. Analysts believe the earlier Ascend 910B chip matches or slightly exceeds the AI performance of Nvidia’s A100 GPU.
Ad
Huawei claims its Ascend chips were used to train about half of China’s 70 most powerful large language models last year. Chinese tech giant Baidu reportedly ordered 1,600 Ascend 910B chips worth $62 million, preparing for potential future limits on Nvidia purchases.
China must become less dependent on US chips
Huawei appears to be developing a strategy to bolster China’s domestic AI industry and reduce reliance on US chips. The company is trying to sell the chips bundled with other components like networking and storage solutions, according to a source.
Additionally, Huawei has formed a group of Chinese chipmakers aiming to produce high-bandwidth memory chips for AI by 2026 as an alternative to Nvidia’s offerings.
However, Nvidia’s slimmed-down H20 chip specifically for China is also selling well after a slow start, according to an SCMP source, in part because Chinese AI companies are interested in Nvidia’s technical support and maintenance services. Nvidia is expected to ship more than one million H20 GPUs in China this year, generating sales of around $12 billion.
While it’s unclear whether Huawei can seriously challenge Nvidia’s market position, with its much more powerful Blackwell chips already in the starting blocks, the competition is likely to drive AI development in China and reduce the country’s reliance on US technology.
Recommendation