China’s progress with domestic AI chips

On March 31, Chinese media reported that Shenzhen city began operating a smart computer cluster containing 10,000 Ascend 910C chips produced by Huawei. This cluster has a computing capacity of 11,000 petaflops (million billion calculations per second).

Combined with the 3,000 petaflop cluster launched last year, the facility now has a total computing capacity of 14,000 petaflops. Nearly 50 organizations have signed framework agreements to utilize the cluster’s computing capacity, raising the combined booking rate to 92%.

Zhang Luncheng, Vice President of startup company X Square Robot, assessed that upgrades in the scale and quality of Shenzhen’s computing capacity have made the city a leading national hub. More importantly, it shows an improvement in self-control.

According to DeepSeek research published earlier this year, the Ascend 910C has about 60% the capacity of Nvidia’s H100. The chip is designed by Huawei and manufactured by SMIC.

Huawei launched a new version of AI chip called Ascend 950PR last year. According to ReutersChinese technology giants such as ByteDance and Alibaba have plans to order new chips. Huawei is preparing to launch about 750,000 950PR models and 600,000 Ascend 910C models this year.

Shenzhen also aims to increase its real-time AI computing capacity to 80,000 petaflops this year. The China Academy of Information and Communications Technology estimates that by the end of June 2025, China’s total computing capacity will reach 962,000 petaflops, accounting for 21% of the world, an increase of 73% over the same period in 2024.

A report from IDC published on March 31 also showed that Chinese GPU and AI chip manufacturers account for nearly 41% of the country’s artificial intelligence server market by 2025. The above achievement was achieved in the context of Beijing becoming more and more cautious about dependence on foreign chips, pushing government agencies and companies to apply domestic alternatives.

According to data, the total number of AI acceleration chips provided by Nvidia, AMD and Chinese chip manufacturers will reach about 4 million units in 2025. Nvidia still holds the leading position in the market, shipping 2.2 million chips and holding 55% of the market share, while AMD has about 160,000 chips and holds 4% of the market share.

The rest comes from Chinese companies. Among them, Huawei leads with 812,000 AI chips, accounting for half of the total amount of domestic branded chips. Alibaba’s T-Head design unit ranked second, shipping 265,000 chips. Baidu’s Kunlunxin and Cambricon each shipped 116,000 chips, ranking third and fourth respectively. Hygon and two GPU startups MetaX and Iluvatar CoreX account for 5%, 4%, and 3% of the total AI accelerator cards, respectively.

Analysts say that the market shares of Nvidia and AMD may fluctuate in the near future, when domestic companies step up production. Before that, according to Financial Timesthe world’s leading AI chip manufacturer, is said to have stopped its plan to produce the H200 chip – which was designed to serve Chinese customers – despite receiving the “green light” from the US government. The Information said that earlier this year, China was said to ask technology companies not to order Nvidia’s H200 chip and aim to equip domestic AI chips.