WSJ: Huawei’s New AI Chip Ascend 910D to Challenge Popular Nvidia H100

April 29,二级目录快排策略 2025  12:15

Huawei is preparing to test its new processor for artificial intelligence training, the Ascend 910D. The company hopes this chip will partially replace high-performance solutions from U.S. tech giant Nvidia in the domestic Chinese market, according to The Wall Street Journal (WSJ).

According to sources, Huawei is already in talks with Chinese tech companies about trialing the new AI processor. The first chip samples are expected to be available for testing by the end of May. As the development is still at an early stage, a series of tests will be needed for a full assessment of its performance. Nevertheless, Huawei expects that the Ascend 910D could outperform Nvidia’s widely used H100 chip in AI training tasks.

This development comes amid tightening U.S. export restrictions on the shipment of advanced Nvidia chips like the H20 to China. These restrictions have created new opportunities for Huawei and other domestic manufacturers, who continue to strengthen China's position in the semiconductor sector despite sanctions. Huawei has become one of the leaders in creating domestic alternatives to Nvidia’s chips and has previously demonstrated its ability to bypass U.S. sanctions by launching its flagship Mate 60 smartphone powered by a Chinese-made processor. The release of this device in 2025, which coincided with a visit to China by then-U.S. Commerce Secretary Gina Raimondo, caused significant concern in Washington.

This year, Huawei plans to deliver more than 800,000 Ascend chips from previous generations (models 910B and 910C) to major state-owned telecommunications companies and private AI developers, including ByteDance, the owner of TikTok. Following the restrictions on Nvidia’s H20 chip deliveries, some of these clients are already discussing the possibility of increasing their orders for the Ascend 910C.

Although Chinese chips may individually lag several years behind their Western counterparts, Huawei and other domestic manufacturers are actively exploring ways to boost overall system performance, such as by linking together multiple chips. For example, the CloudMatrix 384 computing system, unveiled in April, has been cited by analysts as potentially outperforming Nvidia’s flagship solutions in certain tasks, despite potentially higher power consumption.

Experts note that even if a single Huawei GPU falls short of Nvidia’s latest models, the sheer number of chips used in a system could compensate for the performance gap. Analysts also acknowledge challenges with energy efficiency but believe that for China’s current priorities, these issues are not critical.