Huawei Unveils Ambitious AI Chip Roadmap, Challenging Nvidia

In a significant move breaking years of industry silence, Chinese technology giant Huawei on Thursday revealed an aggressive three-year roadmap for its Ascend AI chips, signalling its intent to directly compete with chipmaking behemoth Nvidia.

The announcement underscores Huawei's pivotal role in China's push for a domestic semiconductor manufacturing industry, aiming to lessen its dependence on a US-dominated supply chain.

Following the launch of its Ascend 910C in the first quarter of the year, Vice Chairman Eric Xu disclosed plans for two variants of its successor, the Ascend 950, to debut next year. This will be followed by the 960 version in 2027 and the 970 in 2028.

"Computing power has always been, and will continue to be, key to artificial intelligence, and even more so to China's AI," Xu stated at the annual Huawei Connect conference in Shanghai.

Crucially, Xu revealed that the Ascend 950 chip will be powered by Huawei's own proprietary high-bandwidth memory. This marks a breakthrough, overcoming a key technological bottleneck China faced for years, previously limited to South Korean and US suppliers.

Huawei also unveiled plans for new computing power supernodes, the Atlas 950 and Atlas 960. Described by Xu as the world's most powerful, these supernodes are designed to support 8,192 and 15,488 Ascend chips, respectively. These new systems are successors to the Atlas 900, also known as the CloudMatrix 384, which utilises 384 of Huawei's current 910C chips.

According to research group SemiAnalysis, Huawei's new product reportedly outperforms Nvidia's GB200 NVL72, which incorporates 72 B200 chips, on certain metrics. Huawei attributes this superior performance to its "supernode" architecture, enabling ultra-high-speed chip interconnections.

Read more: Not India, This Is Where Apple Could Make Its Anticipated Foldable iPhone