BUSINESS

Huawei Bets on Chip Clusters to Rival Nvidia

Published: Sep. 22, 2025 6:46 p.m. GMT+8

00:00

00:00/00:00

Listen to this article 1x

Eric Xu, rotating chairman of Huawei, speaks at Huawei Connect 2025 in Shanghai on Sept. 18. Photo: Huawei

Facing U.S. sanctions that have blocked China’s access to advanced semiconductors, Huawei Technologies Co. Ltd. is increasingly relying on clusters of in-house designed, lower-performance chips to achieve similar computing capabilities, the company’s rotating chairman said.

Eric Xu made the remarks during a group interview on Thursday, as Huawei unveiled a three-year roadmap to launch a series of new Ascend chips: the 950 PR in the first quarter of next year, the 950 DT in late 2026, the 960 in late 2027 and the 970 in late 2028.

You've accessed an article available only to subscribers

Subscribe today for just $.99.

VIEW OPTIONS

Unlock exclusive discounts with a Caixin group subscription — ideal for teams and organizations.

Subscribe to both Caixin Global and The Wall Street Journal — for the price of one.

Share this article

Open WeChat and scan the QR code

DIGEST HUB

Digest Hub

Back

Explore the story in 30 seconds

Huawei is using clusters of lower-performance, in-house Ascend chips to overcome US semiconductor sanctions and plans new chip releases through 2028.
Its upcoming Atlas 950 superpod will include over 8,000 Ascend 950 DT chips, projected to deliver 6.7 times Nvidia’s NVL144 system’s power.
Huawei will open source its CANN software platform by year-end, aiming to offer an alternative to Nvidia’s CUDA, enhancing developer adoption.

AI generated, for reference only

Who’s Who

Huawei Technologies Co. Ltd.: Huawei Technologies Co. Ltd. is overcoming US sanctions by using clusters of in-house designed, lower-performance chips to achieve high computing capabilities. They've unveiled a three-year roadmap for new Ascend chips, aiming to launch superpods that integrate many chips. This strategy allows them to achieve computing power comparable to, or even exceeding, current market leaders, despite being a generation behind in individual chip technology. Huawei also plans to open-source its AI software platform, CANN, to foster an ecosystem around its Ascend chips.

Nvidia Corp.: Nvidia Corp. is a leading company in advanced semiconductors, particularly graphics processing units (GPUs). Huawei is developing its own chips and systems, such as the Atlas 950 superpod, which it claims will offer 6.7 times the computing power of Nvidia's NVL144 system. While Huawei's individual chips may be less powerful than Nvidia's, Huawei utilizes a "supercluster" approach to achieve comparable or even superior overall computing capabilities, as noted by SemiAnalysis. Nvidia also has a prominent software platform, CUDA, which Huawei aims to compete with using its open-source CANN platform.

Taiwan Semiconductor Manufacturing Co. Ltd.: Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC) was prohibited from manufacturing chips for Huawei in 2020 due to U.S. sanctions. This restriction impacted Huawei's access to advanced manufacturing technology, as Chinese mainland contract chipmakers lagged behind TSMC.

ASML Holding NV: ASML Holding NV is a Dutch company that produces state-of-the-art extreme ultraviolet lithography equipment. Due to U.S. sanctions, Chinese mainland contract chipmakers cannot purchase this equipment, hindering their ability to compete with leading manufacturers like TSMC. This limitation has prompted Huawei to develop alternative chip strategies.

AMD: According to a report by SemiAnalysis, Huawei's scale-up approach in AI computing is arguably ahead of current AMD products. This suggests a comparison between Huawei's strategies and those of AMD in the realm of high-performance computing, particularly concerning AI.

Apple: The article mentions Apple (苹果公司) in the context of consumer loyalty. Huawei's rotating chairman, Eric Xu, compared the establishment of software ecosystems to consumer loyalty to Apple's iOS platform, suggesting that if Nvidia chips become unavailable, engineers would eventually migrate to Huawei's platform.

AI generated, for reference only

What Happened When