TLDR
- Huawei unveils three-year roadmap to compete with Nvidia in AI chip market using cluster technology
- Company plans to connect up to 15,488 Ascend processors with new UnifiedBus interconnect system
- Huawei claims 62x faster data transmission speeds compared to Nvidia’s upcoming NVLink144 technology
- Individual Huawei chips deliver only 6% performance of Nvidia’s next-generation processors
- Strategy relies on “brute force” approach rather than superior single-chip performance
Chinese technology giant Huawei has announced a comprehensive three-year strategy to compete directly with Nvidia in the artificial intelligence processor market. The announcement, made at Huawei’s annual Connect conference, represents a major shift from the company’s typically secretive approach to product development.
Rotating Chairman Eric Xu presented detailed plans for what Huawei calls its “SuperPod” technology, designed to challenge Nvidia’s current market leadership through innovative clustering approaches. The timing coincides with ongoing US-China tensions over semiconductor technology and trade restrictions.
Huawei openly admits its individual processors cannot match Nvidia chips in raw performance. A single next-generation Ascend 950 processor delivers only 6% of the performance offered by Nvidia’s upcoming VR200 superchip, according to Bernstein analysts.
The company’s strategy centers on connecting massive numbers of processors together using advanced networking technology. Huawei plans to cluster up to 15,488 Ascend-branded AI chips using its newly developed UnifiedBus interconnect protocol.
Revolutionary Interconnect Technology
The UnifiedBus system represents Huawei’s key technological advantage in this competition. The company claims this interconnect can transmit data up to 62 times faster than Nvidia’s planned NVLink144 technology, enabling more efficient communication between processors.
Current Nvidia NVLink72 technology connects 72 Blackwell graphics processing units with 36 Grace central processing units. Huawei’s approach aims to scale much larger, potentially linking up to one million chips in what it terms “super clusters.”
The company also introduced proprietary high-bandwidth memory architecture to improve internal data transfer speeds. This development comes despite US sanctions that cut Huawei’s access to memory industry leaders like SK Hynix since 2020.
Huawei’s roadmap includes future processor generations: Ascend 950, 960, and 970 chips. The Ascend 970, scheduled for 2028, will feature 4 terabit-per-second interconnect speeds compared to Nvidia’s current 1.8 terabit-per-second capability.
Manufacturing Challenges Remain
Despite ambitious plans, analysts express skepticism about Huawei’s manufacturing capabilities. Jefferies analysts noted the company’s failure to deliver promised Ascend 910D processors using 5-nanometer technology due to poor production yields.
Huawei has not advanced beyond 7-nanometer chip designs since introducing them in 2023. The company did not specify manufacturing plans for its new processors, which analysts consider a potential bottleneck.
The lack of advanced chipmaking equipment remains China’s biggest obstacle in reducing dependence on foreign semiconductor technology. US restrictions continue to limit Chinese companies’ access to cutting-edge manufacturing processes.
Bernstein analysts suggest Huawei’s public announcement signals confidence in securing reliable domestic manufacturing capabilities. The company believes it has established supply chains capable of supporting its ambitious AI processor development timeline.
Nvidia shares declined 1% following Huawei’s announcement, after gaining on news of a $100 billion partnership with OpenAI. The Chinese government recently implemented measures targeting Nvidia, including directives preventing Chinese firms from purchasing certain Nvidia components.
Huawei’s public strategy represents China’s broader push for semiconductor independence and domestic technology development in response to ongoing US trade restrictions.
