Huawei Declares "Acceleration, Not Slowdown, for AI"... Challenges NVIDIA
New Chip Unveiled, Performance to Double with Each Generation
Connecting One Million Chips to Overcome U.S. Restrictions
Expanding HBM Utilization, Korea Must Strengthen Its Response
Huawei of China has declared its intention to overcome U.S. restrictions by building a massive computing system capable of connecting up to one million artificial intelligence (AI) chips. This move is effectively a declaration of war against NVIDIA, the leading U.S. company in the field. Despite recent calls to slow the pace of AI development, Huawei's announcement signals a determination to “hit the accelerator” on AI advancement. Huawei has also indicated plans to expand the use of high bandwidth memory (HBM), presenting a new challenge for the South Korean semiconductor industry to respond.
On the 17th, visitors toured the 'Huawei Connect 2026' exhibition held in Shanghai, China. On this day, Huawei announced its AI chip strategy to compete with U.S. company Nvidia. Photo by AFP Yonhap News
View original imageOn September 17 in Shanghai, at 'Huawei Connect 2026', Huawei unveiled its next-generation AI chip, the Ascend 960, along with a system that links these chips on a large scale. The company's strategy is to compensate for individual chip performance limitations by bundling more chips and memory together to boost overall system performance. Huawei announced it would release the Ascend 960 sooner than originally planned and will continue to launch new models every year: Ascend 970 in 2028 and Ascend 980 in 2029. For each new generation of chips, Huawei aims to double performance.
Huawei’s ambitions go beyond semiconductor development and signal the broader direction of AI advancement in China. Eric Xu, Rotating Chairman of Huawei, commented that while there are debates about slowing the pace of AI in the U.S., China, on the contrary, is at a stage where it must press forward even further with AI development.
Huawei’s aggressive approach is not just a message to the U.S.; it also presents a challenge to South Korea. The company emphasized that Korea must not be complacent with its current HBM market share, but instead must strive to continue widening the gap in terms of next-generation HBM bandwidth, capacity, power efficiency, and custom design.
The Atlas 960E SuperPoD system unveiled by Huawei can interconnect up to 4,096 chips to function as a single system. It also houses up to 1PB (petabyte) of HBM. By clustering multiple such systems, Huawei envisions operating up to one million processors as a single, massive AI supercomputer.
David Wang, Vice Chairman and Rotating Chairman of the Board at Huawei, is delivering a keynote speech at "Huawei Connect 2026" on the 17th. Photo by Huawei
View original imageDue to U.S. restrictions, advanced HBM products from Samsung Electronics and SK hynix cannot be freely supplied to Huawei. During this period, Huawei has been developing its own HBM technology.
The driving force behind this intensified strategy is the U.S. export controls. As it is difficult for Huawei to acquire the latest NVIDIA GPUs such as "Rubin" and advanced semiconductor manufacturing equipment from ASML, simply improving the performance of individual chips is not enough to compete with the U.S. Instead, Huawei seeks to link a greater number of relatively lower-performing chips more quickly, while advancing both HBM and communications technologies, in order to boost overall system performance. Leading foreign media outlets have also analyzed that Huawei is seeking to overcome the limitations of advanced manufacturing processes by leveraging massive chip interconnectivity and networking technologies.
In this process, the importance of memory has grown significantly. The Ascend 960DT is equipped with 288GB of HBM. The memory bandwidth, which refers to the rate at which data flows, reaches 9.6TB per second. With U.S. restrictions making it difficult for Huawei to freely source advanced South Korean HBM, the company is moving to secure not only AI chip technology but also memory capabilities to build an independent AI semiconductor supply chain within China.
This does not mean that Huawei will directly manufacture DRAM. However, the company is making clear moves to design and optimize HBM specifications to meet the needs of its AI chips. Chinese DRAM companies such as Changxin Memory Technologies (CXMT) are also accelerating their efforts to develop HBM.
This trend suggests not only that Chinese HBM makers are catching up, but also that there could be a significant challenge to the competitiveness of the South Korean semiconductor industry, which is led by Samsung Electronics and SK hynix. If Huawei develops AI chips, HBM, and chip interconnect technologies together, new memory architectures and utilization methods could take root in China, potentially eroding Korea's current leadership in HBM technology.
Kim Joungho, Professor at the School of Electrical Engineering at KAIST, warned, "We must be vigilant about China's investment in HBM," and argued that South Korea needs to accelerate its own HBM development efforts.
Hot Picks Today
"Two Spend 500,000 Won in a Flash, Yet Wait Two Hours?" ... 'Dreamland' Shunned by Kids and Foreign Tourists
- The Era of Getting 200 Million Won After 30 Years Is Over... A Japanese Bank's "Perpetual Department Head" Who Just Clocked In Cries Out
- Spain Hit by Its Worst Tragedy... Lee Kangin, Moreno and Real Madrid Join Soccer World in Mourning
- They Said, “Make It Your Home for Life” - Then Suddenly Kicked 200 Passengers Off... What Happened to the Around-the-World Cruise?
- "130 Rice Bowls Sold Out in a Day"... 'Unfiltered Face' Adds 8 Million Won to Night Market Sales: China's Bungeunwoo
The Korean government is also focused on speeding up, rather than slowing down, AI development. Baek Kyunghoon, Deputy Prime Minister and Minister of Science and ICT, stated on his social networking account, "Korea has no time to delay."
© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.