Huawei plans to show its next AI accelerator, the Ascend 960, this week in Shanghai, with rotating chairman Wang Tao set to do the honors. The 960 is scheduled to hit the market in 2027 and follows today's Ascend 950 line. An even newer version, the Ascend 970, is slated for 2028, part of a three-year push the company has previously outlined to catch up to Nvidia. Huawei has said each generation will deliver roughly double the computing performance of the last.
Internally, Huawei leadership has been blunt about the ambition. Chairman of the supervisory board Guo Ping told new hires that the aim is for Ascend to run every AI model in China and abroad. As he put it, "We are closing the gap by innovating chip architectures," adding that Apple's early processor efforts were not dominant but won out through architecture redesigns and software optimization.
U.S. export controls have limited access to cutting-edge components. SK Hynix, Samsung Electronics, and Micron have halted shipments of their latest AI memory to Huawei and other Chinese chipmakers under those rules. Nvidia's highest-end Blackwell series is not offered directly into China, and while some top-tier Nvidia parts still slip into the country through smuggling, the volumes are small.
So Huawei is leaning into system-level engineering. It recently unveiled a technique called LogicFolding to squeeze more performance from chips built without the industry's most advanced equipment. It also introduced a SuperPod design that will soon let up to 100,000 Ascend processors operate as a single cluster, tied together by its UnifiedBus networking to move data more quickly. The company argues that these integrations can offset a shortfall in single-chip performance compared with Nvidia's best.
Morgan Stanley's Charlie Chan put it this way in a research note: "System-level competitiveness matters more than ever." He also wrote, "The effective gap is narrowing through multi-die design, advanced packaging, rack-scale system architecture, optical networking, and software-hardware co-optimization."
Technology shifts remind investors to prioritize steady plans that protect and grow savings. Join Briefs Finance CEO Jaspreet Singh on September 29th for a FREE live investor workshop, How to Profit From A Dollar That's Losing its Value, where he shows how we're spotting investment opportunities as the dollar falls. Save your spot.
Despite Huawei's advances, leading Chinese AI labs still primarily train on Nvidia hardware, then run their finished models on more locally made chips. A flagship rollout features DeepSeek, which intends to put no fewer than 160,000 of Huawei's Ascend 950DT units into a newly built data center located in Inner Mongolia, where they will be used to run its models. The 950DT variant favored by DeepSeek features faster, higher-capacity memory tuned for heavy inference workloads.
Production is the choke point. Huawei has been unable to meet the surge in compute demand, and it recently notified customers that pricing on the Ascend 950DT was going up 60 percent due to tight supply of key parts. The broader market is large and growing: Morgan Stanley pegs China's compute spend at as much as 646 billion yuan, or $96 billion, by 2030.
Separately, Huawei says the Ascend 950 series is the first time its AI accelerators are paired with memory chips designed in-house, though it has shared few manufacturing details. Beyond China, Huawei is exploring opportunities in countries such as Malaysia and Egypt as it leads Beijing's push for self-sufficiency in critical tech and promotes a Chinese stack abroad.
The near-term question is whether Ascend 960 brings enough muscle to train cutting-edge models and whether Huawei's cluster-and-software strategy keeps chipping away at Nvidia's advantage. Watch three signals: on-time 2027 availability, large-scale rollouts like DeepSeek's build in Inner Mongolia, and how quickly Huawei can expand manufacturing.
If Huawei scales, it pressures pricing and supplier dynamics across AI infrastructure in China and in markets open to its stack. That does not just matter to chip giants. It filters down to who wins cloud contracts, whose tools developers choose, and how quickly AI services get cheaper for businesses and consumers.
When industries change, keeping a diversified approach helps safeguard your financial future. Our CEO Jaspreet Singh is hosting a FREE live investor workshop, How to Profit From A Dollar That's Losing its Value, on September 29th. Sign up free to join him live.
