Huawei Sets 2027 AI Chip Roadmap Alongside New 4,000-Processor System

Huawei has launched the Atlas 960 SuperPoD, a new AI computing cluster, and set a 2027 timetable for two additional AI chips. The presentation puts Huawei’s bet on large, tightly connected systems at the center of its effort to build advanced AI infrastructure under U.S. semiconductor restrictions.

At Huawei Connect 2026 in Shanghai, the company introduced the Atlas 960 as the successor to its Atlas 950 system. Huawei says the new cluster, powered by Ascend 960 chips, improves training performance by 2.3 times and inference performance by 2.5 times for models with 10 trillion parameters. Training is the process of building a model; inference is the work of using a trained model to generate predictions or responses.

The comparison Huawei is making is between generations of complete computing systems. The Atlas 950 launched late last year, according to the Associated Press, while the Atlas 960 arrives with an upgraded interconnect: UnifiedBus links processors, and Huawei has added near-packaged optics to the SuperPoD architecture for the first time. Huawei says the addition enables higher speeds and lower energy use.

Huawei said UnifiedBus and its Hi-ONE optical system can connect as many as 4,000 processors in one SuperPoD. That scale addresses a basic systems problem: the largest AI workloads require many chips to operate together, so the network joining them can shape the usefulness of the overall machine as much as the individual accelerator.

What Huawei scheduled next

• The 960DT AI chip is planned for the first quarter of 2027.
• The Ascend 960PR AI chip is planned for the third quarter of 2027.
• Huawei also outlined Ascend 970 and 980 chip series for 2028 and 2029, respectively.

Huawei rotating chairman David Wang said the company has shipped more than 1,000 UnifiedBus-linked systems to more than 370 customers. That gives the interconnect strategy a reported deployment base, though Huawei did not identify the customers or disclose how those systems are configured. It is evidence of distribution, not a demonstrated comparison with Nvidia’s hardware or software platform.

Huawei’s announcement comes as U.S.-led controls limit China’s access to some of the most advanced AI chips and chipmaking equipment. Chinese AI companies are increasingly adopting domestic chips, but analysts cited by the Associated Press say advanced AI model training in China still often relies on U.S. chips, including Nvidia’s. That makes the Atlas 960 claims strategically important without settling whether Huawei can replace the wider Nvidia ecosystem.

Huawei has offered a performance comparison against its own previous cluster, a processor-connection ceiling, and a roadmap extending through 2029. The unresolved comparison is external: whether those ingredients can persuade enough AI developers and customers to use Huawei’s systems for the most demanding work, rather than treating domestic hardware as a constrained alternative.