Key Highlights

  • Huawei moved up the launch of its next-generation Ascend 960DT AI chip to the first quarter of 2027, three quarters ahead of schedule.
  • The company’s chairman said Chinese Demand for its AI computing hardware already exceeds what it can produce, limiting international expansion.
  • Huawei’s chairman separately urged Chinese AI labs to accelerate development, contrasting with Silicon Valley’s calls for a coordinated slowdown.
  • Huawei’s Data Center product will connect as many as 4,096 Ascend processors, up from 1,024 in the previous generation.

Capacity Constraints Drive an Accelerated Timeline

Huawei is pulling forward its next-generation AI chip roadmap as Chinese demand for its computing hardware already outstrips what the company can produce, executives said at the company’s annual Huawei Connect conference. The Ascend 960DT chip, aimed at Training AI models, will now launch in the first quarter of 2027, three quarters ahead of its original schedule, while a companion inference chip will follow in the third quarter.

Rotating chairman Eric Xu told reporters that production capacity for Huawei’s AI processors falls significantly short of domestic market requirements, effectively ruling out any substantial international sales expansion for now. The new chips will deliver twice the computing performance of Huawei’s current flagship product.

Building an Alternative to Nvidia

At the heart of Huawei’s effort to challenge Nvidia is its networking technology, which the company says allows it to build larger processor clusters than its US rival without compromising bandwidth. Huawei’s upcoming data center product will connect as many as 4,096 Ascend processors, compared with 1,024 in the previous generation, allowing customers to train and run larger AI models. Xu estimated Ascend has overtaken Nvidia in Market Share within China, while acknowledging reliable estimates remain difficult to obtain.

"We can’t accept a destiny that we cannot control," Xu said, adding that Huawei and the Chinese government are pushing for full self-sufficiency across the chip Supply chain, given US export controls preventing the company from using leading-edge contract manufacturers such as Taiwan’s TSMC.

A Contrasting Call on AI Safety

Xu also urged Chinese AI companies to accelerate development to catch up with frontier US labs, contrasting with recent Silicon Valley calls for a coordinated slowdown. Chinese researchers needed to "increase the speed of development so they can also see the dangers of AI development" like their US counterparts, Xu said. His comments came days after Anthropic and OpenAI called for slower development following stark warnings over AI safety.

Software Gap Narrowing

Huawei’s chief scientist for semiconductors said changes in how large AI models are trained are eroding the advantage of Nvidia’s CUDA software platform, which has long locked developers into the US chipmaker’s processors. "CUDA is not as important as it was two years ago," he said. Xu said at least six or seven Chinese AI labs are aiming to train models containing 10 trillion to 40 trillion parameters over the next two years.

Outlook

Whether Huawei can close the capacity gap with Nvidia will likely depend on how quickly China’s domestic chip supply chain matures, particularly given continued US export restrictions on leading-edge manufacturing. The accelerated timeline underscores the urgency Chinese firms feel, though production bottlenecks in memory and advanced fabrication remain significant constraints on realizing that ambition.