Huawei has unveiled its next-generation artificial intelligence computing hardware, positioning the new Ascend 960 SuperPoD cluster as its most direct challenge yet to Nvidia's dominance of the global AI market. Chairman Wang Tao announced the system at Huawei's annual summit in Shanghai on Thursday, according to the company's official announcement, and said it is scheduled to reach the market in 2027 despite US restrictions.

The company said the Ascend 960DT chip will be available in the first quarter of 2027, while the Ascend 960PR will arrive in the third quarter. Tao also introduced an upgraded iteration of Huawei's UnifiedBus interconnect technology, and the company has added near-packaged optics (NPOs) to its SuperPoD architecture to address the dual challenges of speed and energy consumption. By integrating the UnifiedBus interconnect with its Hi-ONE optical systems, Huawei said it can now tie 4,000 processors together in a single SuperPoD, creating deep, high-speed optical paths throughout the network. The launch was reported by the South Morning Post.

Huawei's goal: Ascend chips that power every AI model

Guo Ping, chairman of Huawei's supervisory board, said earlier this month that the ultimate goal is for the new Ascend chips to power every AI model globally. He argued that Huawei can narrow the technology disparity by rethinking chip architecture, much as Apple overcame its early limitations in chipmaking by focusing on custom design and software efficiency. The Shenzhen-based company, he added, is aiming for total compatibility and will deepen its capabilities and maximize value in the areas it has prioritized.

“We expect the company to achieve even better results, providing digital infrastructure solutions to global customers, offering diversified solutions for the development of artificial intelligence, and contributing to the rise of China's electronics industry,” Guo said.

Beyond the new launches, Huawei recently secured a major commitment from DeepSeek. The AI startup is building a data center in Inner Mongolia that will run on a fleet of 160,000 Ascend 950DT processors.

The Ascend 960 is the next step up from Huawei's current Ascend 950 chips, and the company plans to release the Ascend 970 in 2028 — keeping pace with a three-year pledge to challenge Nvidia by doubling processor performance with every iteration.

A system-level strategy to offset chip shortcomings

Huawei has previously acknowledged that its hardware falls short of Nvidia's in terms of pure speed and power. To make up the difference, it has said it will prioritize leveraging sheer hardware volume, superior networking, and state policy support to win over customers, emphasizing that system-level adaptations and networking can offset the limitations of individual chip performance.

Bernstein analysts led by Qingyuan Lin responded favorably to the Shenzhen-based chipmaker's approach, writing: “Huawei's willingness to publicly articulate its AI roadmap represents a strong signal of confidence in the resilience of its future local foundry supply.”

Morgan Stanley analyst Charlie Chan also stated recently: “System-level competitiveness matters more than ever. The effective gap is narrowing through multi-die design, advanced packaging, rack-scale system architecture, optical networking, and software-hardware co-optimization.”

A demanding AI market backdrop

Huawei's latest roadmap comes as demand for AI computing continues to grow, intensifying pressure on Chinese technology companies to access advanced AI processors. The company is working to build an alternative AI hardware ecosystem that is less dependent on foreign vendors and more locally rooted, making it easier for Chinese developers to access domestic computing capabilities.

The Ascend 960 will matter not only for its processor performance but also for how tightly Huawei weaves together networking, optical technology, and software. By linking thousands of processors, the company aims to build large computing systems capable of handling demanding AI workloads — an effort that grows more important as US export restrictions limit Chinese companies' access to Nvidia's most advanced AI processors.

Huawei's strategy is to improve system-level performance rather than individual chip speed alone. Its progress, however, will depend on its ability to secure enough advanced manufacturing capacity and to persuade AI developers to adopt its hardware and software ecosystem.

The performance gap remains substantial. According to Bernstein analysts, Huawei's next-generation Ascend 950 delivers only a fraction — around 6% — of the performance of Nvidia's upcoming VR200 superchip. Huawei, for its part, says it is exploring ways to cluster up to one million chips to theoretically close that gap, and notes that its in-house high-bandwidth memory architecture can improve data transfer between a processor's various components.

Meanwhile, Chinese regulators have for some time limited local tech giants' ability to purchase Nvidia's chips, including the H200 series. More recently, however, Chinese authorities have allowed a few batches of Nvidia's H200 chips into the mainland to ensure that local AI developers can continue training their models. According to sources, ByteDance and Tencent recently received the green light to import about 10,000 H200 processors. Even without official authorization, Nvidia's most advanced processors continue to reach the Chinese market, though in limited quantities.