Huawei Unveils SuperPoD Interconnect to Connect Up to 15,000 Ascend AI Chips

Huawei unveiled the SuperPoD Interconnect technology, which allows the integration of up to 15,000 graphics cards into a single cluster, including its own Ascend AI processors. The announcement took place on September 18, 2025, at the Huawei Connect conference in Shenzhen.

The technology is designed to scale computing systems and accelerate neural network training. Essentially, it is an equivalent to Nvidia NVLink, providing high-speed data exchange between AI chips.

Huawei is focusing on scalability: individual Ascend chips currently lag behind Nvidia’s solutions in performance, but combining them into large clusters will allow offering clients comparable power for working with large language models and other artificial intelligence tasks.

The presentation took place the day after Chinese regulators banned the country’s largest tech companies from purchasing Nvidia equipment, including RTX Pro 6000D servers. These measures are aimed at supporting the national semiconductor industry and reducing dependence on American technologies.

Chinese authorities stated that local manufacturers have already reached a performance level comparable to the reduced versions of Nvidia chips for the Chinese market. Against this backdrop, Huawei’s move demonstrates Beijing’s intention to strengthen technological independence in the critically important AI sector.

Analysts note that Huawei is not only compensating for the shortage after the restrictions on Nvidia supplies but also forming its own high-performance computing ecosystem, which could become the foundation for the future development of China’s AI sector.