Huawei unveils AI chip Supercluster, calls it world's most powerful computing power

In a bold move signaling its intent to lead the global artificial intelligence race, Chinese technology giant Huawei has announced the launch of a new AI chip supercluster, which it claims will be the world's most powerful computing infrastructure for years to come.

The announcement, made at the company's annual HUAWEI CONNECT conference, underscores a new strategy focused on building massive, interconnected systems to overcome challenges posed by international sanctions and secure a dominant position in the AI hardware market.

A focus on architecture over individual chip performance

During his keynote, Eric Xu, Huawei's Deputy Chairman of the Board and Rotating Chairman, unveiled the company's "SuperPoD" and "SuperCluster" architecture, built on its self-developed Ascend AI chips. A SuperPoD is described as a single, logical machine comprised of thousands of physical machines that can learn and reason as one. SuperClusters, in turn, are large-scale computing clusters made up of multiple SuperPoDs.

The new offerings include the Atlas 950 SuperPoD, which supports 8,192 Ascend NPUs (Neural Processing Units), and the even more powerful Atlas 960 SuperPoD, with 15,488 Ascend NPUs. Huawei asserts that these systems lead the industry in total computing power, memory capacity, and interconnect bandwidth.

To create the superclusters, Huawei announced the Atlas 950 SuperCluster and Atlas 960 SuperCluster, which will scale to over 500,000 and one million Ascend NPUs, respectively. The company claims these are poised to outperform all other computing clusters currently on the market, including those from rivals like Nvidia and xAI.

Rising above U.S. sanctions

Huawei's strategy is a direct response to U.S. sanctions, which have restricted its access to the most advanced semiconductor manufacturing processes. While the performance of a single Huawei Ascend chip may lag behind top-tier competitors like Nvidia's, the company has focused on developing a groundbreaking interconnect technology it calls "UnifiedBus." This protocol allows for the seamless, high-speed connection of massive numbers of chips, effectively creating a single, colossal computing unit.

"Because of U.S. sanctions, we cannot manufacture at TSMC, so the computing power of a single Huawei chip still lags behind Nvidia's," Xu reportedly said. "However, with more than 30 years of experience in connecting people and machines, we've made breakthroughs in interconnect technology that allow us to build 10,000-card-scale super nodes—making ours the most powerful computing systems in the world."

The company also announced a three-year roadmap for its Ascend chips, with new generations slated for release in 2026, 2027, and 2028. Each new series, according to Xu, is designed to double the computing power of its predecessor, ensuring a continuous stream of innovation and competitiveness.

Looking forward

Analysts note that while Huawei's claims are ambitious, its long-standing expertise in networking and systems integration gives it a strong foundation to build upon. The release of its new AI infrastructure and chip roadmap is a clear signal that the company is not only focused on survival but is actively aiming for a leadership role in the global AI landscape.