Chinese tech giant unveils Ascend-based systems linking over one million chips to counter U.S. rivals

Huawei has launched its most ambitious artificial intelligence (AI) computing solution to date, unveiling new supernode and supercluster technologies designed to rival Nvidia’s dominance in the chip sector.

Huawei’s New AI Architecture

On September 18, Huawei introduced its Atlas 950 SuperCluster, capable of connecting more than 500,000 Ascend AI chips, with future versions scaling to over one million units. The system builds from “supernodes,” each supporting up to 15,488 chips by 2027, bundled into “superpods” and then into larger clusters. The design compensates for lower individual chip power by leveraging scale.

Rivaling Nvidia’s NVLink

Huawei’s approach directly targets Nvidia’s NVLink technology, which enables high-speed communication between GPUs. Eric Xu, Huawei’s rotating chairman, claimed the Atlas 950 supernode will deliver 6.7 times more computing power than Nvidia’s NVL144 system expected in 2026, and 1.3 times the performance of Elon Musk’s planned xAI Colossus.

Strategic Timing Amid Sanctions

The launch comes as Beijing pushes self-reliance in semiconductors. China has already restricted Nvidia’s RTX Pro 6000D for AI use, signaling support for domestic alternatives. “The competition has undeniably arrived,” Nvidia acknowledged in a statement, highlighting growing pressure from Chinese competitors.

Roadmap of Ascend Chips

Huawei outlined a rollout of new Ascend processors over the next three years:

• Ascend 950PR (2026)
• Ascend 950DT (late 2026)
• Ascend 960 (2027)
• Ascend 970 (2028)

Each generation aims to double computing capacity, expanding Huawei’s AI ecosystem across telecoms, manufacturing, and digital infrastructure.

Market and Industry Response

Research firm SemiAnalysis noted earlier this year that while each Ascend chip delivers only about one-third the performance of an Nvidia processor, Huawei’s CloudMatrix system outperformed Nvidia’s by bundling five times as many chips. Analysts caution, however, that Huawei may be overstating its technical lead.

Rising Pressure on Nvidia

Nvidia faces increasing regulatory and market headwinds in China, with local firms ordered to stop testing and ordering certain Nvidia chips. CEO Jensen Huang has described Huawei as a “formidable” competitor, while Chinese regulators continue a probe into Nvidia’s practices. The rivalry underscores the broader geopolitical and technological contest shaping the future of AI computing.

Huawei’s supercluster strategy reflects a shift in global AI competition—trading single-chip supremacy for massive, scalable clusters. For China, it signals progress toward semiconductor independence. For Nvidia and global markets, it represents both a challenge and a warning: the balance of power in AI computing may no longer rest on U.S. soil alone.

Sources: The Business Times (2025) , CNBC (2025)

Keywords: Huawei AI Chips, Nvidia Rivalry, SuperCluster Technology, Ascend Chips, China Semiconductors, Artificial Intelligence