Key Points:
- World’s most powerful AI system with 500K Ascend chips.
- Claims 6.7× more power than Nvidia’s NVL144 system.
- Huawei plans chip upgrades through 2028 to dominate AI.
Chinese technology company Huawei announces its latest move in artificial intelligence computing, unveiling the Atlas 950 SuperCluster, which will operate on its in-house Ascend series chips. The system is expected to launch as early as 2026, with the company promising substantial increases in computing capacity across its new platforms.
Nvidia, one of the world’s leading AI chipmakers, acknowledged the development, stating that “the competition has undeniably arrived and is gaining momentum.” According to Nvidia, customers will ultimately decide based on the efficiency of available technology stacks for both commercial applications and open-source models.
Huawei Expands AI Infrastructure with Ascend Chips
Huawei announces a roadmap that includes three new versions of its Ascend chips, rolling out between now and 2028. The company’s goal is to double computing performance each year, using system design to overcome hardware limitations.
At the core of this infrastructure is the Atlas 950 supernode, supporting 8,192 Ascend chips, while the full Atlas 950 SuperCluster will incorporate more than 500,000 chips. A more advanced version, the Atlas 960 supernode, is scheduled for 2027 and will accommodate 15,488 chips per node, pushing the total count for the supercluster beyond 1 million chips.
Huawei announces that the Atlas 950 supernode would deliver 6.7 times more computing power than Nvidia’s NVL144 system, also expected in 2026. Looking further ahead, Huawei said the 950 SuperCluster could even outperform Elon Musk’s xAI Colossus supercomputer, boasting 1.3 times greater computing power.
Research findings appear to support some of Huawei’s claims. A study by SemiAnalysis earlier this year noted that Huawei’s CloudMatrix system outperformed Nvidia’s platforms by using a greater number of chips, despite each Ascend processor being less powerful individually.
Industry Implications and Growing Competition
Huawei announces its latest developments at Huawei Connect 2025, the company’s annual technology event in Shanghai. At the same event in 2023, Huawei introduced the Atlas 900 SuperCluster, which has since been deployed in industries such as telecommunications and manufacturing. According to the company, more than 300 Atlas 900 A3 supernodes have already been delivered to over 20 customers worldwide.
Industry experts suggest Huawei’s push represents a significant step forward in the AI computing market. George Chen, partner at The Asia Group, commented that while Huawei may present ambitious targets, its determination to become a global AI leader is “not to be underestimated.”
Meanwhile, Nvidia continues to face challenges in key markets, with increasing pressure from rivals developing large-scale AI infrastructure. Despite this, Nvidia remains a dominant force in the industry, with its GPUs widely regarded as the standard for AI model training and deployment.
As competition intensifies, both companies are likely to accelerate innovation. Huawei announces its focus on building scalable superclusters with massive chip integration highlights a trend toward expanding computing capacity by volume. Nvidia, on the other hand, continues to refine its GPU performance and ecosystem advantages, maintaining strong partnerships across cloud, enterprise, and AI research sectors.
“Computing power has and will continue to be the key for AI,” said Huawei’s rotating chairman, Eric Xu, during his presentation at the event. “We are confident that our systems will play a vital role in shaping the next era of artificial intelligence.”
Sources:
