TMTPOST -- Chinese telecommunications giant Huawei unveiled a sweeping expansion of its artificial intelligence computing ambitions on Thursday, announcing new supercomputing systems powered by its in-house Ascend chips. The move signals escalating competition with U.S. chipmaker Nvidia, the global leader in AI processors.
AI-generated image
At its annual Huawei Connect conference in Shanghai, the company introduced the forthcoming 「Atlas 950 SuperCluster,」 which it said will begin rolling out as early as next year. The announcement underscores Huawei』s determination to challenge U.S. dominance in advanced AI hardware despite Washington』s tightening export controls on semiconductors.
「The competition has undeniably arrived and is gaining momentum,」 an Nvidia spokesperson said. 「Customers will choose the best technology stack for running the world』s most popular commercial applications and open-source models.」
U.S. restrictions have cut China off from the most advanced chips needed to train large-scale AI models, forcing Chinese firms to rely on domestically developed processors or stockpiles of older foreign components. Huawei, long targeted by U.S. sanctions, has turned to large-scale clustering of its less powerful but increasingly refined Ascend processors. By combining vast numbers of these chips in interconnected networks, the company aims to match or surpass the performance of systems built on Nvidia』s hardware.
Huawei said it plans to release three new versions of its Ascend chips through 2028, pledging to double computing power with each iteration. At the base level, its new Atlas 950 supernode will connect 8,192 Ascend chips. A full Atlas 950 SuperCluster would integrate more than 500,000 chips.
A future upgrade, the Atlas 960 system slated for 2027, is projected to support nearly 15,500 chips per node and scale to more than 1 million Ascend processors across a full supercluster. Huawei claimed these would represent the world』s most powerful AI systems by raw computing power.
Huawei did not shy away from comparisons with its American rival. Eric Xu, the company』s vice chairman and rotating chairman, said the Atlas 950 supernode would deliver 6.7 times more computing power than Nvidia』s upcoming NVL144 system. He went further, predicting Huawei would 「be ahead on all fronts」 compared with Nvidia』s 2027 offering — and claimed the Atlas 950 SuperCluster would even exceed the computing capacity of Elon Musk』s xAI Colossus project.
While independent verification of Huawei』s claims is limited, industry experts said the company』s ambition is clear.
「Huawei』s announcement on its computing breakthrough is well timed with recent increasing emphasis by the Chinese government on self-reliance on China』s own chip technologies,」 said George Chen, partner and co-chair of the digital practice at The Asia Group. While noting that Huawei might be overstating its technical edge, he added, 「the ambition to be a world AI leader cannot be underestimated.」
Research from SemiAnalysis earlier this year found that Huawei』s CloudMatrix system outperformed Nvidia』s equivalent despite each Ascend processor offering only about one-third the power of an Nvidia chip. The advantage came from scale: Huawei deployed roughly five times as many chips in its clusters.
「Computing power has and will continue to be the key for AI,」 Xu emphasized during his keynote speech in Shanghai.
Huawei』s new announcements build on earlier efforts. Two years ago, it introduced the Atlas 900 SuperCluster, which it still markets with 「thousands」 of Ascend chips. On Thursday, the company said it had already deployed more than 300 Atlas 900 A3 supernodes to over 20 clients across telecoms, manufacturing, and other sectors.
This momentum reflects a broader push within China to substitute foreign technology with homegrown alternatives. Beijing has increasingly pressed domestic firms to use local suppliers and invest in national semiconductor capabilities. Huawei』s breakthroughs are likely to be touted as proof of progress in these efforts.
Huawei』s offensive comes as Nvidia faces growing headwinds in China, once one of its most important markets. Earlier this week, China extended an antitrust probe into the U.S. company over alleged monopolistic practices. On Wednesday, Nvidia』s shares fell more than 2% after the Financial Times reported that Beijing had ordered Chinese tech giants to halt testing and purchases of the RTX Pro 6000D, one of Nvidia』s new AI-focused chips.
Nvidia Chief Executive Jensen Huang expressed frustration over the developments. 「I am disappointed to hear this news,」 he told reporters, while reiterating that Huawei represents a 「formidable」 competitor.
The geopolitical context adds another layer of complexity. The U.S. and China concluded trade talks in Spain this week that reportedly included discussions over the fate of TikTok, the video app owned by Beijing-based ByteDance. But technology competition — particularly around semiconductors and AI — remains one of the most contentious issues in the bilateral relationship.
For Nvidia, the challenge is twofold: navigating U.S.-China political tensions while defending its market leadership against Huawei』s scaling strategy. While Nvidia still enjoys an edge in chip performance and ecosystem maturity, Huawei』s aggressive clustering approach could erode that lead in specific applications.
For Huawei, the path is equally fraught. Building systems at such enormous scale requires not only chip production but also innovations in networking, cooling, and software integration. Questions remain about whether the company can deliver on its bold promises under continued sanctions.
Even so, the stakes are enormous. Computing power is the currency of the AI race, and Huawei』s declaration that it will double capabilities annually through 2028 raises expectations across the industry. Nvidia may remain the benchmark, but Huawei』s announcement makes clear that the fight for AI leadership is now firmly a two-way contest.
