Huawei’s AI chip plan designed to break US dependence

Chinese tech giant Huawei has unveiled its once-secretive AI chip development program, an announcement that could reshape expectations across the fast-moving, high-stakes global industry.

On September 18, Huawei chairman Eric Xu announced a three-year roadmap for the company’s Ascend series of processors at the Huawei Connect 2025 conference in Shanghai.

Coming soon after the Chinese government decided to block Chinese companies from using Nvidia processors, Huawei’s announcement will heighten competition between Chinese and American AI efforts while signaling a significant reduction in China’s dependence on American technology.

Explaining the roadmap, Xu told the audience that, following the launch of the Ascend 910C earlier this year, Huawei will release three new chip series – the Ascend 950, 960, and 970 – in 2026, 2027, and 2028, respectively.

“Generally, we will follow a one-year release cycle and double compute with each release,” he said. That pace of development, if achieved, is comparable to that of leading US chipmakers Nvidia and AMD.

Xu also said Huawei will offer proprietary high-bandwidth memory (HBM) solutions for variants of the Ascend 950, revealing that, along with Chinese memory maker CXMT, Huawei has been working hard to escape dependence on HBM imported from Korea – a target of US sanctions.

Xu’s keynote address, titled “Groundbreaking SuperPoD Interconnect: Leading a New Paradigm for AI Infrastructure,” outlined Huawei’s plans “to create a new computing architecture, and develop computing SuperPoDs and SuperClusters, to sustainably meet long-term demand for computing power.”

“SuperPoDs,” Xu continued, “have become the main form factor of products for large-scale AI infrastructure. They are becoming the new norm. A SuperPoD is a single logical machine, made up of multiple physical machines that can learn, think, and reason as one. As computing demand continues to grow, so will SuperPoDs.” A SuperPoD functions as a single computer, while SuperClusters connect multiple SuperPods.

In March 2025, Huawei launched its Atlas 900 A3 SuperPoD, which incorporates up to 384 Ascend 910C chips. According to Xu, it is currently the largest SuperPoD in the world.

“Maybe you’ve heard about the CloudMatrix384,” Xu said. “It’s a cloud service instance that Huawei Cloud has built on top of our Atlas 900 A3 SuperPoDs. Since its launch, we’ve deployed more than 300 Atlas 900 A3 SuperPoDs to serve over 20 customers in sectors like ISP [Internet Service Provider], telecoms, and manufacturing.”

This squares with the assessment of Dylan Patel and his colleague at SemiAnalysis, who found that Huawei’s Ascend 910C AI processor is significantly more impressive when used in the company’s CloudMatrix 394 rack-scale AI data center solution – a complete system consisting of 384 Ascend 910C processors, along with servers, networking, storage, power management and cooling.

In their estimation, the CloudMatrix 394 “competes directly” with Nvidia’s top-end GB200 Grace Blackwell Superchip.

The “engineering advantage,” they wrote, “is at the system level, not just at the chip level, with innovation at the accelerator, networking, optics, and software layers… Huawei is a generation behind in chips, but its scale-up solution is arguably a generation ahead of Nvidia and AMD’s current products on the market.”

To power its new SuperPoDs, Huawei developed a new interconnect protocol called UnifiedBus.

“Today, we are also releasing the technical specifications for UnifiedBus 2.0. You might wonder why we’re starting things out with version 2.0. Our research on UnifiedBus actually began back in 2019,” Xu said. “For reasons everyone here is familiar with, we don’t have access to advanced process nodes, so we’ve decided to focus our efforts on making breakthroughs by combining chips – essentially connecting more computing resources.”

The reasons everyone at Huawei Connect 2025 was familiar with are that US sanctions have prevented Chinese semiconductor producers from using advanced EUV lithography, and Chinese equipment makers have not yet been able to replicate or develop their own version of the technology.

Huawei is working on the assumption that China “will lag behind in semiconductor manufacturing process nodes for a relatively long time” and that “sustainable computing power can only be achieved with process nodes that are practically available.” Fortunately for Huawei and for China, fabricating the Ascend processors does not require 5nm or smaller design rules.

The Atlas 900 A3 SuperPoD uses UnifiedBus 1.0. UnifiedBus 2.0 is an open protocol that Huawei hopes will be widely adopted, creating an open UnifiedBus ecosystem.

In Xu’s assessment, “With a series of novel systems innovations, we’ve managed to develop a solid interconnect technology for SuperPoDs – one that provides the high reliability, all-optical interconnect, high bandwidth, and low latency required for large-scale SuperPoDs.”

Huawei’s technological advance did not appear out of the blue but rather is the result of years of work, largely out of the view of the media and under the US government’s radar. And that tech is aimed not only at developing successful products, but also at setting new industry standards.

The Atlas 950 SuperPoD will feature up to 8,192 Ascend chips – 20 times more than the Atlas 900 A3 SuperPoD – and is expected to be available in the fourth quarter of 2026. The Atlas 950 SuperClusters, set to launch at the same time, will consist of 64 Atlas 950 SuperPoDs. In Huawei’s estimation, the system will “even outstrip [Elon Musk’s] xAI’s Colossus, now the world’s largest computing cluster.”

According to Xu, “The Atlas 950 SuperCluster will unequivocally be the world’s most powerful computing cluster. From existing dense and sparse models with over 100 billion parameters, to future models with over 1 trillion or even 10 trillion parameters, the Atlas 950 SuperCluster will be a compute powerhouse for model training, driving efficient and steady innovation in AI.”

But the Colossus2 data center in Memphis, Tennessee, is already being scaled up, and Microsoft/OpenAI, Oracle, Meta, Alphabet (Google) and Amazon/Anthropic are also in the race. So it may be more accurate to say that Huawei will be going head-to-head with, not frantically chasing, the American industry leaders.

Follow this writer on X: @ScottFo83517667