According to a report by Asia Times on September 21, Huawei announced a major development. At the "Huawei Global Connect 2025" conference, it released the latest development roadmap for its Ascend series of AI chips, announcing that it will successively launch three new generation AI chips, Ascend 950, 960, and 970, over the next three years.

Huawei stated that it will continue to advance AI chip iteration at a pace of one generation per year, with computing power doubling each time, and fully integrate the chip architecture into its SuperPoD supercomputing platform, achieving the strategic implementation of "chip as infrastructure."

The biggest highlight of this release is that Huawei not only achieved performance improvements in the chips themselves but also plans to build a complete AI computing power solution covering chips, systems, data centers, and software ecosystems through domestic high-bandwidth memory, self-developed interconnection protocols, the SuperPoD system, and CloudMatrix cloud computing instances.

This is not only a response to U.S. sanctions but also a systematic breakthrough.

Currently, China is countering some U.S.-made AI chips, while the U.S. has also restricted exports of key EDA tools and EUV equipment to China. Under this dual blockade, Huawei chose to avoid process technology and restructure its system, marking that Chinese enterprises are no longer just followers in the computing power race but are beginning to try to set their own rules.

Huawei AI Chip

Different from previous point-based breakthroughs, the biggest feature of Huawei's AI chip is its strong system-level integration capability and replaceable ecological structure.

Firstly, it is not simply launching a single chip but using the Ascend chip as the underlying engine of the entire AI computing platform. With SuperPoD as the basic unit, hundreds to thousands of chips can be interconnected through UnifiedBus interconnection technology to form super nodes.

This architecture not only bypasses the limitations of advanced process technologies but also levels the performance gap between chips through system scheduling and optical interconnection.

At the same time, Huawei has built a complete software and hardware ecosystem: including its self-developed CANN compiler, MindSpore deep learning framework, openPangu large model system, AI training scheduling system, as well as domestically produced storage, switching, cooling, and power supply hardware.

This means that an institution deploying Huawei's SuperPoD system can completely move away from the U.S. technology stack, achieving full domestic production throughout the process.

Another aspect is that by collaborating with domestic HBM manufacturers such as Changxin, Huawei is also addressing the last remaining gap.

AI Chip Rendering

The reason why Huawei can truly make this happen is a very critical factor: the pressure from the U.S. was too great.

Since 2018, Huawei has become the most thoroughly suppressed company in the global tech industry - chips were cut off, ARM was restricted, Android was separated, EUV lithography equipment was blocked, and EDA design tools were prohibited... Any key link that could affect chip design and manufacturing was not overlooked by the U.S.

But it was precisely under such an extremely harsh environment that Huawei completed a systematic structural evolution, completely shifting to the domestic architecture path, actively avoiding process bottlenecks, and focusing on the overall system.

At the same time, Huawei realized that if it only made chips, it would easily be choked again. Only by building the entire ecosystem could it have the qualification to stand unshakable in this technological decoupling war.

This strategic transformation has created a closed-loop, self-controlled, and complete AI foundation.

In a way, Huawei did not develop a certain technology due to being pressured, but rather became an enterprise community with national computing power platform capabilities due to being pressured.

This ability gives it the qualification to define boundaries and challenge rules.

Sino-U.S. Tech Rivalry

Looking at the big picture, the establishment of Huawei's AI chip system is no longer just about corporate success, but rather a strategic benchmark for the direction of China's entire chip industry development.

It breaks the past cognitive path where advanced process technology was the only direction, proving that even without access to cutting-edge lithography machines in the long term, China can still achieve strong computing power output through architectural optimization, system integration, and engineering scaling, truly opening up the imagination for asymmetric breakthroughs in the entire industry.

Huawei has expanded the chip industry from the previous vertical chain of design, foundry, and packaging/testing to a full industrial collaboration network covering chips, nodes, clusters, cloud platforms, and ecosystems, prompting the Chinese chip industry to begin horizontal expansion and internal integration, moving from point breakthroughs to systemic construction.

Additionally, it should be noted that Huawei's computing power products, represented by SuperPoD, are becoming an important supplement to national computing infrastructure, providing a key foundation for China to build an independent, secure, and self-controlled AI ecosystem.

U.S. chip technology is still leading, and it can still pressure China, but now Huawei, a company that has not been completely cut off from the chip industry, is already able to produce advanced chips, and in the future, it will certainly not need the U.S.

So, goodbye, Huang Renxun!

Original article: https://www.toutiao.com/article/7552749975025828371/

Disclaimer: This article represents the views of the author and is welcome to express your opinion below with the 【up/down】 button.