Although Huawei first gained global recognition as a smartphone maker, the chip designer is evolving into a foundational technology pillar for China. Much like Intel (NASDAQ:) is a strategic asset for the US, Huawei Technologies (HWT.UL) is broadening its mandate across semiconductors, AI, telecommunications, and ultimately, ensuring China’s self-reliance amid stiff US export controls.
With DeepSeek, China has already demonstrated it can upset the AI stock arena. According to the Wall Street Journal’s sources, a deeper disruption could be underway with Huawei’s new AI chip Ascend 910D. Purportedly, the D version is set to outperform Nvidia’s H100 chips, the series that trained most of today’s large language models (LLMs).
Previously, we covered that 14% of Nvidia’s (NASDAQ:) revenue comes from China, and likely up to 33% if Singapore is included as a transshipment point. With Huawei’s latest push, could Nvidia find itself in a situation similar to Tesla (NASDAQ:) against Chinese BYD?
Current Ranking of AI Chips
At present, Nvidia’s flagship B200 chip, built on Blackwell architecture, has no match in terms of performance. With that being said, it also matters how chips are deployed. Case in point, Hewlett-Packard Enterprise’s El Capitan currently ranks as the world’s fastest supercomputer, employing AMD’s EPYC and Instinct MI300A accelerators.
But owing to Nvidia’s full stack approach that integrates hardware, networking, and software, the company’s products became go-to solutions for AI workloads in data centers. In the first wave of the AI hype, Nvidia mostly sold A100 accelerators (Ampere architecture) to hyperscalers such as Meta (NASDAQ:), Microsoft (NASDAQ:) and Alphabet (NASDAQ:), followed by H100s (Hopper architecture).
This all changed in October 2022 when the U.S. Commerce Department implemented export controls for both the A100 and H100 series. Less than a year later, Nvidia nerfed both into A800 and H800 to continue its China sales.
DeepSeek R1 leveraged H800 for training (pre-deployment) combined with Huawei’s Ascend 910C chip for inferencing (post-deployment). According to the DeepSeek team, the Ascend 910C gives approximately 60% performance of H100, built with a 7nm node process. For comparison, both Hopper and Blackwell accelerators are built with TSMC’s 4nm process, enabling greater transistor density for more computational power.
Just on time for even stricter export control, Huawei’s Ascend 910C chips are purportedly set for mass volume production this May. In the next phase of AI scaling, Ascend 910D is still not competing with Nvidia’s H200 or B200, but is sandwiched between H100 and H200.
Is China’s AI Chip Self-Reliance Inevitable?
China still has trouble with the mass deployment of AI chips at scale. To deploy Ascend 910C, which combines two Ascend 910B dies, Huawei had to erect a number of shell companies to fool TSMC (NYSE: TSM) into manufacturing them, which was first exposed by The Information.
It is clear that China is in the intermediary period where large smuggling networks are needed to neutralize the Commerce Department’s chip restrictions. Long-term, however, China created its own AI industrial complex by combining the efforts of Huawei as chip designer, SMIC as chip manufacturer, and CXMT/XMT as high-bandwidth memory (HBM) manufacturer.
According to the Center for Strategic & International Studies (CSIS), it also appears that China stockpiled over a year’s worth of HBM supply.
In other words, China has to build up its own ecosystem of cutting-edge chip companies. The US relies on its hegemonic power to control Taiwan’s TSMC as well as South Korea’s SK Hynix for HBM needs. And all three chip designers, Nvidia, AMD (NASDAQ:), and Intel, are headquartered in California.
As both Intel and TSMC work on deploying chips using the 2nm node process this year, China is still one generation behind, looking to launch 5nm tech instead. Of course, the US also has leverage against Dutch ASML Holding (AS:), which makes it difficult for China to deploy the most advanced chip-making equipment.
Yet, a recent report from William Huo suggests that China’s SMIC can deploy 5nm chips regardless. Based on these trends of China’s advance, it is reasonable to assume that any export chip controls against China will be neutralized by 2030.
In the meantime, China can use its 90% hold on rare minerals refining, used in delicate electronics, as a powerful leverage.
***
Neither the author, Tim Fries, nor this website, The Tokenist, provide financial advice. Please consult our website policy prior to making financial decisions.
This article was originally published on The Tokenist. Check out The Tokenist’s free newsletter, Five Minute Finance, for weekly analysis of the biggest trends in finance and technology.