Huawei’s efforts to fill the void left by Nvidia in China’s AI market are facing significant obstacles due to performance issues and slower connectivity speeds with its Ascend AI chips.
Huawei has positioned its Ascend chips as a viable alternative to Nvidia’s hardware, which has been restricted from export to China under U.S. export controls enforced by the Biden administration. However, reports suggest that Chinese AI firms using Ascend have raised concerns about the hardware’s performance, citing stability issues—an essential factor for developers who need reliable hardware for long-duration AI model training.
Additionally, users have noted that Ascend suffers from slower inter-chip connectivity, a crucial feature that allows developers to cluster hardware for efficient AI inference and training. Slower connectivity extends the time required for training processes, which is a critical disadvantage.
Beyond hardware, the software that supports Ascend, called Cann, has also been a source of frustration. Cann is Huawei’s alternative to Nvidia’s well-established Cuda platform, designed to assist developers with data processing. However, even Huawei’s internal teams have expressed concerns about Cann, describing it as “difficult and unstable to use,” according to the Financial Times. An engineer familiar with Baidu’s experience with the Ascend chips also reported frequent crashes with the hardware.
Huawei promoted its latest AI chip, the Ascend 910B, as competitive with Nvidia’s A100 GPU in terms of performance. However, Nvidia has since moved on to its more advanced H100 chip, with the Blackwell series set to launch in 2025, making Huawei’s offering appear less competitive.
Despite these challenges, Zhang Ping’an, CEO of Huawei Cloud Computing Technologies, expressed optimism in July at the World Artificial Intelligence Conference in Shanghai, asserting that China can still lead in AI despite restrictions on access to top-tier hardware. Demand for Huawei’s Ascend chips remains strong, with the company reporting a 34% revenue increase in the first half of 2024.
Yet, Huawei is struggling to meet that demand. Chinese chip manufacturers face difficulties acquiring the machinery needed for chip production due to import restrictions. According to a July report from Korean news site Chosun.biz, Semiconductor Manufacturing International Co. (SMIC), which produces chips for Huawei, has encountered manufacturing problems, with four out of five processes reportedly defective.
Analysts remain cautious about Huawei’s potential to rival Nvidia in the AI chip space. Lian Jye Su, chief analyst at Omdia, noted that Huawei’s AI chipset is “not known for its user-friendliness” due to the relative immaturity of its software platform, Cann, which was introduced in 2018 compared to Nvidia’s Cuda, which has been in development since 2006.
However, Su added that Huawei’s strong client service could help mitigate some of these issues. “They are known to provide responsive and comprehensive on-site support to their clients,” Su said, adding that the enforced decoupling of Nvidia’s chips from the Chinese market has given Huawei an opportunity to thrive despite its shortcomings.
Looking ahead, Huawei may soon face competition from Nvidia once again. Nvidia is reportedly developing AI chips tailored specifically for the Chinese market, designed to comply with U.S. export restrictions. These modified Blackwell chips would be slower than the mainline versions, but Nvidia is eager to regain ground in a market that once contributed 26% of its revenue, a figure that has since dropped to around 17%.
Alvin Nguyen, senior analyst at Forrester, commented: “It’s not surprising that Huawei is facing difficulties in replacing Nvidia. Nvidia’s software ecosystem, particularly Cuda, has been around for a long time and gives them a strong advantage. However, Huawei’s close collaboration with customers may help them gradually narrow the gap.”
Nguyen further noted that displacing Nvidia will require time and substantial effort from Huawei, both in hardware and software development. However, ongoing export controls present opportunities for Huawei to remain competitive as long as Nvidia is restricted from selling its top-tier products in China.
