Huawei’s AI Ambitions Hit Major Roadblocks
.jpg "Huawei’s AI Ambitions Hit Major Roadblocks")
China’s effort to rival U.S. computing power in AI is encountering significant challenges, with Huawei’s push to develop domestic AI chips facing serious setbacks.
According to a recent Financial Times report, customers using Huawei’s Ascend series chips have reported performance issues and difficulties transitioning from Nvidia’s market-leading products.
Software Struggles and Performance Issues
Huawei has positioned itself as China’s leader in creating a homegrown alternative to Nvidia’s AI chips, particularly following the U.S. tightening export controls on high-performance chips last year.
However, industry insiders, including AI engineers and researchers, say Huawei’s chips still fall short of Nvidia’s, especially during the initial training phase of AI models. This is largely due to limitations within Huawei’s software ecosystem, particularly its Cann software, which users describe as “difficult and unstable.”
Poor documentation is another significant issue, making it hard for users to troubleshoot errors, further complicating the adoption of Huawei’s technology.
Production Challenges
Beyond software, Huawei is also facing major production hurdles. A report from a Chinese outlet in July highlighted the difficulties in ramping up production of the Ascend 910B AI processors. The Semiconductor Manufacturing International Co. (SMIC), responsible for producing these chips, is struggling with low yields—only 20% of the chips produced are functioning as intended. This low success rate is hampering Huawei’s ability to meet demand and advance its AI chip development.
SMIC had aimed to produce 500,000 units of the Ascend 910B chip this year, but equipment failures and outdated manufacturing tools have prevented the company from reaching this goal.
Impact of U.S. Sanctions and Future Prospects
These challenges are compounded by ongoing U.S. trade restrictions, which have forced Chinese manufacturers to rely on older, less efficient equipment. While competitors like Samsung and TSMC use advanced Extreme Ultraviolet (EUV) technology for chip production, Huawei and SMIC are limited to using older Deep Ultraviolet (DUV) machines. This technological gap leads to higher costs, more defects, and frequent equipment malfunctions.
Despite these obstacles, Huawei remains committed to advancing its AI chip development. The company is taking a multi-faceted approach to tackle the issues, including sending engineering teams to help customers transition from Nvidia’s Cuda software to Huawei’s Cann.
However, the road ahead is uncertain. There have been rumors that Huawei is working on more powerful AI chips, like the Ascend 910C, with a possible launch in September. These new chips were reportedly based on a 5nm process, representing a significant technological leap. But given the current production and software difficulties, the timeline and specifications of these new chips remain unclear.
.jpg&description=Huawei’s AI Ambitions Hit Major Roadblocks){kind=link}