When you buy through affiliate links in our content, we may earn a commission at no extra cost to you. Learn 
China’s push to match U.S. computing power in AI is facing hurdles, with Huawei facing software and production challenges slowing its progress.?
According to a recent Financial Times report, customers of Huawei’s Ascend series chips are experiencing performance issues and struggling to transition from industry-leading Nvidia products.
Huawei Software Woes and Performance Issues
Huawei has emerged as China’s frontrunner in developing a domestic alternative to Nvidia’s AI chips, particularly after the U.S. tightened export controls on high-performance silicon in October last year.
However, multiple industry insiders, including AI engineers and researchers, report that Huawei’s chips are still significantly inferior to Nvidia’s, especially for the initial training phase of AI models. This is largely due to Huawei’s software ecosystem, particularly its Cann software, which has several limitations.
Customers and even Huawei’s employees describe Cann as “difficult and unstable to use.” One anonymous researcher highlighted poor documentation as a major issue making it difficult to find the source of errors when they occur.
Beyond software issues, Huawei is grappling with significant production challenges.
In July, a report from a Chinese outlet revealed that Huawei is struggling to expand production of its Ascend 910B AI processors due to insufficient yields at Semiconductor Manufacturing International Co. (SMIC).
The production efficiency for these chips remains extremely low, with only 20% of chips produced by SMIC operating as intended. This low yield rate is impacting Huawei’s ability to further meet demand and develop its artificial intelligence business.
SMIC had initially predicted the production of 500,000 Ascend 910B chip units this year. However, equipment failures and outdated chip-making tools have prevented the firm from achieving this goal.
Impact of US Sanctions, Huawei’s Response and Future Outlook
?The difficulties faced by Huawei and SMIC are aggravated by ongoing U.S. trade restrictions on the Chinese market.
Chinese manufacturers are forced to rely on older, less efficient equipment as they cannot access state-of-the-art chip-making machinery from companies like ASML.
While competitors like Samsung and TSMC use advanced Extreme Ultraviolet (EUV) equipment for chip manufacturing, Huawei and SMIC are limited to older Deep Ultraviolet (DUV) machines.
Wired called it “The $150M Machine Keeping Moore’s Law Alive.”
Meet ASML’s EUV Machine – one of humanity’s greatest engineering feats.
It took 17 years and €6B to develop and can print 20,000 transistors on an area the width of a human hair.
Here is how it works:
1/ 50,000… pic.twitter.com/gnijAIbQtC
— Business Breakdowns (@bizbreakdowns) July 14, 2023
This technological gap results in higher production costs, increased defect rates, and frequent equipment failures.
Despite the array of challenges, Huawei demonstrates resilience and adaptability in its approach to AI chip development.
The Chinese tech giant has implemented a multi-faceted strategy to address the issues at hand.
One key initiative involves dispatching teams of engineers to assist customers in transitioning from Nvidia’s Cuda to Huawei’s Cann software.
However, the path forward remains compounded with obstacles. Recent rumors suggested that Huawei was developing more powerful AI chips, specifically the Ascend 910C, with a potential debut this September.
These chips were reportedly based on a 5nm process, marking a significant advance. Yet, given the current production and software issues, the timeline and specifications of these purported chips are now shrouded in uncertainty.
