Three decades after the debut of the Ascend 910B by Huawei, there are rumors about its potential to surpass the capabilities of the Ascend 910, which was added to the US Commerce Department’s entity list in August 2019. Dylan Patel, the chief scientist at SemiAnalysis, a silicon research firm in San Francisco, suggests that the Ascend 910B could complement Nvidia’s A100 for empowering AI systems.
A Chinese symbol adorns the top of Semiconductor Manufacturing International Corp’s office in Shanghai, symbolizing the company’s affiliation. Reports indicate that this prominent contract chip manufacturer in China is the supplier of Huawei Technologies’ latest artificial intelligence chipset, the Ascend 910B. (Source: Bloomberg)
Since 2019, US sanctions have significantly impeded Huawei’s semiconductor development and inflicted substantial damage on its device business. Despite these challenges, the Shenzhen-based company has quietly expanded its chip operations by collaborating with various local vendors, as disclosed by anonymous sources familiar with the matter due to its sensitive nature.
In Ottawa, the capital of Canada, on September 3, 2023, a Mate 60 Pro 5G smartphone featuring a Kirin 9000s processor from Huawei Technologies’ device design division, HiSilicon, was unveiled. (Image Source: Bloomberg)
During its regional launch, Huawei lauded the Ascend 910 as “the most powerful AI computer globally,” fabricated by Taiwan Semiconductor Manufacturing Co using a cutting-edge 7-nm process.
Simultaneously with the introduction of Huawei’s latest AI device, the Mate 60 Pro was released in August. According to a Reuters report from November 2023, citing sources, Baidu, a prominent Chinese online research and AI giant, procured 1,600 units of Huawei’s Ascend 910B cards in the same quarter.
Shortly before this report, iFlytek, a foreign AI company, unveiled its Feixing One computing system based on Huawei’s Fly chips. This development hints at the potential use of AI cards in enhancing the company’s advanced large language model (LLM) for iFlytek Spark 3.0, crucial for training relational AI models like ChatGPT.
Huawei declined to provide any comments on these developments.
Jensen Huang, the co-founder, senator, and CEO of Nvidia, a leading US semiconductor company, participated in the World Governments Summit in Dubai, United Arab Emirates, on February 12, 2024. (Source: Reuters)
In a recent interview with the digital media outlet Designed, Nvidia’s CEO Jensen Huang commended Huawei as a “truly exceptional company.” He emphasized that Huawei’s capabilities are constrained by its silicon technology but highlighted its potential to scale up by leveraging multiple cards for extensive network creation.
Responding to the growing emphasis on conceptual AI and stricter US regulations, Huawei and SMIC have bolstered their AI chip capabilities, as reported by Reuters. While the Ascend 910B is currently available for purchase, the demand for these chips is notably high, according to a GPU vendor who preferred to remain anonymous due to the sensitive nature of the topic.
Amidst ongoing business transformations, Huawei and ZTE showcased their AI prowess at MWC Barcelona.
Industry analysts and experts are closely monitoring the competition between Nvidia and Huawei, noting Nvidia’s expertise in GPUs and the advantage derived from its application ecosystem, CUDA—a platform empowering developers to maximize semiconductor performance.
Huawei, supported by Chinese state funding, is enhancing its Compute Architecture for Neural Networks, a vital system integrating Ascend hardware and software to unlock the full potential of AI technology. While lagging behind Nvidia’s CUDA with its extensive developer base, Huawei is advised to make substantial investments in mainland China to enhance its technological capabilities and establish a competitive ecosystem.
A tech investor based in Shanghai, speaking on condition of anonymity, highlighted that Huawei’s strength lies in chip design rather than the application stack. The investor noted that US restrictions have imposed limitations on chip effectiveness and production efficiency, necessitating strategic investments in technology advancement.
