### China’s AI Aspirations Confront Hurdles with Huawei’s Ascend Chips
China’s endeavor to compete with the United States in the field of artificial intelligence (AI) is facing considerable challenges, especially concerning computing power. Central to this difficulty is Huawei, a prominent Chinese tech firm striving to create AI chips that can stand toe-to-toe with industry titan Nvidia. Nevertheless, Huawei’s ambitions have been hindered by software complications and performance issues, leading to frustration among numerous clients.
#### The Contest for AI Dominance
As the United States intensifies export restrictions on high-performance silicon, Chinese firms are compelled to look for domestic substitutes for Nvidia’s robust AI chips. Huawei has positioned itself as a key player in this contest, with its Ascend series of AI chips gaining traction among Chinese AI organizations. These chips are notably utilized for inference, an essential process in AI applications such as OpenAI’s ChatGPT, which formulates replies to user inquiries.
In spite of its rising acclaim, Huawei’s Ascend chips have drawn criticism regarding their performance, particularly during the initial training phases of AI models. Industry experts have highlighted various concerns, including stability challenges, slower inter-chip communication, and the limitations associated with Huawei’s software platform, Cann.
#### The Software Dilemma: Cann vs. Cuda
One of the primary hurdles for Huawei is crafting a software platform that can rival Nvidia’s Cuda. Cuda is widely recognized as Nvidia’s “secret ingredient,” providing an accessible interface that significantly expedites data processing. In contrast, Huawei’s Cann software has been deemed cumbersome and error-prone.
Even within Huawei, there is some discontent concerning Cann. A researcher from the company commented on the software being “challenging and unreliable,” adding that inadequate documentation complicates issue identification and resolution. This has impeded the development timeline, as skilled developers must delve into the source code to troubleshoot concerns.
The situation is exacerbated by reports that Huawei’s chips frequently crash, particularly during AI development tasks. This has resulted in dissatisfaction among customers, including major Chinese tech firms like Baidu, iFlytek, and Tencent.
#### Huawei’s Strategy: On-Site Assistance and Price Hikes
To address these issues, Huawei has implemented various measures to aid its clientele. The company has deployed teams of engineers to help clients transfer their training code from Nvidia’s Cuda to Huawei’s Cann. This proactive strategy has been positively acknowledged by certain customers, who value Huawei’s dedication to service.
“Huawei is exceptional in customer service, so it’s no surprise they have engineers present at their significant clients, assisting with chip usage,” noted a former Baidu staff member.
Huawei’s substantial workforce, with over 50 percent of its 207,000 employees engaged in research and development, has been a critical advantage in this initiative. Additionally, the company has established an online platform for developers to share feedback on potential software enhancements.
However, the hurdles aren’t confined solely to software. Huawei has also increased the price of its Ascend 910B chip, utilized for training AI models, by 20 to 30 percent. This price adjustment is likely a reaction to supply challenges, as Chinese firms struggle to acquire cutting-edge chipmaking machinery from Dutch company ASML owing to export limits.
#### The Future Path: Navigating Demand and Development
Despite these obstacles, Huawei has experienced robust demand for its AI chips. The company reported a 34 percent surge in first-half revenues, although it did not disclose a breakdown of sales across various businesses. As per Huawei executive director Zhang Ping’an, over 50 foundational AI models have been “trained and refined” on the Ascend chip.
iFlytek, a Chinese AI corporation, has also noted success with Huawei’s chips, declaring that its large language model has been trained solely on Ascend chips. Huawei’s engineers played a pivotal role in integrating this technology at iFlytek’s headquarters in Hefei, eastern China.
While Huawei’s AI chips have made notable progress, the company still confronts a formidable challenge in its pursuit of matching Nvidia’s preeminence in the AI chip sector. The ongoing challenges of developing stable and user-friendly software, in conjunction with supply limitations, will continue to test Huawei’s competitiveness on the international stage.
As China advances its ambition to become a frontrunner in AI, the success of firms like Huawei will be vital. However, surmounting current challenges will necessitate not just technological breakthrough but also strategic collaboration and support from the broader Chinese tech landscape.
