In an exciting development, the emergence of DeepSeek’s artificial intelligence (AI) models is poised to enhance the competitive landscape for Chinese chipmakers, including Huawei, allowing them to better contend with the more potent U.S. processors in the domestic market. Chinese companies like Huawei have long grappled with the challenge of keeping pace with Nvidia in producing high-end chips capable of rivaling the American company’s products for training models, a crucial process wherein algorithms learn from data to improve decision-making accuracy.
DeepSeek’s models, which prioritize “inference” – the phase during which an AI model draws conclusions – focus on optimizing computational efficiency rather than solely relying on raw processing power. Analysts anticipate that this approach will help narrow the performance gap between Chinese-made AI processors and their more powerful U.S. counterparts. Recent statements from Huawei and other Chinese AI chipmakers such as Hygon, EnFlame backed by Tencent, Tsingmicro, and Moore Threads indicate their intention to support DeepSeek models, although specific details are currently scarce.
The open-source nature and affordability of DeepSeek’s technology are expected to accelerate AI adoption and foster the creation of practical applications, aiding Chinese companies in overcoming U.S. export restrictions on their most advanced chips. Even prior to the spotlight on DeepSeek this year, products like Huawei’s Ascend 910B were favored by companies like ByteDance for less computationally demanding inference tasks, where trained AI models make predictions or perform functions like chatbots.
Numerous Chinese firms spanning industries from automotive to telecommunications have announced plans to integrate DeepSeek’s models into their products and operations. Lian Jye Su, a chief analyst at tech research firm Omdia, noted that Chinese AI chip vendors excel in inference workloads, benefiting from localized expertise and industry-specific knowledge, though they still face stiff competition from Nvidia in AI training due to the latter’s superior GPU technology.
Despite Chinese AI chips being cost-competitive for inference tasks within China, Bernstein analyst Lin Qingyuan emphasized that Nvidia’s chips remain superior globally, even in inferencing. While U.S. export restrictions limit the entry of Nvidia’s most advanced training chips into China, the company is permitted to sell less powerful training chips for use in inference tasks by Chinese customers. Nvidia emphasizes the importance of its hardware and software ecosystem, including CUDA, a key platform enabling general-purpose computing on Nvidia GPUs, in maintaining its competitive edge.
Huawei stands out for its proactive approach in seeking alternatives to Nvidia’s technologies, signaling a broader trend toward diversification and innovation in the Chinese AI chip market.
Considering a CUDA equivalent known as Compute Architecture for Neural Networks (CANN) is under development, experts have expressed concerns regarding the challenges in convincing developers to switch from CUDA. Omdia’s Su highlighted that the software performance of Chinese AI chip companies currently falls short. He pointed out that CUDA’s extensive library and wide range of software capabilities necessitate substantial long-term investments. The information was reported by Liam Mo and Brenda Goh, with editing by Sam Holmes.
