By Brenda Goh and Yuhan Lin
SHANGHAI (Reuters) – An intriguing scenario is unfolding in China as a multitude of companies, spanning from chip manufacturers to cloud service providers, eagerly embrace DeepSeek’s innovative artificial intelligence (AI) models. This enthusiastic response has prompted analysts to celebrate what they describe as a “watershed moment” for the industry.
Moore Threads and Hygon Information Technology, notable for producing AI chips and positioning themselves as competitors to Nvidia, announced on Monday that their computing clusters and accelerators are now compatible with DeepSeek’s R1 and V3 models. In a show of appreciation, Moore Threads posted on WeChat, recognizing DeepSeek’s advancements with domestically produced graphic processing units (GPUs) as potentially igniting the spark for China’s AI sector.
On a separate occasion, Huawei Technologies, renowned for its own line of AI chips, disclosed a collaboration with AI infrastructure startup SiliconFlow to make DeepSeek’s models accessible to customers through its Ascend cloud service. Huawei emphasized that the performance of these models is on par with those running on top-tier chips globally.
The integration of DeepSeek’s models with Huawei’s Ascend chips is being hailed as a pivotal moment by Bernstein analysts, who observed that DeepSeek’s achievement demonstrates that competitive large language models (LLM) can be effectively deployed on China’s ‘good enough’ chips, thus reducing dependence on state-of-the-art hardware from the United States. This sentiment was reinforced by the impending release of chips from Cambricon and Hygon, further enhancing China’s AI capabilities.
Cloud divisions of Alibaba, Baidu, and Tencent have also joined the bandwagon by offering access to DeepSeek’s models through their respective services. In a notable development last month, DeepSeek introduced a cost-effective AI assistant that consumes minimal data, quickly surpassing its U.S.-based competitor ChatGPT in downloads from Apple’s App Store within days, triggering a widespread decline in tech stocks.
Previously, DeepSeek garnered global attention with a research paper in December highlighting that the training of DeepSeek-V3 required significantly less computing power, valued at less than $6 million, compared to the substantial investments made by tech giants like Meta and Microsoft using Nvidia’s H800 chips.
The success of DeepSeek has not gone unnoticed, with China embracing the startup, based in Hangzhou, and its founder, Liang Wenfeng, as pop culture icons. While Microsoft and Amazon have also integrated DeepSeek’s models into their cloud services, certain countries such as Italy and the Netherlands have raised privacy concerns and taken measures to block or investigate DeepSeek’s AI application.
(Reporting by Brenda Goh and Lin Yuhan; Editing by Clarence Fernandez)
