Feb 5, 2025

From Moore Threads to Iluvatar Corex, China’s leading chip designers are rushing to adopt DeepSeek’s latest models. — SCMP DeepSeek has given China’s artificial intelligence (AI) push a shot in the arm, as the country’s chip developers and cloud service providers rush to support the start-up’s increasingly popular models. Moore Threads Technology, a graphics processing unit (GPU) design company created by former Nvidia China general manager Zhang Jianzhong, said in a WeChat post on Tuesday that it would “pay tribute to DeepSeek” by “using locally-made GPUs to set China’s AI ecosystem on fire”. The chip firm said DeepSeek’s open-source V3 and R1 models had “greatly promoted” AI development and provided “inspiration” for developers. “To push forward the development of the domestic AI ecosystem, Moore Threads will open its proprietary KUAE GPU intelligent computing cluster to fully support the distributed deployment of DeepSeek’s V3 and R1 models,” the company wrote, referring to its full-stack solution for AI data centres based on its own chips. Moore Threads’ pledge comes days after China’s tech champion Huawei Technologies said it was working with Beijing-based AI infrastructure start-up SiliconFlow to make DeepSeek’s models available to end users through the telecoms giant’s Ascend cloud service. It offers computing performance matching that of DeepSeek models running on global premium GPUs, according to Huawei. Huawei’s Ascend cloud service relied on its home-grown Ascend solution for compute resources, which could involve various types of hardware including the company’s self-developed server clusters, AI modules and accelerator cards, its website said. Gitee AI, a Shenzhen-based one-stop service website for AI developers, also said it was offering four DeepSeek-R1-based models through servers powered by GPUs from Shanghai-based chip designer MetaX. “The combination of DeepSeek-R1, MetaX GPUs and the Gitee AI platform has realised the full use of domestically developed and made-in-China technologies from chips and platforms to computing power and models,” Gitee AI said in a WeChat post on Sunday. Gitee AI is also collaborating with Shanghai-based GPU start-up Iluvatar Corex, which said on Tuesday it was working with partners to ensure quick and easy access to DeepSeek-R1. “The launch of DeepSeek-R1 was like a bolt of lightning that crushed long-standing technical barriers and injected new vitality into China’s AI industry,” Iluvatar Corex said in a WeChat post. Besides GPU developers, China’s major cloud service providers – including Alibaba Group Holding, Tencent Holdings, Huawei and Baidu – worked overtime during the Lunar New Year holiday to support DeepSeek’s new models on their respective platform, as demand from Chinese consumers and businesses soared. Alibaba owns the South China Morning Post. While AI chips produced by Chinese developers remain less sophisticated than products from US GPU giant Nvidia, the success of DeepSeek has provided hope that China could narrow its gap with the US by focusing on so-called inference processors, which are designed to support generative AI tools that have already been trained. Inference chips are lighter-weight, more efficient and easier to develop than training chips. Deeper integration between chip developers and software engineers in China is also expected to enable the smooth operation of home-grown open-source models and allow domestic AI systems to become less reliant on US chips. – South China Morning Post Open Modal