Xiaomi Builds GPU Cluster with 10,000 Cards, Increasing Investment in AI Large Models

Reports indicate that Xiaomi is actively building its own GPU cluster with thousands of cards to increase investment in large AI models. This initiative, which has been ongoing for several months, is led by Xiaomi's founder Lei Jun.

Xiaomi's involvement in the AI sector is not a recent development. In April 2023, Xiaomi's AI Lab officially established a large model team, headed by Luan Jian, who reports to Wang Bin, the vice chairman of Xiaomi's Technology Committee and the director of the AI Lab. Luan Jian has extensive experience in AI, having previously served as the chief voice scientist and head of the voice team at Microsoft Xiaoice.

Signs of Xiaomi's increased focus on large AI models have been evident. Earlier, it was reported that Luo Fuli, a key developer of the open-source large model DeepSeek-V2, would join Xiaomi, possibly working in the AI Lab and leading the large model team. The innovative architecture of DeepSeek-V2, particularly in reducing the cost of using large models, is significant for Xiaomi's AI model development.

Lei Jun has expressed his views on large AI models and Xiaomi's commitment to this field in multiple public appearances. He stated that Xiaomi has been working in AI for many years, with teams such as the AI Lab, Xiao Ai, and autonomous driving. For large models, Xiaomi is fully committed and determined to embrace them.

According to insiders, the main focus of Xiaomi's large model team since its establishment has been on lightweight and local deployment. Currently, Xiaomi has successfully run large models on mobile devices, achieving performance close to cloud-based models in some scenarios. Additionally, Xiaomi is actively developing new technologies and products, planning to showcase its latest achievements in AI large models in the future.

Moreover, Xiaomi's AI team has been expanding. Since forming the AI team in 2016, the company has seen multiple expansions, with the current team size exceeding 3,000 people. Its AI capabilities cover areas such as vision, acoustics, speech, NLP, knowledge graphs, machine learning, large models, and multimodal directions, and are gradually being integrated into various business segments, including smartphones, automobiles, AIoT, and robotics.

Regarding the plan to build a GPU cluster with thousands of cards, Xiaomi has not yet commented. However, industry experts believe that this move will significantly enhance Xiaomi's R&D capabilities and competitiveness in the field of large AI models.