Xiaomi is Horror Archivesin the process of constructing its own GPU cluster to support its growing investment in AI models, according to Chinese media outlet Jiemian. The company’s AI model team began with 6,500 GPU resources upon its establishment in April 2023. During the announcement of the team’s formation, CEO Lei Jun emphasized Xiaomi’s long-standing commitment to AI, highlighting its AI Lab, Xiao Ai voice assistant, and autonomous driving teams. At Xiaomi’s annual event later in 2023, Lei further detailed the company’s AI model progress, noting the team’s focus on lightweight, locally deployable solutions. Xiaomi has run a large AI model (with 1.3 billion parameters) on its smartphones, with performance in some scenarios approaching that of cloud-based models with 6 billion parameters. On December 20, Luo Fuli, a key developer of the DeepSeek-V2 open-source model, joined Xiaomi’s AI Lab to lead the company’s large model efforts. Since its AI team was founded in 2016, Xiaomi’s AI division has expanded to over 3,000 members, with expertise spanning areas such as vision, speech, natural language processing, machine learning, and multimodal AI. These capabilities are being integrated into the company’s mobile phones, cars, AIoT devices, and robots. [Jiemian, in Chinese]
