Xiaomi to Build GPU Cluster to Advance AI Large Models

Xiaomi is reportedly building its own GPU cluster to support the development of large-scale AI models. This ambitious initiative, which has been underway for several months, is being personally overseen by Xiaomi Chairman Lei Jun, much like the company's earlier ventures into automotive technology.

Xiaomi officially launched its AI Lab's Large Model Team in April 2023, initially equipped with 6,500 GPUs. The company has been deeply invested in AI for several years, with Lei Jun expressing strong commitment to AI's potential. He stated, “For large models, we are fully committed and determined to embrace them. We are developing some exciting technologies and products that we will showcase once fully refined.”

In May 2024, Xiaomi announced its large language model, "MiLM," had officially passed regulatory approval, marking a historic step forward in the company's AI journey. At the World Internet Conference in Wuzhen last month, Lei Jun highlighted Xiaomi's early investment in AI, citing the company's smart voice assistant, Xiao Ai, which boasts 120 million active users. AI is now deeply integrated across Xiaomi's various business sectors.

The AI Lab's Large Model Team has grown rapidly, with over 1,200 researchers dedicated to AI R&D as of last year. Xiaomi's Vice President and CFO, Lin Shiwei, also revealed that the company has been actively exploring strategic investments and collaborations within the AI large model ecosystem, having reviewed the leading teams in the industry.

Additionally, in August 2023, Xiaomi's Senior Vice President and Mobile Division President, Zeng Xuezhong, disclosed plans to launch AI-powered applications on mobile devices, leveraging deep technical partnerships with Qualcomm and MediaTek. These applications, which rely on on-device large models, promise to offer consumers an advanced AI experience.

Later that month, Lei Jun shared that Xiaomi had successfully run a demo of a mobile-based large model (with 1.3 billion parameters), achieving cloud-like performance in certain scenarios. The company also unveiled its self-developed 6 billion-parameter model.

Since forming its first visual AI team in 2016, Xiaomi has expanded its AI capabilities six times, with over 3,000 people now working across areas like computer vision, voice recognition, knowledge graphs, NLP, machine learning, and multimodal AI. Unlike many competitors, Xiaomi's focus is on creating lightweight, locally deployable AI models.

In a significant development, industry reports have revealed that Luo Fuli, a key developer behind the DeepSeek open-source large model DeepSeek-V2, is joining Xiaomi. Luo will lead the AI Lab and oversee the expansion of Xiaomi's large model team, further solidifying the company's push into advanced AI research and development.