Xiaomi Launches Xuanji O3, O100, and D100 Chips

ChainCatcher report: Lei Jun has announced that Xiaomi’s new Xuanjie family chips—Xuanjie O3, Xuanjie O100, and Xuanjie D100—have all completed silicon validation. The Xuanjie O3 is an AI flagship SoC designed for Xiaomi’s premium products and will make its debut on the Xiaomi 18 Fold. The Xuanjie O100 is a high-bandwidth AI accelerator chip with 1.22 TB/s memory bandwidth, while the Xuanjie D100 is China’s first 3nm AI chip for intelligent driving with high computational power; both will enter commercial use next year. These three chips collectively address Xiaomi’s end-side AI computing needs across its entire “person-car-home” ecosystem. The Xuanjie O3 is built on a 3nm process, measures 133mm², and integrates 24 billion transistors—a 26% increase over the previous generation. It achieves an AnTuTu benchmark score of 5.22 million. Its CPU features a 10-core all-big-core design with a maximum clock speed of 4.35GHz, delivering Geekbench single-core scores of 3,945 and multi-core scores exceeding 15,000. The GPU includes 16 cores with ray tracing support, significantly improving graphics performance while reducing power consumption by 64%. This chip is the first in the industry to support LPDDR6, features a 200 TOPS NPU Tensor computing capability, and includes dedicated optimizations for large language model inference. The Xuanjie O100 is the world’s first 6nm wafer-level 3D-stacked AI accelerator chip, utilizing Wafer-on-Wafer packaging with hybrid bonding to vertically stack two layers of AI-optimized high-speed DRAM with one layer of high-performance NPU, achieving a bandwidth of 1.22 TB/s. When paired with the O3, it enables a maximum local AI inference speed of 330 tokens/s. Xiaomi has been rebuilding its large-scale chip development capabilities for over five years, investing more than RMB 21 billion and assembling a team of nearly 3,000 engineers.