Xiaomi AI Cube Prototype Debuts With Three Xring Chips for Local AI Computing
Xiaomi has revealed an ambitious new engineering prototype called the AI Cube, a compact mini PC built to run advanced AI models directly on the device. Announced during Xiaomi’s Xring chip technology conference on August 24, the system highlights the company’s growing focus on custom silicon, local AI processing, and high-performance computing hardware.
The Xiaomi AI Cube is not a typical mini PC. Instead of relying on a single all-in-one processor, Xiaomi has equipped the prototype with three in-house Xring chips: the Xring O3, Xring O100, and Xring D100. Together, these chips are designed to handle local AI workloads, high-bandwidth memory tasks, and demanding compute operations.
According to Xiaomi, the AI Cube can deliver up to 150 W of continuous performance. This positions the device as a serious local AI system aimed at running large language models and on-device AI applications without depending entirely on cloud servers.
At the center of the system is the Xring O3, which acts as the main processor. Xiaomi says this chip will also appear in the upcoming Xiaomi 18 Fold. The Xring O3 features a 10-core CPU, a 16-core G2-Ultra NX GPU, and a 200 TOPS NPU for artificial intelligence workloads. It is also built to support Xiaomi’s MiMo on-device AI models, showing the company’s intent to bring more AI features directly to its hardware ecosystem.
For data-heavy AI tasks, Xiaomi uses the Xring O100. This chip is designed for high-bandwidth near-memory computing and uses 6nm 3D wafer-level stacked packaging. Xiaomi claims the O100 includes 28,672 effective data lines and can reach up to 1.22 TB/s of near-memory computing bandwidth. That level of bandwidth could be important for running large AI models efficiently on a compact machine.
The third chip, the Xring D100, is aimed at high-performance AI and intelligent driving workloads. Built on a 3nm process, it includes a 20-core CPU and a 16-core NPU. Xiaomi says the D100 supports up to 160 GB of unified memory and can locally deploy AI models with up to 200 billion parameters.
That makes the AI Cube especially interesting in the growing market for local AI PCs. Compact AI systems from companies such as NVIDIA and Asus have also pushed into this category with high unified memory capacities and support for large local models. Xiaomi’s approach stands out because it combines three of its own chips inside one mini PC, rather than building around a single main processor.
The AI Cube is also designed to handle different types of AI models. Xiaomi says the prototype can deploy both 120B and 3B large models locally and switch between fast and slow systems depending on the workload. This could allow the device to balance speed, power consumption, and model complexity depending on the user’s needs.
Beyond the internal hardware, Xiaomi has given the AI Cube a striking physical design. The chassis uses an aerospace aluminum unibody structure with 33,874 CNC precision perforations. This design likely helps with heat dissipation while giving the device a futuristic look suited to an AI-focused mini PC.
The AI Cube remains an engineering prototype for now, so it is not yet a consumer product. Xiaomi has not announced pricing or a release date. However, company materials suggest that the Xring O100 and Xring D100 chips are planned for commercial launch in 2027.
If Xiaomi brings the AI Cube or a similar system to market, it could become an important step in the rise of compact local AI computers. With growing demand for privacy-friendly AI, faster on-device processing, and reduced reliance on cloud computing, Xiaomi’s custom Xring chip strategy may give the company a stronger position in the next generation of AI hardware.
