Xiaomi unveiled the Xring D100 on 24 August, presenting it as China's first high-computing-power smart-driving chip on a 3nm process, manufactured by TSMC. It arrived alongside the Xring O100, a 6nm neural processing chip, and the Xring O3 smartphone processor.
What was specified
The D100 carries a 20-core high-performance CPU and a 16-core NPU, and supports up to 160GB of unified memory. Xiaomi says development is complete and commercial deployment is slated for 2027. The headline capability claim is that it can run models of up to 200 billion parameters locally.
What the common framing gets wrong
This is being written up as Xiaomi's self-driving chip running 200-billion-parameter models in a car. The 200B claim was demonstrated on a desktop AI box, not in a vehicle — a different thermal envelope, a different power budget and a different memory configuration from anything that ships behind a windscreen. A part that can host a model on a desk is not thereby shown to run it at automotive temperatures, on automotive power, inside a latency budget where the answer has to arrive before the car does.
The three numbers that are missing
For a driving accelerator, the conventional headline figure is TOPS. Xiaomi did not give one. It also did not name the launch vehicle or give an installation timetable beyond the year. "Development complete, commercial use 2027" is a statement about a schedule, not about a product anyone can buy or benchmark, and the absence of a throughput number in an announcement otherwise dense with core counts is conspicuous rather than incidental.
The strategic read
The consistent thread is vertical integration: Xiaomi is designing across phones, general AI acceleration and now driving, following the pattern set by Apple, Samsung and Huawei. Reuters reported the TSMC manufacturing relationship on the basis of sources rather than a Xiaomi statement, which is worth noting alongside a launch where the company was otherwise specific about what it wanted specified.