Huawei’s Kirin 9050 runs a 30-billion-parameter AI model offline.

Inference stays on the handset, with no internet connection or data centre needed.

A mixture-of-experts design activates about 2 billion parameters at a time, and the Da Vinci processor uses low-bit quantization.

Huawei’s account gives no speed or battery figures for the model.

How each outlet framed it
TechNode leans favorable
frames the Kirin 9050 model as a sovereign on-device AI milestone, noting the 2B active-parameter figure was disclosed earlier at Huawei's October briefing

Sources: TechNode