Mobile AI hardware race: NPUs and UFS 5.0 boost on‑device intelligence

Smartphone makers are redesigning chips to run artificial‑intelligence models locally, adding dedicated Neural Processing Units (NPUs) that promise lower latency, better privacy and reduced power use. Apple has expanded its Neural Engine, Qualcomm is equipping Snapdragon silicon with powerful NPUs, and Samsung is enhancing Exynos processors for AI‑driven photography and assistants.

At the same time, Japanese memory supplier Kioxia announced commercial UFS 5.0 flash modules – 1 TB and 512 GB versions – that deliver up to 10 GB/s sequential read and 9 GB/s write speeds. The new HS‑Gear6 mode and eighth‑generation BiCS FLASH aim to eliminate the memory bottleneck that has limited on‑device AI performance, with mass production slated before the end of 2026.