Glossary

NPU

Neural Processing Unit — a chip dedicated to running machine-learning inference efficiently.

Definition

An NPU is hardware specialized for the math machine-learning models need: lots of low-precision multiply-accumulate operations, run in parallel. NPUs are now common in flagship phones and modern PCs, and they let AI features run on-device instead of in the cloud.

On-device AI matters for privacy and latency. Image cleanup, transcription, summarization and assistants that run on the NPU can stay entirely on the device — a different security model than cloud AI.

Example

Your phone's voice typing transcribes you in real time without sending audio to a server. The transcription model runs on the phone's NPU.

Frequently asked questions

Is NPU AI safer than cloud AI?

By default yes — data doesn't leave the device. But the model still has access to whatever you give it.

Do I need an NPU?

Not for most tasks. It becomes useful as more applications integrate on-device AI features.