What is an NPU, and is it worth paying for?
An NPU (neural processing unit) is a fixed-function accelerator built into a modern SoC to run neural-network maths — matrix multiplication and convolution — at a fraction of the power a CPU or GPU would use. In 2026 laptop and phone chips it is rated at 40-80 TOPS.
What it is for
The NPU targets sustained, low-power, on-device inference: live captions and translation, background noise removal, webcam effects, on-device image generation, and small language models. It runs these continuously without draining the battery or spinning up fans.
What it is not for
It does not accelerate large-model inference the way a GPU does — memory bandwidth and capacity, not TOPS, gate that. A 40 TOPS NPU will not run a 70B model.
| Workload | Best on |
|---|---|
| Always-on features, small models | NPU |
| Interactive 7-13B local LLM | GPU / unified memory |
| 70B+ local LLM | High-VRAM GPU or unified memory |
| Training / fine-tuning | Discrete GPU (CUDA) |
Is it worth paying for? If you keep a laptop for years and use its built-in AI features daily, yes — the efficiency is real. If your plan is to run large local models, spend the money on memory instead.