An NPU (neural processing unit) means a small chip built only for AI, usually found inside phones and newer laptops. It lets AI features run on the device itself, using little battery, instead of sending your data to a company’s servers.
Think of a kitchen. The main processor is the chef who can cook anything. The graphics chip is a big, power-hungry oven. The NPU is a small, efficient appliance made for one job, like a kettle: it does that job quickly without heating the whole house.
In a sentence: “My new laptop has an NPU, so it can blur my background on video calls without draining the battery.”
What it means
An NPU is a specialist chip for AI maths. IBM, which sells computing systems to businesses, explains how it differs from the other chips in a computer.1
The selling point is efficiency. Like a graphics chip (GPU), an NPU does many small sums at once, but it uses less power doing them. IBM says NPUs “can perform similar parallel processing with far better power efficiency.”1 “Neural” refers to neural networks, the kind of maths that AI models are made of.
A simple example
Live captions and video-call touch-ups are typical NPU jobs. In a phone or laptop, the NPU usually sits on the same piece of silicon as the other processors. IBM says phones and laptops usually combine the NPU with other processors on one chip, called a “system-on-chip (SoC)”.1
Microsoft sells a class of Windows laptops, called Copilot+ PCs, built around an NPU. It says the chip enables “AI tasks like real-time translation, image generation, and intelligent search to run locally.” Running locally means on the laptop, not in the cloud. Its list of such features includes live captions that translate speech, “even offline”, and effects that improve video calls.2
Why it matters to you
- Privacy. A task done on your device’s NPU does not need to send your words or photos away to be processed. Check each app, though: many AI features still use the cloud even on a device with an NPU.
- Big chatbots still live in the cloud. An NPU handles small, quick jobs. The large models behind leading chatbots run on racks of big chips in data centres. Local or cloud weighs the two.
- “AI PC” labels. If a laptop is sold on its NPU, treat that as one feature among many, and check whether the apps you use can actually make use of it.
Often confused with
| Word | How it differs |
|---|---|
| GPU | Differs: a bigger, more powerful chip first made for graphics. It trains and runs large models, but uses much more power. See GPU meaning. |
| CPU | Differs: the main, general-purpose processor. It can run AI, just less efficiently. |
| AI chip / AI accelerator | Differs: a broad label for any chip that speeds up AI. IBM notes the overlap: NPUs are “also known as an AI chip or AI accelerator”.1 |
Questions people ask
NPU vs GPU: what is the difference?
A GPU is a powerful general chip first made for graphics; an NPU is a smaller chip made just for AI. IBM says NPUs do similar parallel work “with far better power efficiency”.1 See GPU meaning.
What does an NPU do on a laptop?
It runs AI features on the machine itself. Microsoft lists tasks “like real-time translation, image generation, and intelligent search” running locally on its Copilot+ PCs.2
Is an NPU the same as an AI chip?
Mostly, yes: the names overlap. IBM says NPUs are “Also known as an AI chip or AI accelerator”.1
Learn more on this site
- Local or cloud: where does AI live next? — the case for AI on your own device.
- What AI actually runs on — the big chips and data centres at the other end.
- Run AI on your own computer — trying local AI on a laptop or desktop.
- Inference meaning — the job an NPU does when it runs a model.
- The AI glossary and AI for beginners — the basics, in plain words.
Related on this site: Small language model (SLM) meaning · GPU vs NPU vs TPU
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What is an NPU?
- A small chip built only for AI, found in phones and newer laptops
- The main processor that runs everything
- A cloud service for big chatbots
Show the answer
Answer: A small chip built only for AI, found in phones and newer laptops. An NPU lets AI features run on the device itself, using little battery.
2. How does an NPU compare with a GPU?
- An NPU uses far more power than a GPU
- Both do many small sums at once, but an NPU uses much less power
- They are exactly the same chip
Show the answer
Answer: Both do many small sums at once, but an NPU uses much less power. Like a kettle next to a big oven, the NPU does its job efficiently, while the GPU trains and runs large models.
3. Does a device with an NPU run every AI feature on the device?
- Yes, nothing ever goes to the cloud
- Yes, as long as the battery is full
- No, many AI features still use the cloud, so check each app
Show the answer
Answer: No, many AI features still use the cloud, so check each app. The NPU handles small, quick jobs; large chatbot models still run in data centres.
Related terms: GPU · Local LLM / on-device AI · Quantization · Small language model (SLM) · VRAM
- 1IBM, What is a Neural Processing Unit (NPU)?, undated, read at source 1 Oct 2026“Differing from general-purpose central processing units (CPUs) or graphics processing units (GPUs), NPUs are tailored to accelerate AI tasks and workloads”. ↑a“NPUs can perform similar parallel processing with far better power efficiency.” ↑b“most consumer applications, such as smartphones, mobile devices and laptops, combine the NPU with other coprocessors on a single semiconductor microchip known as a system-on-chip (SoC).” ↑c“Also known as an AI chip or AI accelerator, NPUs are typically used within heterogeneous computing architectures that combine multiple processors.” ↑d“Differing from general-purpose central processing units (CPUs) or graphics processing units (GPUs), NPUs are tailored to accelerate AI tasks and workloads”. “NPUs can perform similar parallel processing with far better power efficiency.” ↑e“Also known as an AI chip or AI accelerator, NPUs are typically used within heterogeneous computing architectures that combine multiple processors.” ↑f
- 2Microsoft, Copilot+ PCs & Windows PCs: Differences?, 19 Dec 2025, read at source 1 Oct 2026“enabling AI tasks like real-time translation, image generation, and intelligent search to run locally.” and “Windows Studio Effects: Enhance video calls with intelligent lighting, gaze correction, and voice clarity.” ↑a“enabling AI tasks like real-time translation, image generation, and intelligent search to run locally.” ↑b
The comparisons and examples on this page (the kitchen and the kettle, the advice on checking apps and shop labels) are this site’s own ways of explaining, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK