Open any laptop advert in 2026 and you will see a number you have probably never thought about: 40 TOPS, 45 TOPS, 80 TOPS. It sits next to the processor speed and the memory, and it belongs to a chip called the NPU — the neural processing unit. It is the reason your next laptop can blur your background on a video call, transcribe speech live, and edit photos without ever sending your data to the cloud. Here is what an NPU is, what that number means, and whether you should care.

What an NPU actually is

A neural processing unit is a piece of silicon designed to do one job very well: the maths behind neural networks. AI models run on the same small calculation repeated billions of times — multiply numbers together, add them up, and do it again. A normal processor can do that maths, but it is a generalist, built to do everything from opening files to running spreadsheets. An NPU is a specialist, built to do that one calculation over and over, using very little power.

Think of it like a kitchen. The CPU is the head chef: it can cook anything, but it does one dish at a time. The GPU is a team of line cooks: it can chop thousands of vegetables at once, which is why it became the workhorse of AI training. The NPU is a machine built to make exactly one dish, thousands of times a day, using almost no electricity. It is not more powerful than the GPU — it is more efficient at one narrow job.

That efficiency is the whole point. AI features that run all day — background blur on a video call, noise removal, live captions — would drain your battery if the main processor handled them. The NPU does them in the background, quietly, for a fraction of the power.

NPU vs CPU vs GPU

CPU GPU NPU
Built for General tasks, one after another Thousands of parallel calculations Neural-network maths, repeated
Best at Running apps, the operating system Gaming, graphics, training AI models On-device AI features, all day
Power use Moderate High Very low
AI role Runs small AI tasks, coordinates everything Trains big models, runs heavy AI Runs everyday AI features efficiently
Example Intel Core, AMD Ryzen, Apple M-series NVIDIA GeForce, AMD Radeon Qualcomm Hexagon, Apple Neural Engine, Intel AI Boost

The three chips do not compete — they share the work. Modern laptops have all three, and the operating system sends each job to the chip best suited to it. That is why the NPU is usually part of the main processor rather than a separate chip you can buy on its own.

What does TOPS mean?

TOPS stands for trillions of operations per second — how many simple AI calculations the NPU can do in one second. Forty TOPS means forty trillion operations a second, which sounds enormous until you remember that each operation is a tiny piece of arithmetic.

There are three catches that adverts do not mention:

  1. It is a peak figure, not a measured one. TOPS is the theoretical maximum under ideal conditions — every unit busy, all the time. Real work is rarely that tidy, and AMD itself notes in its own footnotes that the figure is a maximum in an optimal scenario and "may not be typical".
  2. Precision changes the number. TOPS is usually quoted at INT8, where each number is squeezed into a single byte. Some vendors quote different precisions, so two chips with the same TOPS rating may not be equally fast.
  3. Vendors count different things. Some quote the NPU alone; others quote the whole platform — CPU plus GPU plus NPU added together. For Microsoft's Copilot+ badge, only the NPU figure counts, but adverts do not always make that clear.

So treat TOPS as a rough size guide, not a precise speed test — the same way you would treat megapixels on a camera. It tells you something, but it is not the whole story.

The 40 TOPS line: Copilot+ PCs

The number everyone quotes comes from Microsoft. To earn the Copilot+ PC label, a Windows laptop needs an NPU rated at 40 TOPS or more, along with at least 16 GB of memory and 256 GB of storage. The label unlocks Windows AI features that run on the laptop itself: Recall (a searchable timeline of what you have seen on screen), live captions with real-time translation, Windows Studio Effects for video calls, and Cocreator image tools in Paint.

Here is where today's chips stand, using the manufacturers' own published peak figures for the NPU alone:

Chip family NPU rating (peak) Copilot+ eligible
Qualcomm Snapdragon X Elite 45 TOPS Yes — the first Copilot+ PCs, 2024
Intel Core Ultra 200V (Lunar Lake) 40–48 TOPS Yes
Qualcomm Snapdragon X2 Elite 80–85 TOPS Yes
AMD Ryzen AI 300 series Up to 50 TOPS Yes
AMD Ryzen AI 400 series Up to 60 TOPS Yes
Intel Core Ultra Series 3 Up to 50 TOPS Yes
Apple Neural Engine (M-series) Varies by chip Apple uses its own AI features, not Copilot+

Qualcomm's flagship figure nearly doubled in about eighteen months — from 45 TOPS in the first Snapdragon X Elite to 80–85 TOPS in the X2 Elite. The race is moving fast, and the 40 TOPS line that felt ambitious in 2024 is now the entry ticket.

Note that some features need less. Microsoft says Windows Studio Effects — automatic framing, background blur, eye-contact correction and voice focus — runs on any NPU rated at 10 TOPS or more. The 40 TOPS threshold is a tier badge for the full Copilot+ set, not the minimum for every AI feature.

What does an NPU actually do for you?

Today, the visible benefits are modest but real:

  • Video calls: background blur, automatic framing, eye-contact correction and noise removal, all without draining the battery.
  • Live captions: real-time transcription and translation of calls and videos, processed on your device so the audio never leaves it.
  • Photo tools: removing objects, restyling images and generating pictures in apps like Paint's Cocreator.
  • Search: Recall lets you find something you saw on screen earlier by describing it — though this feature has raised privacy questions, and Microsoft has made it opt-in with extra security controls.
  • Battery life: because the NPU sips power, these features can run in the background all day without the fan spinning up.

The key point: an NPU only helps when the app is built to use it. The AI services most people know — ChatGPT, Claude, Gemini — run their calculations on distant servers, so your laptop's NPU makes no difference to them at all. The NPU matters for features that run on your own machine.

Do you need an NPU?

For most buyers in 2026, the honest answer is: you will get one whether you ask for it or not. Every current laptop chip from Qualcomm, Intel and AMD includes an NPU, and Apple has built a Neural Engine into its chips since 2017. The question is not whether to buy an NPU, but whether to pay extra for a faster one.

  • If you buy a Copilot+ PC (40+ TOPS): you get the full Windows AI feature set, and a laptop that will keep getting new on-device features as Microsoft ships them.
  • If your laptop has a smaller NPU (10–16 TOPS): video-call effects and basic AI tools work fine; you simply miss the Copilot+ exclusives.
  • If you mostly use web AI tools: the NPU barely matters. Your money is better spent on memory, storage and screen quality.
  • If you care about privacy: on-device AI keeps your voice, face and documents off the cloud — the NPU's quiet superpower.

The NPU is not a revolution you will feel on day one. It is a slow shift: more and more of the clever work your computer does will happen on your desk instead of in a data centre, using less power and keeping your data closer to home. The TOPS number is just the industry's way of measuring how ready your laptop is for that shift.

Takeaway checklist

  • An NPU is a specialist chip that runs neural-network maths using very little power, so AI features can run all day without draining the battery.
  • TOPS means trillions of operations per second — a peak theoretical figure, usually at INT8 precision, and not directly comparable between vendors.
  • 40 TOPS is Microsoft's Copilot+ line, bundled with 16 GB of RAM and 256 GB of storage; it unlocks Recall, live captions, Studio Effects and Cocreator.
  • Today's chips: Snapdragon X Elite 45 TOPS, X2 Elite up to 85 TOPS, Intel Core Ultra 200V 40–48 TOPS, AMD Ryzen AI up to 60 TOPS.
  • It only helps when apps use it — ChatGPT and Gemini run in the cloud, so your NPU makes no difference to them.
  • Check the fine print: some adverts add CPU, GPU and NPU together; for Copilot+, only the NPU figure counts.
  • Buy for the whole laptop, not the number — memory, storage and screen matter more than the gap between 45 and 60 TOPS.

How GPUs Actually Work · Tensor Cores vs CUDA Cores: What's the Difference?

Hamza

Founder & Editor, TechToday

Hamza is the founder and editor of TechToday. Hamza writes about artificial intelligence, computing, security and the technology shaping everyday life, with a focus on explaining complex ideas in plain, honest language — clear answers for curious readers, without the hype.