What is an NPU, and why do new laptops advertise AI speed?

Look at the spec sheet of a recent laptop and you will often find a line you would not have seen a few years ago: NPU, followed by a number of TOPS. Manufacturers highlight it as proof that the machine is ready for artificial intelligence. Here is what an NPU actually is, what it does for you and how to judge the number.

The short answer

An NPU, or neural processing unit, is a part of the processor designed to run the kind of math that AI models rely on. That math is mostly large amounts of simple multiplication and addition, repeated many times. An NPU does it with very little electricity, which matters in laptops and phones where battery life is precious.

How it differs from the CPU and the GPU

Your computer already has two other processing parts that can run AI work:

• The CPU is flexible and good at many different tasks, but it is not efficient at huge amounts of repeated math.
• The GPU is very good at that kind of math and is what most AI training and heavy local AI use today. It also uses a lot of power.
• The NPU is built for steady AI work at low power. It is not as fast as a high-end graphics card, but it can run all day without draining the battery.

You can think of the NPU as the efficient specialist for everyday AI jobs.

What it is used for

The most common uses today are practical and small:

• Blurring your background and improving your voice and image on video calls
• Live captions and translation
• Searching your own files and photos by describing them
• Running smaller AI models directly on your device, without sending data to the internet
• Photo cleanup and some editing features

Because these tasks run in the background, the benefit is mostly longer battery life and smoother performance rather than a dramatic new capability.

Reading the TOPS number

TOPS stands for trillions of operations per second. It is a measure of how much AI math the NPU can perform at its peak. Microsoft, for example, set a minimum of 40 TOPS for laptops that carry its Copilot+ PC label. A higher number means more potential speed, but be careful with comparisons:

• TOPS depends on the type of number used in the calculation, so figures from different makers are not always comparable.
• Peak numbers are not real-world numbers. Memory speed and software support limit what you will actually see.
• A big number is useless if the programs you use do not take advantage of the NPU.

Do you need one?

For most people, an NPU is a nice extra and not a reason to buy a laptop by itself. It makes sense to care if you attend many video calls, travel and want battery life, or want AI features to work offline and keep your data on the device. If you mainly browse, write documents and watch videos, you will not miss it.

If you are buying a new machine, look at the things that matter every day first: battery life, screen quality, memory and storage. Then treat the NPU as a bonus that will gain value as more software supports it.

What to watch next

Software is catching up with the hardware. Over the next few years more programs will run parts of their AI features locally, and the NPU will matter more. For now, check which features of the laptop you are considering actually use the NPU, and ignore marketing that simply repeats a big number.