An NPU, or neural processing unit, is a part of a device’s main chip set aside for the maths that AI models need. It is much smaller and less powerful than a data-centre GPU, but it uses far less energy, so it can run AI features on a battery without the phone or laptop running hot.
You meet NPUs in phone and laptop launches, usually with a speed figure in TOPS. Microsoft’s Copilot+ PCs, launched in May 2024, were sold on chips capable of “40+ TOPS”. NPUs let features such as live captions or background blur on video calls run on the device rather than in the cloud.