A processor for AI workloads
An NPU (neural processing unit) is a processor built to run neural network workloads efficiently. Found in phones, tablets, and computers, it handles speech recognition, image processing, background blur, and summarization on the device.
Vendors use different names — Apple calls its version the Neural Engine. On Windows, a high-performance NPU is one of the requirements for a Copilot+ PC.
How it differs from CPU and GPU
The CPU is the general-purpose processor that runs the OS and applications. The GPU excels at parallel computation and is used for graphics as well as training and running large AI models.
The NPU specializes in the calculations AI models repeat constantly, and it does them continuously at low power. Work is divided across CPU, GPU, and NPU according to what each does best rather than being handed entirely to one.
Relationship to on-device AI
Cloud AI sends input over the internet to be processed on large machines. On-device AI uses the NPU and other local hardware, so it works without connectivity, responds quickly, and keeps data from leaving the device.
Apple Intelligence and Windows AI features include both local and cloud processing. Having an NPU does not mean every AI feature runs entirely on the device.
What TOPS means
NPU performance is often quoted in TOPS, or trillions of operations per second. It is a rough indicator, but a higher number does not speed up every AI workload proportionally.
The model in use, software support, memory, and power settings all affect real performance. When choosing a computer, check whether the features you want are supported on that NPU rather than comparing TOPS alone.
Worth knowing
An NPU does not make answers more accurate; it makes the computation more efficient. On a machine with a powerful NPU, results still depend on the model and the application.
The label AI PC on its own does not tell you which features run locally. Check the OS, applications, models, and the NPU performance they require.