When buying a computer for AI, four acronyms appear repeatedly: CPU (Central Processing Unit), GPU (Graphics Processing Unit), NPU (Neural Processing Unit) and RAM (Random Access Memory).

A CPU runs the computer and can execute smaller AI models, sometimes slowly. A GPU performs many calculations in parallel and is especially useful for model inference and image or video generation. An NPU is a specialised accelerator built into some newer chips, but its advertised performance may not translate into support for the models you want. RAM holds active data and, on some machines, part of the model.

Here is a practical comparison. Searching a few hundred public documents may run acceptably on a modest computer. Running a larger language model over an extensive local archive may need substantially more memory. Generating a short video is a different workload again: GPU memory and supported software can be decisive.

A model also has to be compatible with the available software. A headline NPU figure does not mean your chosen model runner can use it. GPU memory (VRAM, Video Random Access Memory) and memory bandwidth may matter more than marketing labels.

Before shopping, identify the tasks, the models that can perform them and the memory they need. Ask for expert advice on compatibility rather than buying a computer because it has “AI” on the box. Planetary.Blue can help members compare realistic configurations.