Plateclear
Menu

Library › Buyer guides

Choosing a computer for AI work

Free Updated 2026-10-19

As of 2026-10-19. Hardware, models and prices change quickly: check the date at the top, and verify every spec and price on the maker's site before buying. We don't earn anything from any purchase.

If you only use AI in the browser, an app, or an assistant that calls a provider

Any recent laptop works, including for the Discord and Telegram assistant kits (Claude Code itself asks for about 4 GB of RAM and macOS 13 or a recent Linux). Spend on a good screen and keyboard, not on a graphics card.

If you want to run models on your own machine

  • Memory matters most. The model has to fit in memory to run at a usable speed. A rough guide from Ollama's library: a 3-billion-parameter model is a 2 GB download, an 8B model about 5 GB, a 14B model about 9 GB, a 32B model about 20 GB — and you need some room on top for the conversation. See Local model starter (members) for a full table.
  • Apple Silicon Macs share one pool of memory between processor and graphics, so the whole pool (minus what the system needs) is available to a model. Current Apple specs pages list, for example, Mac mini configurations from 16 GB (up to 32 GB) and an M5 Pro configuration from 24 GB (up to 64 GB), and MacBook Air (M5) from 16 GB (up to 32 GB). [check: configurations and prices on apple.com]
  • Windows and Linux PCs: the graphics card's own memory (VRAM) is what counts. NVIDIA's current cards range from 8 GB to 32 GB — see Graphics cards for local AI (members).

Memory can't be upgraded later on Macs, so buy the amount you'll need.

If you want a server that is always on

A small cloud server or a spare mini PC is enough for an assistant that calls an AI provider.

Sources