One AI Model on My Mac Mini Takes Up 19GB. That's Bigger Than My First Laptop's Hard Drive.

There is a single file on my Mac Mini M4 that takes up 19GB. It is one AI model, deepseek-r1:32b, running locally through Ollama.

For reference, my first laptop had a 10GB hard drive. The entire computer. Operating system, programs, every file I owned. This one model is almost twice that.

What you get for 19GB

It runs entirely on the machine. No internet connection, no account, no subscription, no data leaving the desk. I can unplug the ethernet and it still answers.

That last part is the whole pitch. Every prompt you type into a cloud AI service is a prompt you sent to someone else's computer. Most of the time that is fine. Sometimes it is not, and having a local option means you get to decide which is which instead of defaulting.

What it costs

Nothing in dollars. Ollama is free, the models are free to download.

It costs you disk and patience. I have seven models pulled right now and they add up fast. And a 32 billion parameter model on a Mac Mini is not instant. It thinks. If you are used to cloud speed, the first few responses feel slow.

When local actually wins

Use local when: the data is sensitive, you are offline, you are running something in a loop where cloud costs would stack up, or you just want to know it still works if a company changes its pricing next quarter.

Use cloud when: you need the strongest possible answer on a hard problem, or you need it fast.

I use both. That is not a cop out, it is the actual answer. The models I run locally handle the routine, private, high-volume stuff. The hard reasoning goes to a frontier model.

The setup

Ollama installs in about a minute on macOS through Homebrew. Pulling a model is one command. There is no configuration step, no API key, no signup.

That is the part that surprises people. The barrier to running real AI on your own hardware is not technical skill anymore. It is knowing that you can.

The 19GB is just the price of admission, and disk is cheap.