Local AIAI mini PC review
Mac mini (M4) review: the easiest way to run your own AI
Apple's small desktop turns out to be one of the best-value machines for running AI at home — if you buy the memory right at checkout and accept that you can never change it later.
Is the Apple Mac mini (M4) worth buying?
Yes — with one condition. Buy the Mac mini with 24 GB or 32 GB of memory, never the 16 GB base model, because you choose it once and can never upgrade. So configured it's around $800–1,000, runs mid-size AI models well, stays silent, and costs under $10 a year to leave on. The largest models still favour a 128 GB PC.
The least fussy way into local AI. Fast where it counts, absurdly efficient, and quiet. The catch is Apple's pricing on the two things you need more of: memory and storage.
Where it's good
- Apple's shared memory design is exactly what AI models want
- Silent, and about 4 watts at idle — the most efficient option here
- The software situation for AI on Macs is now genuinely good
- Doubles as a perfectly nice normal computer
Where it bites
- You choose the memory once, at checkout, forever — no upgrades ever
- The 16 GB base model is the wrong buy for AI; 24 GB or 32 GB costs more
- 256 GB of storage fills up fast and Apple's upgrade is overpriced
- Very large models still favour a 128 GB PC over any Mac mini
Some links below are affiliate links. How this works.
I did not expect to be recommending a Mac for this. But the Mac mini with Apple’s M4 chip has turned into one of the best-value ways to run AI at home, and the reason is the way it handles memory.
The one thing that makes it good at this
In most computers the processor has its own memory and the graphics chip has its own, and they copy data back and forth. Apple puts both on one shared pool of fast memory — think a studio apartment instead of two rooms down a hallway. AI models love this, because the model has to sit in memory that the graphics chip can reach, and on a Mac that’s just… all of it.
The practical result: a Mac mini with 32 GB can comfortably run models that would need an $800 graphics card in a PC. And it does it while drawing about as much power as a lightbulb.
What it’s like to live with
Silent. Genuinely — there’s a fan, I’ve never heard it. It sits on the corner of my desk and draws about 4 watts doing nothing, maybe 30 under a heavy model. Over a year of running around the clock that’s under ten dollars of electricity.
The software caught up in the last year. The same tools people use on a PC — Ollama, LM Studio — run natively on the Mac now, and setup is the same twenty minutes: install the app, download a model, start chatting.
The catch, and it’s a real one
You pick your memory when you buy it and you can never change it. Not “it’s hard” — it’s soldered to the board. So the decision at checkout is the whole decision.
The base Mac mini has 16 GB. For normal computing that’s fine. For AI it’s tight — you’ll be limited to smaller models. Moving to 24 GB is $200 more, 32 GB is $400 more, and for this purpose that money is well spent. Buy the memory you want for the next five years, because there’s no second chance.
Storage is the other soft spot. 256 GB disappears once you’ve downloaded a few models, each of which can be 5–30 GB. Apple wants a lot to give you more internally. Don’t pay it — a $100 external SSD holds your model library fine.
Who should buy it — and who shouldn’t
Buy it if you want the simplest, quietest, cheapest-to-run way to have your own AI, and you’re willing to spend around $800–1,000 to get a version with enough memory. This is the one I’d hand to someone who just wants it to work.
Buy something else if you want to run the very largest models — a 128 GB AMD machine will hold things the Mac mini can’t — or you’re on a tight budget, where the Beelink SER8 gets you most of the way for $520.
The specs, in plain English
- Chip
- Apple M4 (10-core CPU, 10-core GPU)
- Memory
- 16 GB unified (base); 24 GB +$200; 32 GB +$400Apple's memory is shared between the processor and graphics in one fast pool, which is exactly what AI models want. The catch: you pick the size at checkout and can never change it. For local AI, get 24 GB or 32 GB, not the 16 GB base.
- Storage
- 256 GB SSD (base); upgrades are expensive256 GB fills up fast once you download a few models. Budget for an external SSD rather than Apple's upgrade price.
- Networking
- Gigabit (10GbE +$100), Wi-Fi 6E
- Idle power
- ~4 W
- Load power
- ~30 WOne of the most efficient machines you can run a model on — about $8 a year left idling.
The bottom line
If you want to run your own AI with the least possible hassle, buy a Mac mini with 24 GB or 32 GB of memory and a cheap external SSD. It'll cost around $800–1,000 all in, it'll be silent, and it'll sip power. Just don't buy the base model and expect it to keep up.