Which Mac or mini PC should you buy to run AI at home? (UK prices)

A £1,299 Mac mini will now run Qwen3.8 27B, one of the most popular open AI models around, with your wifi switched off. A £2,829.99 mini PC from a Chinese maker you've probably never heard of will hold a 70B model. And Apple will sell you 128 GB of memory for £400 less than 96 GB.

Last updated 22 October 2026. Written by Annie, a search and digital marketing specialist. Facts and prices checked against the sources listed on this page. See the editorial policy.

Buying before Christmas? Black Friday is 27 November 2026. The Black Friday AI deals page tracks verified UK price drops on every machine in this guide, with the configurations worth waiting for.

Buying hardware for local AI is full of quirks like that. Get it wrong and you've spent £2,499 on a machine that can't load the model you bought it for. I've done the maths and checked the UK prices, so you don't have to.

What matters most when you buy for local AI? Memory

Forget the chip. Memory decides which AI models you can run, and on every machine in this guide you can't add more later.

Here's the rule of thumb for models squeezed to 4-bit, the standard way to run AI at home:

• 16 GB: small models up to 8B, and 14B at a squeeze. Fine for drafting, summaries and chat
• 32 GB: the sweet spot. Mid-size models like Qwen3.8 27B, capable enough for coding help and research
• 64 GB: 70B models, the largest that still fit in one affordable box
• 128 GB and up: the giants, including big mixture-of-experts models

If you remember one thing, make it this: spend on memory before speed.

My three picks: the best Mac and mini PC for local AI

Best value: Mac mini M6 with 32 GB, £1,299. It runs 27B models, it's silent, and it's the cheapest way into proper local AI. Order the 32 GB version. The £899 base model has 16 GB, and that's the upgrade you'll regret skipping.

For 70B models on a budget: GMKtec EVO-X2 with 128 GB, £2,829.99. It's the cheapest 128 GB machine on sale in the UK, about £2,270 less than the cheapest 128 GB Mac. The catch: it ships from China, it's happiest on Linux, and it answers more slowly than a Mac with a Max chip. Rather stay on a Mac? The Mac mini M5 Pro with 64 GB (£2,699) runs 70B models too.

No compromises: Mac Studio M5 Max with 128 GB, £5,099. More than twice the memory speed of the AMD boxes, so answers arrive faster, and everything works out of the box. A painful price, but it's the one to buy if budget isn't the issue.

Every machine I've reviewed

Each of these 12 machines has its own page, with the models it runs, published speeds and where to buy it in the UK. They are sorted by starting price, checked on 17 October 2026.

How much memory does each AI model need?

Find the model size you want, then read across. A 4-bit file is smaller and quicker; 8-bit keeps a little more quality but needs nearly twice the room. For a specific machine or model, the local LLM memory calculator does the sums for you.

The maths, if you want it: a 4-bit file is about 0.6 GB per billion parameters (4.85 bits each, llama.cpp's common Q4\_K\_M format); 8-bit is about 1.06 GB. Allow another 2-4 GB for a long conversation.

The Mac catch nobody mentions: by default macOS lets the graphics side use about 75% of memory, so a 32 GB Mac gives a model roughly 21-24 GB and a 128 GB Mac about 96 GB. Advanced users can raise this with one Terminal command (iogpu.wired_limit_mb), at the risk of starving the rest of the system (source).

Mixture-of-experts models (names with "A3B", "A22B" and so on) must fit whole in memory, but only the active part works on each word, so they run much faster than their total size suggests.

Which Mac should you buy for local AI?

The cheapest way to each amount of memory, from Apple's UK store:

Four Apple quirks worth knowing

• The £2,499 trap. The entry Mac Studio has a fast M5 Max chip but only 36 GB. For £200 less, a Mac mini M5 Pro gives you 48 GB. For AI, memory wins
• 128 GB costs less than 96 GB. 96 GB only comes with the Ultra chip (£5,499). 128 GB comes with the Max (£5,099)
• Upgrades are steep. Apple charges £400 to go from 16 GB to 32 GB, and £2,000 to go from 48 GB to 128 GB on an M5 Max
• Refurbished isn't a shortcut right now. Apple's UK refurbished store had one Mac with 64 GB or more when I checked: a 128 GB MacBook Pro at £8,279

Want a laptop? The MacBook Air tops out at 32 GB. The cheapest 128 GB laptop is the 14-inch MacBook Pro M5 Max at £6,699.

Prices from Apple's UK store for Mac mini, Mac Studio and refurbished Macs, checked 16 October 2026.

Are Chinese AI mini PCs any good? The boxes taking on the Mac

A wave of small desktop PCs from Chinese makers now packs up to 192 GB of memory. They use AMD's Ryzen AI Max chips, which share one pool of memory between processor and graphics, the same trick that makes Macs so good at this.

Not here yet: Chuwi showed a 192 GB UniBox at IFA in September but has no UK price, and AOOSTAR's version is only sold in China.

Before you buy from a maker's UK store

• Check the total at checkout. GMKtec says one of its prices excludes VAT, so confirm VAT and import charges before you pay
• Look for a UK returns address and the warranty length
• Ignore prices in older reviews. Many quote less than you'll pay today

Prices checked 16 October 2026 on each maker's store and Scan.

Mac or mini PC: which is better for local AI?

The mini PC wins on memory per pound. The Mac wins on speed and sheer ease. Here's where each pulls ahead.

Choose the mini PC if you want to load 70B-class models on a budget and don't mind Linux.

Choose the Mac if you already use one, want it to work out of the box, or care about speed. Check the memory before buying: Bitdoze notes that a 36 GB Mac Studio pre-order can't hold a 70B model.

Source for the speed and memory figures: Bitdoze, best mini PC for local LLMs (prices checked 21 September 2026), and the Mac memory limit above.

Which AI models should you run first?

Qwen3.8 27B is the model of the moment: over 16,000 likes and 6.9 million downloads on Hugging Face, and it runs on a 32 GB Mac.

Bonsai 2 also comes in an MLX version built for Apple Silicon. Skip the "uncensored" and "abliterated" re-uploads that crowd the trending list: they come from anonymous uploaders and strip out safety training.

Source: Hugging Face trending models, pulled 16 October 2026. Memory needs are worked out as in the table above.

Still not sure? Answer three questions

Your answers tell you how much memory you need, then which machine gets you there for the least money.

1. What do you want the AI to do?
   • Chat, summarise, draft emails → small models (8-14B) → 16-24 GB
   • Coding help, research, agents that use tools → mid-size models (27-32B) → 32-48 GB
   • The biggest open models, or several models at once → 70B and up → 64-128 GB+
2. Mac or PC?
   • Mac: simplest setup, quiet, holds its resale value
   • PC: more memory for the money with an AMD AI mini PC
3. What's your budget?
   • Under £1,000 · £1,000-£2,000 · £2,000-£3,500 · £3,500+

The verdict

Buy memory first. For most people, the 32 GB Mac mini at £1,299 is the sweet spot: quiet, simple and big enough for the models that matter right now. Want 70B models? The GMKtec EVO-X2 is the bargain, and the 64 GB Mac mini M5 Pro is the easy option.

Frequently asked questions

How much RAM do I need for a local LLM?

16 GB runs small models up to 8B. 32 GB runs mid-size models like Qwen3.8 27B. 64 GB runs 70B models at 4-bit, and 128 GB covers 70B at 8-bit and models up to about 120B. On a Mac, and on the AMD mini PCs in this guide, that is the machine's shared memory; on a PC with a separate graphics card, it is the card's own memory that counts.

Can a Mac mini run a 70B model?

Yes, with 64 GB of memory. The Mac mini M5 Pro with 64 GB costs £2,699 and runs 70B models at 4-bit.

Is a Mac or a PC better for local AI?

A Mac is faster and simpler to set up. An AMD mini PC gives you more memory for the money: the GMKtec EVO-X2 with 128 GB costs £2,829.99, against £5,099 for the cheapest 128 GB Mac.

Is 16 GB enough to run AI locally?

For small models, yes: 8B models run comfortably and 14B at a squeeze. For anything bigger, buy 32 GB.

Can I add more memory later?

No. Memory is built in on Apple Silicon Macs and soldered on the AMD mini PCs in this guide, so buy what you need up front.

Can I run an LLM locally on a Mac?

Yes. Any Apple Silicon Mac (M1 onwards) can run a small model, and because the graphics chip shares the Mac's memory, a well-specified Mac runs big ones too. Install Ollama or LM Studio, download a model that fits your memory and chat with it offline. The step-by-step setup is in run a local LLM on your Mac.

Which MacBook is best for running LLMs?

The 14-inch MacBook Pro with an M5 Pro or M5 Max chip. A MacBook Air tops out at 32 GB, which is enough for 27B models, but the MacBook Pro goes up to 128 GB, and the cheapest 128 GB laptop is the 14-inch MacBook Pro M5 Max at £6,699. Buy the most memory you can afford. The options and UK prices are on the MacBook Pro M5 Pro and M5 Max page.

What's the best LLM I can run locally?

The biggest one that fits your memory with room to spare. On 32 GB, Qwen3.8 27B is the model of the moment. On 64 GB you can run Llama 3.3 70B at 4-bit. On 128 GB you reach 70B at 8-bit and models up to about 120B. The models table above lists the ones to try first, with the memory each needs.

What is the best computer for running local LLMs?

For most people, the Mac mini M6 with 32 GB at £1,299: quiet, simple and big enough for 27B models. For 70B models on a budget, the GMKtec EVO-X2 with 128 GB at £2,829.99. If money is no object, the Mac Studio M5 Max with 128 GB at £5,099. Prices were checked in October 2026.

Why is everyone buying Mac minis for AI?

Because the Mac mini is the cheapest way to get a lot of memory that the graphics chip can use. Apple Silicon shares one pool of memory between processor and graphics, so a £1,299 Mac mini with 32 GB holds a 27B model and the 64 GB Mac mini M5 Pro holds a 70B model, in a silent box that fits in one hand. Getting 64 GB of graphics memory into a PC means two high-end graphics cards, a big case and a loud fan.

This week: Apple's own M5 Ultra against M3 Ultra figures, Meta's two-Mac-Studio demonstration and a 69 per cent jump in RTX 5090 prices are covered in local AI this week (13 to 21 October 2026).

Not sure private AI is worth the hardware at all? The private AI guide sets out the three routes, the costs and the limits before you buy.

Read next

• Is it safe to run Chinese AI models on your Mac?
• Run a local LLM on your Mac
• Ollama vs LM Studio vs Jan
• Self-hosted LLMs: what they cost and when they're worth it
• Hermes Agent: the self-improving AI agent, explained