One question sorts
every AI conversation.
AI is in more of our Mac conversations every week. The confidence gap at the table comes down to one distinction. Get it straight, and the recommendation almost makes itself.
When a customer says "AI",
what do they usually mean?
AI is moving fast, and it is not one thing. About half of people are more concerned than excited about it.
Most mean the chatbot. That is online AI, and it is the easy one to answer. The gap is real, but it closes fast once we know which of the three we are talking about.
The customer is not vague on purpose. The industry used one word for ten different things. Our job is to find out which one they mean.
The line we hear most.
"Almost certainly your current one. Let me ask you one thing."
Most customers use online AI, which runs fine on any Mac. We only spec up when the AI has to run on the Mac itself. That is the next screen.
The customer is not
ignorant. They are unsure.
What share of people say they are more concerned than excited about AI?
About half. And that wariness is growing, especially among younger people.
This is the person walking into our store. They are not sure what AI means for them, and they half expect it to be a problem. Calm beats clever.
One word. Three meanings.
When a customer says "AI", they mean one of three things: a chatbot they talk to, a feature that helps them, or a vague sense they should have it.
Apple did not lead
on chatbots.
That is true, and the team feels it every time a customer asks. Apple chose a different thing.
General chatbot
Knows a lot about everything. Runs on someone else's servers. Keeps your conversation.
Personal intelligence
Knows you and your context. Runs on your device. Keeps your data yours.
What does your customer
actually value?
Most customers, when it is put this plainly, choose private. That is Apple's whole bet.
Apple Intelligence runs on the device by default. It only sends work to Apple's servers, called Private Cloud Compute, when a request needs a bigger model, and that processing is designed to use the data once and keep nothing.
The privacy objection.
"With Apple Intelligence, most of it runs on your device, so your data never leaves. When it needs more power, Apple designed the server side to use your request once and keep nothing. Not even Apple can read it."
Say "designed to", not "guaranteed". It is honest and it is exactly what Apple's own engineers state.
Where does the AI live?
Online AI
The provider runs the model. The Mac just runs the app or a browser tab. Every Mac in the store does this well.
Local AI
The model is on the Mac, running on its own chip and unified memory. This is the only time configuration really matters.
Say it together"Would the AI run on the Mac itself,
or through a service like ChatGPT?"
Two specs carry the
local conversation.
Unified memory
How big the model can be. The model, your documents, your apps and the system all share it. You cannot add more later, so it comes first.
Memory bandwidth
How fast the model replies. If an Air answers in a second, an M5 Pro takes about half that, an M5 Max a quarter, and M3 Ultra less again.
The Macs, for local AI.
| Mac | For | Local AI note |
|---|---|---|
| MacBook Neo | First Mac, everyday, online AI | 8GB is not for meaningful local work |
| MacBook Air | Thin all-rounder, occasional local | 24 or 32GB when local AI is a named need |
| MacBook Pro | Sustained work, local AI on the move | M5, M5 Pro, or M5 Max by size of local work |
| Mac mini | Best-value desk setup | M4 Pro with 48GB for regular local AI |
| Mac Studio | Stationary specialist | M4 Max or M3 Ultra for the largest private models |
The quick close.
If you remember one thing.
"Would the AI run on the Mac itself, or through a service like ChatGPT?"
Where the facts come from.
- Apple, Private Cloud Compute security blog (security.apple.com)
- Apple, Apple Intelligence and MacBook Pro product pages (apple.com)
- Apple, March 2026 product training reference (this project)
- Pew Research Center, 2026 findings on AI concern and AI-written web content