The Mac isn’t selling as fast as Apple can make them because of artificial intelligence needs as some would have you believe, but AI is giving businesses plenty of reasons to buy more of them.
You may have noticed that the lead times on both the Mac mini and the Mac Studio have been a bit much over the last year or so. In fact, right now, if you wanted to buy a fully maxed-out Mac Studio, there’s a 10- to 12-week wait.
There’s also an $18,299 price tag before taxes. Just saying.
There’s some overlap as to why both of those things are happening. A major reason is that we’re seeing unprecedented supply chain constraints as both storage and memory components become harder to reliably and inexpensively source.
But there is another part of the story. And it’s a story worth delving into, because it is, unfortunately, going to affect all of us in one way or another.
So let’s talk about Mac mini, Mac Studio, and artificial intelligence.
Yes, people are using the Mac for AI
Suppliers are seeing an increased demand for the Mac mini and Mac Studio. And yes, AI is definitely a reason behind why that’s happening.
For some, there’s this idea that Mac, especially Apple’s pint-sized powerhouses, is covertly becoming the backbone of artificial intelligence.
Let me be clear here. There’s an entire ocean between “AI is driving Mac demand” and “Mac is becoming foundational AI infrastructure.”
Even Apple is aware of this. In fact, Apple’s entire pitch is that the Mac mini and Mac Studio are perfect for doing smaller tasks, something akin to an AI-based “chore.”
We’ve already known that this has been the case for several months. But when large AI outfits start buying up Apple’s gear in bulk, it gets people talking.
According to The Information, OpenAI has purchased tens of thousands of Mac minis and Mac Studios. And Anthropic leases its Mac minis through Amazon Web Services.
But if these Macs aren’t being used for infrastructure, even when purchased in bulk like this, what are they actually being used for?
Its the little things, sometimes
OpenAI isn’t chaining a bunch of Mac minis together to make a supercomputer, and there are no obvious plans to do so. That doesn’t, however, mean that there’s not an Apple-shaped spot in the pipeline.
One of the best things about a Mac mini, and this goes doubly for a Mac Studio, is that it’s a lot of energy-efficient computing power in a small footprint. This makes them ideal for hyper-specific use cases that require a lot of repetition.
One such use is reinforcement learning, a process in which AI needs to learn through trial and error. As IBM explains, “In reinforcement learning, autonomous agents learn to perform a task by trial and error in the absence of any guidance from a human user. It particularly addresses sequential decision-making problems in uncertain environments…”
There are a lot of reasons to do this on a Mac mini or Mac Studio. Power is part of it, sure, but the other part is that AI is going to need to learn to interface with macOS anyway, and this is an economical way of doing it.
Well, relatively economical.
A fully maxed out Apple Studio currently costs $18,299 before taxes. Don’t worry, it can (and will) get more expensive.
And yes, there are a few instances of companies clustering Mac hardware together. The Information found that Mount Thor is building a Mac-based Neocloud, and EXO Labs software does allow multiple Macs to run models too large for an individual machine.
These are pretty niche applications at the moment. And they’re certainly not indicative of a broader shift in AI infrastructure.
Apple isn’t positioned to displace what Nvidia can offer at scale. Nvidia makes purpose-built hardware for the massive, interconnected computing systems. Apple makes the Mac mini, a desktop-class computer you can put into your back pocket.
But that doesn’t mean that Apple can’t ruffle Nvidia’s feathers elsewhere.
For the Mac, by the Mac
MacStadium surveyed about 300 U.S-based developers in the mid-market and enterprise organizations to find out just what they were using their Mac mini and Mac Studios for. And it is the exact thing that I’ve personally suspected for a long time:
Mac developers want to use a Mac to develop for the Mac. They aren’t using the 16GB Mac mini for big AI loads, and the starting SSD space is too small for anything serious
Historically, we know that the entry-level Mac mini is about 80% of the sales of the model. That was the same through the entire M4 Mac mini run. The entry-level model was just as constrained for most of 2026 as the upgraded configurations or the M4 Pro Mac mini.
Most of these outfits are pretty small, with 80% falling somewhere between 5 and 50 engineers. But they’re running a surprising amount of hardware; 55% of the teams surveyed use 26 or more physical Macs for continuous integration/continuous deployment (CI/CD).
CI/CD is, in layman’s terms at least, the automated process of taking developers’ code, building and testing it, and ultimately preparing it for release. And because these teams develop for Apple platforms, that process requires access to Mac hardware.
Image credit: MacStadium
One pretty critical point here is that even though these engineers are using Macs, most aren’t primarily relying on self-hosted AI. More than a third are using hosted AI services like OpenAI or Claude.
Less than a quarter are primarily self-hosting private models, and another 23% are using a mix of self-hosted and hosted.
But that doesn’t mean Mac isn’t valuable. In fact, it’s probably more valuable.
About 90% of those polled report increased pull requests and commits since they’ve started using AI coding tools. And AI adoption has increased Mac infrastructure costs for 83% of survey respondents.
MacStadium puts it plainly:
“More code means more builds. More builds mean more Macs.”
This practically means that even if the Mac isn’t being used to run the AI itself, AI is still going to drive Mac sales. Especially in those crucial enterprise markets.
Second verse, same as the first
I do want to be incredibly clear here. This isn’t new demand; it’s increased demand.
Most of the teams surveyed had a decent amount of Macs used specifically for CI/CD. The average was 81 physical Macs, despite the typical team employing only around 20 Apple-platform engineers.
Of course, there’s a bit of selection bias baked into these results. MacStadium specifically surveyed infrastructure professionals at organizations whose development teams build software for Apple platforms.
But even as AI increases demand for Mac infrastructure, most companies aren’t responding by simply buying more Macs.
It’s not supply that is the limiting factor at enterprise-level outfits | Image credit: MacStadium
About half of MacStadium’s respondents said they plan to move from self-hosted Macs to managed or cloud solutions over the next year. Compare that with less than a fifth, or 18%, who said their primary plan is to increase Mac capacity.
The distinction between Silicon and Mac
Apple itself is a pretty solid illustration of the difference between what a Mac Studio can offer and what dedicated AI infrastructure looks like. When Apple needs serious infrastructure, it purpose-builds it.
We’ve already known about this. For Private Cloud Compute, Apple just makes its own servers using its own chips to handle heavy-duty AI workloads that exceed what an iPhone or Mac can process locally.
Inside an Apple server – Image Credit: @hsuchingpo/X
And Apple doesn’t appear to be gearing up to offer those servers to anyone outside of the company, either. The Information says that businesses have asked Apple for access, but Apple has turned them down.
Which, at the very least, does create a market for Mount Thor and EXO Labs.
The Information also says that Apple didn’t see this coming, and that there is nobody watching the store. We know for a fact that is absolutely not the case.
Apple founded the “Pro” group, and out of it, immediately, came the iMac Pro, and last Intel Mac Pro tower. That group still exists, and it is why the Mac Studio is what it is.
They also claim that there isn’t anybody inside Apple doing developer relations at all. That will probably come as a surprise to the developer relations team that has about 65 people that we could quickly find working in this capacity, that Mike Wuerthele and I confirmed were on the job.
Not having a team at all is vastly different than a team having multiple responsibilities. What’s true is the latter, not the former.
An Apple-shaped niche
Just because Apple’s boxy little Macs aren’t going to beat Nvidia at its own game doesn’t mean they don’t have their own spot in the AI pipeline. They very clearly do.
Silicon’s unified memory offers some advantages for specific AI workloads. Specifically, cases where both the CPU and GPU need to access the same pool of memory, making it easier to fit a big AI model into memory locally.
I do find it funny, personally, that everyone seems somewhat surprised that Apple has “accidentally” stumbled into hardware success. Apple hasn’t done anything accidentally here.
Apple Silicon has been a thing for a while now. So has the Neural Engine. Apple’s entire modus operandi for at least the last decade has been designing heavy hitters that don’t use a lot of power.
Acting as though Apple, a company that just had its 50th anniversary, did not see the need to make a chip that could do more with less is absurd.
Sure, Apple may not have anticipated that a company like OpenAI would be buying tens of thousands of Macs for reinforcement learning. That would have been one hell of a gamble if it had.
Tens of thousands is somewhere between “barely observable in the data” and “background noise” in total M4 Mac mini sales across its lifetime.
But the company has been in the game long enough to know that smaller computers are going to have to go big or go home. The surprise is in where the devices are being used, not that they’re being used at all.
