How Apple Stumbled Into AI Hardware Success With the Mac

The hottest products at Apple right now are not the iPhone, the iPad or a buzzy new show on the company’s streaming service. They are two of the lowliest members of Apple’s venerable Mac product line: its boxy Mac mini and Mac Studio computers, which come without monitors, keyboards or mice and are designed for professional users.

The products are currently flying off the shelves, because it turns out that the machines are well suited to running agents—AI software that handles multistep tasks, from editing and testing code to automating email inbox organization and summarizing documents. And they’re also a hit with AI developers who want to save on cloud computing bills by training and running their models locally. In the most recent June quarter, Mac sales grew nearly 29% annually to $10.4 billion, faster than any other product segment at the company.

One sign of the unlikely star status for the products occurred on June 23, when Apple held an unusual event on its Cupertino, Calif., campus, known as Apple Park. Normally, Apple is hyperfocused on selling its hardware and services to consumers, but it devoted the June event—called Business at the Park—to enterprise customers. Executives from Walt Disney Co. and Ford Motor were there, as was Anthropic co-founder Jared Kaplan. Tim Cook and John Ternus—Apple’s outgoing and incoming CEOs, respectively—showed up too.

One of Apple’s loudest messages to the attendees was that its hardware is ideal for handling AI chores locally rather than in the data centers of pricey cloud services. And according to one attendee, Apple’s Mac mini was the “darling” of the event.

For AI developers, Mac minis and Mac Studios have become favorites because Apple loads them with more powerful versions of its M-series computer chips and more memory. Unlike a thin MacBook, which can quickly heat up and throttle chip performance, the boxy Macs have cooling systems that can handle long, complex AI tasks. And while Apple’s in-house silicon chips aren’t as powerful overall as Nvidia’s graphics processing units—which dominate the AI chip market—their single shared pool of memory gives them performance advantages for AI workloads.

Some AI labs like OpenAI have bought tens of thousands of Mac minis and Mac Studios for the task of reinforcement learning, a technique where the AI learns through trial and error, said people familiar with OpenAI. OpenAI is using these Macs for training computer-use agents and is desperate to acquire more of them, the people said. Anthropic also rents Mac minis from Amazon’s cloud service, Amazon Web Services, people familiar with Anthropic’s efforts said.

Some startups are even building out Mac-only cloud services, in anticipation of a future where demand for Apple hardware grows so much that some AI developers will want to run on Macs in data centers rather than locally. One such startup is a new Apple hardware–based neocloud company, Mount Thor, founded by Peter Voell, a former OpenAI compute infrastructure employee. Though it is still mostly operating in stealth mode, Mount Thor’s minimal website describes its offerings as “AI execution environments on Apple hardware.”

The Mac’s popularity for local AI has attracted the attention of Nvidia. While most AI training and inference chores are likely to remain in the cloud, demand for local AI has been on the rise, and Nvidia sees Apple as its biggest competitor in local AI, according to a person who has discussed the competitive threat with Nvidia executives.

Nvidia has even come up with new products to get a bigger piece of that market. Late last year, the chipmaker released DGX Spark, an AI desktop computer with a compact squared-off design similar to that of the Mac mini.

Unfortunately for Apple, the exploding popularity of Macs has come at a time when it and other device makers are suffering from severe supply-chain woes. They are wrestling with a historic shortage of memory chips caused by AI data centers’ voracious need for them. The most advanced models of Mac minis and Mac Studios, which have the chips and memory configurations most appealing to AI developers, have been out of stock for months.

“The memory shortage is impacting Apple, and if enterprises are unable to get their Macs, they’ll migrate to different hardware choices,” said David Stout, CEO of webAI, a startup selling AI tools to enterprises that use Apple hardware.

That’s already started to happen. Some businesses have started to look at alternatives to Apple’s hardware because of the supply limitations for Macs over the past year, said Todd Dailey, a former enterprise marketing manager for Apple’s AI products. Companies and AI developers consistently bring up Nvidia’s DGX Spark as an option and is readily available now, said Dailey, who left Apple in April and is now an independent AI consultant.

Because Mac demand has been so hot, Apple earlier this week announced new Mac mini and Mac Studio models with more powerful chips. It was an unusual time for the company to come out with the products: Apple typically does its end-of-the-year Mac refresh in October or November. In its announcement, it bragged about how the new Mac Studios can be clustered together to make a more powerful system for running giant frontier models.

According to Dailey, the soaring demand for Macs among business customers has been accidental rather than something the company had planned for: Apple doesn’t have a dedicated engineering team for business customers, nor does it have any employees focused on developer relations. “The idea that any team at Apple has an actual plan for embracing enterprise AI is a joke,” Dailey said.

Over the years, Apple has tried selling various enterprise server products, but support for these products has been inconsistent and spotty. The company sold Xserve, an Apple-made server product with Intel processing chips, between 2002 and 2011. It also released a server operating system based on Mac but discontinued that product in 2022.

Apple has recently begun building its own servers using Mac chips. However, they’re strictly for internal use with Private Cloud Compute, the company’s server system for processing heavy-duty AI tasks that exceed the processing capabilities of the iPhone or Mac. Some business customers have asked if Apple can sell them access to these servers, but it has turned them down so far. Instead, the company is hoping partners like webAI and Mount Thor can help push its hardware deeper into the business world, people familiar with Apple’s enterprise strategy said.