AI research labs and emerging AI cloud providers are bulk-purchasing Macs, which have now become Apple's fastest-growing business segment, with sales surging thanks to the wave of AI agent technology. The hottest product from Apple right now isn't the iPhone, iPad, or a flashy new streaming show — it's two unassuming models in the classic Mac product line: the boxy Mac mini and Mac Studio.
These two devices, sold without displays, keyboards, or mice and aimed at professional users, are currently in short supply because they're ideally suited for running AI agents. These AI software programs handle multi-step tasks such as code editing and testing, automatically sorting email inboxes, and summarizing documents. Meanwhile, a large number of AI developers favor these machines for training and running models locally, reducing their dependence on costly cloud services.
In the June quarter just concluded, Mac revenue grew nearly 29% year-over-year to $10.4 billion, outpacing every other Apple hardware line. A telling milestone in this unexpected success story occurred on June 23rd, when Apple hosted a special event called the "Campus Business Conference" at its Cupertino headquarters. The gathering, distinct from Apple's typical consumer-focused launches, was tailored to enterprise clients, drawing executives from Disney and Ford, as well as Jared Kaplan, co-founder of Anthropic, with both outgoing CEO Tim Cook and incoming CEO John Ternus in attendance.
Apple's core message to attendees was straightforward: its hardware is perfectly equipped to handle AI tasks locally, minimizing reliance on expensive cloud data centers. According to one attendee, the Mac mini was the undisputed star of the entire event.
AI developers prefer the Mac mini and Mac Studio because they feature higher-performance M-series chips and large memory capacities. Thin-and-light MacBooks heat up quickly during extended AI workloads, causing the chips to throttle; however, these boxy devices offer superior heat dissipation and can sustain prolonged, complex AI computations without performance dips. Although Apple's custom silicon isn't as powerful overall as the GPUs from NVIDIA that dominate the AI chip market, its unified memory architecture gives it a unique performance advantage in AI workloads.
Sources familiar with the matter reveal that AI laboratories, including OpenAI, have already purchased tens of thousands of Mac minis and Mac Studios for reinforcement learning — a technique where AI iteratively learns through trial and error. OpenAI is using these machines to train computer-operating agents and actively seeking to acquire even more units. Meanwhile, Anthropic reportedly rents Mac mini devices through Amazon's AWS for its research. Several startups are now building Mac-only cloud services, anticipating a surge in demand for Apple hardware in data centers rather than strictly local use.
Mount Thor, a new AI cloud (neocloud) startup founded by Peter Voell, a former OpenAI compute infrastructure engineer, exemplifies this trend. The company, still largely in stealth mode with a minimalist website, describes its business as "AI computing environments powered by Apple hardware." Apple is also promoting EXO Labs' open-source software that can cluster multiple Macs together to run trillion-parameter models locally.
The Mac's ascendancy in local AI has caught NVIDIA's attention. Insiders indicate that while the vast majority of AI training and inference tasks will remain in the cloud, the rising demand for on-device AI has led NVIDIA to view Apple as its biggest competitor in this arena. The chip giant introduced the DGX Spark late last year — a compact, square AI desktop machine with remarkably similar design cues to the Mac mini — specifically to capture this market.
Yet Apple faces a dilemma: while Mac demand climbs steeply, the company, along with other hardware makers, is mired in a severe supply chain crisis. AI data centers are devouring memory chips at unprecedented rates, causing a historic shortage. High-end versions of the Mac mini and Mac Studio favored by AI developers have been out of stock for months. As David Stout, CEO of webAI, a startup selling AI tools for Apple hardware to enterprises, notes, "The memory shortage is hitting Apple. If companies can't get Macs, they'll turn to other hardware options."
That shift is already emerging. Todd Daly, a former enterprise marketing manager for Apple AI products who left in April to become an independent consultant, observes that over the past year, supply constraints have driven many enterprises to evaluate alternatives — frequently listing NVIDIA's DGX Spark, which remains readily available, as a substitute.
In response to the overwhelming demand, Apple released updated versions of the Mac mini and Mac Studio with more powerful chips earlier this week — an unusual timing given that Apple typically refreshes the Mac line in October or November. The launch highlighted that the new Mac Studio can be clustered to form more powerful systems capable of running cutting-edge, massive models.
Daly suggests that this robust enterprise demand is a fortunate accident rather than deliberate strategy; Apple lacks a dedicated enterprise-focused engineering team or developer relations roles. "The idea that there's an internal team specifically planning to embrace enterprise AI is pure fantasy," he says. Apple has historically dabbled in enterprise server products with inconsistent support, such as selling Intel-based Xserve servers from 2002 to 2011 and offering a Mac-based server operating system until its discontinuation in 2022.
More recently, Apple has developed proprietary server hardware based on Mac chips, but these units are designated solely for internal "Private Cloud Compute," handling heavy AI workloads that iPhone and Mac cannot manage. Enterprise customers seeking to rent these servers have been uniformly denied. According to those familiar with Apple's enterprise strategy, the company hopes partners like webAI and Mount Thor will help its hardware gain further traction in the corporate market.
Comments