OpenAI has purchased tens of thousands of Mac minis and Mac Studios in recent months to train computer-use agents, The Information reported, an unusual compute strategy that has helped spark an AI-driven demand surge Apple itself did not see coming.
The bulk purchases, made for reinforcement learning work, have contributed to shortages of Mac mini and Mac Studio configurations that have left some models out of stock for months, according to reporting covered by MacRumors and 24/7 Wall St. For more context on this story, see our ongoing latest AI developments.
Why AI Labs Want Desks Instead of GPU Racks
The logic inverts conventional AI infrastructure wisdom. Training a frontier model demands enormous clusters of interconnected GPUs — a market Nvidia dominates. But training computer-use agents, which are placed inside a desktop, told to complete a task, and scored on the result, has a different profile: it favors breadth across thousands of independent machines rather than raw horsepower concentrated in one supercomputer.
That is where Apple silicon's unified-memory architecture becomes useful. A Mac keeps the CPU, GPU and memory working from the same pool rather than relying on a discrete graphics card and separate system memory, an arrangement well suited to running thousands of parallel desktop sessions where each machine operates independently. It is also, commercially speaking, a cheaper way to scale agent training than renting scarce and expensive GPU hours for workloads that GPUs are not uniquely good at.
OpenAI is not alone in the approach. Anthropic has reportedly rented Apple silicon capacity through Amazon's AWS for similar workloads, according to 24/7 Wall St.
Apple Was Not Ready for Enterprise AI Buyers
The demand caught Apple flat-footed. According to The Information's reporting, the company lacked a dedicated engineering team for business customers and staff focused on developer relations, and had no enterprise AI strategy when organizations began approaching it to buy Macs by the thousands as compute nodes.
Some enterprise customers asked for access to Apple's Private Cloud Compute infrastructure — and were turned down, MacRumors reported. Apple is instead leaning on partners such as WebAI and Mount Thor, which provide AI tooling and execution environments built on Apple hardware.
The company has nonetheless been courting the segment. In June, Apple held a "Business at the Park" event featuring executives from Ford, Disney and Anthropic, where the Mac mini was reportedly the "darling" of the show. Apple has also promoted the ability to link multiple Mac Studios into a single, more capable system for running large frontier AI models — a feature aimed at business and developer customers rather than everyday consumers.
An Unusually Early Refresh — and a Strained Supply Chain
The surge appears to have bent Apple's product calendar. Apple normally releases new Mac models closer to October or November, but last week it refreshed the Mac mini with its M6 chip and the Mac Studio with M5 Max and M5 Ultra processors — an unusually early launch that The Information attributes to the AI-driven boom in Mac Studio and Mac mini sales.
Those new machines are pointed directly at this audience. As AI Buzz Wire covered at launch, the M6 Mac mini and M5 Ultra Mac Studio pair Apple's first 2-nanometer processor with its first quad-die architecture and up to 512GB of unified memory — specifications aimed at running large AI models locally and clustering machines together.
Supply has not kept pace. The demand spike has coincided with a global memory shortage, leaving many higher-end configurations out of stock for months. Some enterprise customers have reportedly turned to alternatives such as Nvidia's DGX Spark, a compact AI desktop with a similar form factor, when Apple's high-end configurations were unavailable.
What It Means for Apple's Numbers — and Nvidia's Moat
There is real money in the trend, though attribution requires care. Apple generated roughly 10.4 billion dollars in Mac revenue in its fiscal third quarter, up about 29 percent year over year, but the company does not disclose how much came from Mac minis and Studios, so pinning the growth directly on AI labs would be premature, as 24/7 Wall St. noted.
The more strategic point is what the trend says about the next phase of the AI buildout. As inference and agent training spread beyond giant pretraining clusters, AI labs are buying compute wherever the economics make sense — and Apple has stumbled into selling silicon to the very companies building the models, without spending a dollar on AI data centers of its own.
That does not make Apple a new Nvidia. Macs are complements to GPU clusters for agentic workloads, not substitutes for the concentrated horsepower that massive model pretraining requires. But with computer-use agents among the fastest-moving frontiers in AI, the idea that a meaningful share of agent training now runs on desk-side machines — rather than data center racks — is reshaping assumptions about what AI infrastructure looks like.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
