OpenAI Buys Tens of Thousands of Macs to Train AI Agents

Apple typically refreshes the Mac in the autumn. This week's earlier-than-usual bump to the Mac mini and Mac Studio came about because The Information reported OpenAI has bought tens of thousands of the compact desktops over recent months to run "reinforcement learning" and to train "computer-use agents," the class of models that operate a machine the way a person would.

The machines ship without monitor, keyboard or mouse, and they are now among Apple's hottest products.

The reason they fit is architectural. Apple silicon's unified memory lets the CPU, GPU and Neural Engine draw from a shared memory pool, a layout well suited to agent workloads that are memory-bound and parallelism-light compared with transformer pre-training. Training a fleet of agents, per the report, can require thousands of independent machines rather than one enormous interconnected supercomputer, which is roughly the shape a rack of Mac minis takes.

Apple's own pitch has caught up. This week the company promoted the ability to link multiple Mac Studios together into a single, more capable system for running large frontier AI models, a feature aimed at business and developer customers rather than a family buying a first computer.

The problem: Apple was not built to serve this customer. MacRumors reports higher-RAM Mac Studio configurations are quoting wait times that have stretched from around two weeks to nearly two months, and Apple reportedly lacks a dedicated engineering team for business customers, with staff focused on developer relations. Two of the AI researchers we track shared the story on the day it went live.