AI Research Labs are purchasing tens of thousands of Mac Minis to train their artificial agents.

According to The Information, OpenAI, Anthropic, and other AI companies are buying tens of thousands of Mac minis and Mac Studios to train computer use agents (CUA), which are AIs that can interact with software, clicking buttons, and doing complex tasks automatically. The high demand for these computers has caused the most powerful Mac Studio models to sell out for several months because of a lack of the memory chips used by Apple in its silicon. Mac sales grew almost 29% during the June quarter to $10.4 billion, driven in part by demand from AI companies.

Why are AI companies interested in the Mac? Because Apple’s silicon uses a unified memory architecture, which enables efficient processing of large model weights during extended inferences without the heat and power costs of discrete GPUs. Rather than purchasing computers directly, Anthropic is using Macs via AWS. The trend is creating new infrastructure, such as Exo, an open-source tool that lets you cluster multiple Macs together to run bigger models, and Mount Thor, a cloud service built by Peter Voell, formerly of OpenAI’s computing infrastructure team, that runs on Apple hardware and targets this market. Nvidia’s DGX Spark is billed as a small but powerful solution that relies on GPUs.