AI giants descend on the Mac mini - OpenAI buys tens of thousands, Anthropic rents

The biggest AI companies have found a new favourite. It is not a graphics card costing tens of thousands of dollars - it is the Mac mini.

According to The Information, OpenAI has bought tens of thousands of Mac minis and Mac Studios in recent months - no monitors, no keyboards, straight into the server room. Anthropic uses the same machines but does not buy them: it rents them through AWS, drawing on its close ties with Amazon (the scale of that rental has not been disclosed). Same idea, entirely different attitude to spending money.

The Macs are used to train agents that operate a computer the way a person does: the model runs inside a real system, looks at the screen, clicks, types, gets feedback and tries again - millions of times over. This is reinforcement learning, the method behind the current wave of AI agents.

Why a Mac in particular? Because of the architecture. In Apple Silicon the processor and the graphics unit share a single pool of memory, and desktop Macs have active cooling, so they survive weeks of work under load. They do not replace the big Nvidia clusters - they fill a niche where memory, access to the operating system and isolated environments matter more than raw GPU power.

The consequences are visible in shops: Supplies of Mac minis and Studios started running short, so Apple announced new models back in late August instead of its usual autumn slot - with shipping only in the third week of September. For the first time in the history of the line, the company is aiming them at business customers, boasting about linking several Mac Studios into a single AI machine. Mac revenue grew 29% year on year, while some customers, put off by the shortages, are already eyeing Nvidia's DGX Spark.

Apple spent years being told it had slept through the AI race. It got in through the back door - with hardware nobody was buying "for AI".

Reinforcement learning - a method of training AI through trial and error: the model takes an action, receives a reward or a penalty, and after millions of repetitions works out for itself which strategies work.