OpenAI Buys Tens of Thousands of Mac Minis, Studios for Reinforcement Learning: Report

OpenAI has bought tens of thousands of Apple Mac mini and Mac Studio computers in recent months to train AI systems that can operate computers, according to The Information. 

The machines are being used for reinforcement-learning workloads, where AI agents practise completing tasks on real computers rather than for the large-scale pre-training of OpenAI’s foundation models.

The purchases involve Macs with large amounts of unified memory, which OpenAI uses to run the models and training workloads needed for computer-use agents. 

The underlying hardware can enable AI models to be trained to navigate software interfaces, interact with applications, and perform tasks on a computer, giving OpenAI a physical environment where agents can repeatedly practise and be evaluated. The scale of OpenAI’s purchases comes as demand for high-memory Macs has increased across the AI industry. 

Mac mini and Mac Studio configurations with large memory capacities have been difficult to obtain, with some models remaining out of stock for months. AI companies are particularly interested in these configurations because Apple’s unified-memory architecture allows the CPU and GPU to access the same large memory pool.

Apple does not disclose sales of individual Mac models, but its broader Mac business generated a record $9.4 billion in revenue in the June quarter, up about 29% year over year.

Apple had already seen AI-driven demand for its compact desktops translate into supply constraints during the June quarter. 

“We’re seeing customers increasingly put those capabilities to work, from using Mac mini as a powerful platform for agentic AI to deploying clusters of Mac Studio systems to run frontier-class models locally,” Cook said on Apple’s July earnings call. He added that Mac faced supply constraints during the quarter driven by “very high levels of demand”.

Recently, the company introduced new Mac mini models with the M6 and M5 Pro chips alongside a refreshed Mac Studio powered by M5 Max and M5 Ultra. 

Other AI companies are pursuing similar approaches. Anthropic has reportedly been renting Mac mini capacity through Amazon Web Services, as AI developers increasingly use Apple silicon for workloads that benefit from its large unified-memory configurations.