OpenAI buys tens of thousands of Mac mini and Mac Studio units to train agents
The Information reports OpenAI purchased tens of thousands of Macs in recent months for reinforcement learning and computer-use agent training.
On August 31, The Information reported that OpenAI purchased tens of thousands of Mac mini and Mac Studio units in recent months to train reinforcement-learning and computer-use agents.
Apple M-series chips' memory capacity and thermal headroom made them the preferred hardware for OpenAI's agent training. The orders helped drive Mac segment quarterly revenue up nearly 29% year-over-year, with some models out of stock for months.
The report notes that Anthropic is also renting large numbers of Mac minis through AWS for agent-related tasks. The appetite for Mac mini in agent training reflects intensifying cross-domain competition for compute resources, extending the hardware race beyond GPUs into consumer-grade chips.
The move parallels OpenAI's same-day announcement of expanded internal testing for the new Astra model, which emphasizes multi-agent orchestration, persistent reasoning, and real-time self-correction, all of which significantly raise the demand for RL samples and compute devices.