OpenAI has purchased tens of thousands of Mac minis and Mac Studios to train computer-use agents through reinforcement learning, according to reporting by The Information.
The scale of the acquisition shows a shift in AI training infrastructure away from sole reliance on data center GPU clusters. Mac devices, which run on Apple's in-house silicon, offer a different computational pathway for training models designed to interact with graphical user interfaces and desktop software. The purchase represents one of the largest known bulk acquisitions of consumer and prosumer hardware by an AI lab for model training.
Computer-use agents are large language models trained to control computers by observing screens, parsing visual elements, and executing actions through keyboard and mouse inputs. Training such models requires hardware that can run actual operating systems and applications at scale. Traditional GPU-heavy training pipelines optimize for mathematical computation. OpenAI is prioritizing real-world software interaction as a core capability.
Apple's M-series chips, which power Mac minis and Mac Studios, have gained adoption among AI developers for local model inference and fine-tuning. Nvidia's dominance in training infrastructure has pushed alternative suppliers to seek differentiation. Apple, which has not released its own large-scale AI training chips, benefits from OpenAI's validation of Mac hardware for this use case.
The exact cost of the acquisition has not been disclosed. Mac minis start at under $600 for entry configurations, while Mac Studios range from $1,999 to $3,999 depending on chip and memory. If the purchase involved 50,000 units at mid-range pricing, the total expenditure could exceed $50 million.
OpenAI has not publicly confirmed the acquisition or detailed its reasoning. The company has previously disclosed partnerships with Microsoft and infrastructure contracts with other cloud providers, but rarely announces specific hardware purchases of this magnitude. The timing coincides with industry-wide competition to build frontier AI models with specialized capabilities beyond text generation.
The purchase size relative to total Mac unit shipments remains modest; Apple sold roughly 24 million Macs globally in 2025. The concentration of devices for a single use case in a single buyer's hands represents a notable concentration of purchasing power in the consumer hardware market.
