AI development companies, including OpenAI and Anthropic, are reportedly deploying large numbers of Mac minis for reinforcement learning purposes.

According to a report by The Information, OpenAI has secured vast quantities of Mac minis and Mac Studios, in some cases buying up existing inventory, while Anthropic is renting large numbers of Mac minis via Amazon's cloud services. These machines are said to be utilized for tasks such as training AI agents, organizing emails, summarizing documents, and testing code.

These machines are used in the reinforcement learning phase, where a model repeatedly attempts an action and corrects it based on feedback. Because Apple Silicon chips feature a design where the CPU, GPU, and Neural Engine share a single memory pool, they are considered efficient for reinforcement learning, which involves stacking countless short inference.

Furthermore, the laboratories are enabling the execution of massive models by clustering these machines to aggregate their memory.

Source: