OpenAI Reportedly Buys Thousands of Macs to Train Computer-Use AI Agents

OpenAI has reportedly purchased tens of thousands of Mac minis and Mac Studios to support the training of computer-use artificial intelligence agents through reinforcement learning, according to information attributed to hokanews and shared by @coinbureau on X. The reported hardware investment highlights the growing importance of physical computing environments in the development of AI systems designed to interact directly with computers.

The report contrasts OpenAI’s approach with that of Anthropic, which reportedly rents similar Mac hardware through AWS. The two approaches illustrate different methods for providing AI systems with access to computing environments that resemble those used by people and businesses.

OpenAI Reportedly Expands Mac Hardware Infrastructure

According to the information shared on X, OpenAI has acquired tens of thousands of Mac minis and Mac Studios for reinforcement-learning workloads focused on computer-use agents.

Computer-use agents are AI systems designed to interact with computers by performing tasks through interfaces such as applications, websites and operating systems. Instead of producing only text or code, these systems can be trained to carry out actions within digital environments.

Reinforcement learning is a machine-learning approach in which an AI system improves its behavior through feedback based on the outcomes of its actions. For computer-use agents, this type of training can involve allowing models to perform tasks in controlled computing environments and learn from whether those actions achieve markets the intended result.

The reported acquisition of large quantities of Mac hardware would provide OpenAI with access to physical machines on which such systems can be trained and evaluated.

Why Mac Hardware Matters for Computer-Use Agents

Training computer-use agents can require environments that closely replicate the computers people use in everyday work. Physical hardware can provide an environment in which an AI system interacts with operating systems, applications and other software in ways that may differ from conventional model training conducted primarily on large computing clusters.

The reported use of Mac minis and Mac Studios suggests that OpenAI is dedicating substantial computing resources to this area of AI development. The hardware could provide multiple independent environments for training and testing agents simultaneously.

However, the specific number of devices involved was not disclosed beyond the description of tens of thousands, and the reported purpose of the hardware was to support reinforcement learning for computer-use agents.

The information does not establish how the machines are configured, where they are located or how they are distributed across OpenAI’s operations.

Anthropic Takes a Different Hardware Approach

The report also highlighted a contrasting strategy from Anthropic. Rather than purchasing comparable Mac hardware outright, Anthropic reportedly rents similar machines through AWS, according to the information shared on X.

Cloud-based hardware rental allows companies to access computing resources without purchasing and maintaining the physical equipment themselves. It can also provide flexibility in scaling infrastructure depending on research and development requirements.

OpenAI’s reported purchase of Mac minis and Mac Studios represents a different infrastructure model, involving direct ownership of the hardware. The two approaches reflect alternative ways for AI developers to obtain access to specialized computing environments.

The comparison comes as major AI companies continue developing systems capable of carrying out increasingly complex tasks on behalf of users. Computer-use capabilities require models to understand interfaces, interpret visual information and execute sequences of actions while responding to changes within a digital environment.

The reported investment in Mac hardware indicates that physical computing environments are becoming part of the infrastructure financial being used to develop these capabilities. At the same time, Anthropic’s reported use of rented Mac hardware through AWS demonstrates that companies can pursue similar research using cloud-based access rather than direct hardware ownership.

The information was shared by @coinbureau on X, citing the report from hokanews. The reported hardware purchases and rental arrangements provide a snapshot of how AI companies are allocating computing resources as they develop computer-use agents through reinforcement learning.

Writer: Victoria Hale
Technology & Blockchain Writer

Victoria Hale writes about blockchain technology, digital infrastructure, and the intersection of emerging technologies with finance. Her articles explore how new protocols and systems are shaping the evolving digital economy.

She prioritises clarity and accuracy when explaining technical developments to a general audience.