OpenAI is buying up Mac computers to train neural networks. The tech giant has purchased tens of thousands of Apple devices, according to authoritative outlet The Information. The purchases were made over the past several months. The company is using the devices for specific agentic tasks. The computers help train AI agents in important skills. The neural networks are learning to independently interact with desktop systems. This technology is known in the industry as computer-use agents. Developers are also applying reinforcement learning methods. Chief rival Anthropic addresses similar challenges in a different way.
Anthropic uses an entirely different technical approach. The competitor simply rents Mac mini capacity from Amazon. The Amazon Web Services cloud service provides the necessary resources. Apple unveiled an updated device lineup on August 25. The new computers are powered by modern chips. The company's engineers developed the M5 Ultra and M6 processors. An analysis of the technical specifications points to a clear division of roles. The computers perform different tasks in modern AI labs. The compact Mac mini form factor is highly convenient for developers.
The company can deploy thousands of independent working environments. AI agents need real operating systems for training. The neural networks learn to click a mouse and type text. The models successfully master control of graphical user interfaces. The Mac Studio M5 Ultra supports up to 512 GB of memory. This capacity is needed to run heavy language models. The flagship model has a memory bandwidth of 1.2 TB/s. The computer can fit the entire open-source GLM-5.3-Flash model. This neural network contains 320 billion working parameters.
The model can occupy up to 300 GB of RAM. A standard desktop would require ten graphics cards. The flagship RTX 5090 cards have only 32 GB of memory each. A multi-GPU setup demands enormous amounts of electrical power. Tim Cook reported a Mac mini shortage on April 30. The Apple CEO directly named the reason for the device shortfall. Demand was being driven specifically by new agentic AI tools. The company adapted its marketing to meet the needs of neural network developers. The new devices are positioned as solutions for local computing.
Apple hardware does not replace Nvidia GPU clusters. Base models are still trained on powerful graphics cards. Mac computers form an entirely new infrastructure layer. They create a virtual office for trained neural networks. AI agents practice completing real-world work tasks. Neither OpenAI nor Anthropic has disclosed exact figures. The companies have kept secret the number of devices purchased and the deal values. Apple is firmly establishing itself in the neural network infrastructure market.