In a world dominated by Nvidia GPU clusters, a quiet but significant transformation is underway. My analysis of recent purchases and market signals indicates that Apple is becoming a key, though not obvious, player in the infrastructure for training agentic artificial intelligence. This is not about replacing data centers, but about creating a fundamentally new layer of computing.

Over the past few months, OpenAI has acquired tens of thousands of Mac mini and Mac Studio computers. This is not a spontaneous decision, but a strategic move aimed at solving highly specialized tasks. The devices are used to train AI agents to interact with desktop operating systems—so-called computer-use agents—as well as for reinforcement learning algorithms. My expertise suggests that this class of tasks requires not so much raw computing power as a realistic and isolated environment for neural networks to "practice."

Notably, OpenAI's main competitor, Anthropic, is solving a similar problem in a different way, renting Mac mini capacity from Amazon Web Services. This confirms that demand for such computing is colossal, while supply in the market is limited.

On August 25, shortly before the purchase information was published, Apple unveiled an updated lineup of devices based on M5 Ultra and M6 chips. An analysis of the stated specifications clearly demonstrates the division of roles in AI labs:

  • Mac mini M6. The compact form factor makes them ideal for deploying thousands of independent working environments. AI agents need real operating systems to learn how to click, type, and manage interfaces.
  • Mac Studio M5 Ultra. Support for up to 512 GB of unified memory provides sufficient headroom for locally running and hosting the "heaviest" language models.

Let me give a concrete example: the flagship Mac Studio M5 Ultra with 512 GB of memory and 1.2 TB/s bandwidth can entirely fit the open-source LLM GLM-5.3-Flash with 320 billion parameters. Depending on quantization, the model can take up to 300 GB of RAM. For comparison, on a regular desktop, such a volume of video memory would require a build of ten flagship RTX 5090 graphics cards, which entails enormous electricity costs and a significant loss of bandwidth.

Remarkably, as early as April 30, 2026, Tim Cook, who served as Apple's CEO, publicly acknowledged the shortage of Mac mini and Mac Studio, directly citing agentic AI tools as the reason. The company has even adapted its marketing, positioning the new devices as solutions for "continuously running local agentic computing."

The key conclusion I draw from this situation is that Apple hardware is not intended for training base models—that role is still performed by Nvidia GPU clusters. Instead, it forms a new infrastructure layer—a virtual "office" where trained neural networks practice performing real tasks. Neither OpenAI nor Anthropic has disclosed the exact number of devices or purchase amounts, which only fuels interest in this new but rapidly growing niche.

My comment: The market underestimates Apple's role in the AI race. If Nvidia sells the "brain" for AI, then Apple could become the supplier of the "hands and eyes" needed to turn that intelligence into real actions. This is a strategic advantage that competitors will find hard to copy.