A compact desktop computer is becoming part of the AI infrastructure story. According to The Information, OpenAI and other AI labs have bought tens of thousands of Mac minis and Mac Studios as they work on computer-use agents.
The shift matters because these systems are not being described as ordinary developer machines. OpenAI uses them to train computer-use agents that autonomously handle multi-step tasks, a category of AI work where the machine environment itself is central to what the agent must learn to operate.
The Mac mini moves into AI training
OpenAI is using Mac minis and Mac Studios for a specific kind of AI work: training computer-use agents. These agents are built to handle multi-step tasks on their own, which means the training environment has to support long-running, interactive work rather than only isolated prompts or short experiments.
The source does not describe the internal training setup in detail. What it does make clear is the scale of demand. OpenAI and rival AI labs are not buying a few machines for testing; they have bought tens of thousands of Mac minis and Mac Studios.
That scale suggests the machines are filling a practical role in AI operations. For computer-use agents, the computer is not just a place where code runs. It is also the kind of interface the agent is expected to understand, navigate, and act within across more than one step.
Supply is now part of the story
OpenAI wants more machines, but availability is a constraint. The most powerful models have been sold out for months because of a memory chip shortage, according to the source article.
That detail is important because it shows demand is running into a hardware bottleneck. AI labs may want to expand these setups, but the availability of the most capable Mac configurations is not unlimited.
Anthropic is also involved in this trend. The company is renting Mac minis through AWS, which points to another route for accessing the hardware without relying only on direct ownership.
Taken together, the buying by OpenAI and other labs, the rentals by Anthropic, and the shortage around the most powerful models all point in the same direction: compact Apple machines have become useful enough for AI work that supply now matters.
Why Mac minis are attracting AI users
The Mac mini is also gaining attention beyond the largest AI labs. The source describes it as picking up steam as a local AI computer, partly fueled by the OpenClaw hype.
The appeal comes from the hardware mix. Strong chips, shared memory, and solid cooling make the Mac mini suitable for long-running AI workloads. For local AI users, that combination can be important because workloads may need to run for extended periods on a small machine.
This does not mean the Mac mini is the only compact AI option. Nvidia's DGX Spark is described as a compact alternative, but it follows a different design path.
The difference is architectural. Nvidia's DGX Spark leans on dedicated GPU power with CUDA and Tensor cores, while Apple uses a unified memory architecture. The source presents these as different approaches rather than direct copies of one another.
- Mac mini and Mac Studios: used by OpenAI and other AI labs for computer-use agent training.
- AWS rentals: a route Anthropic is using to rent Mac minis.
- Nvidia's DGX Spark: a compact alternative based on dedicated GPU power with CUDA and Tensor cores.
- Apple's unified memory architecture: a different hardware approach from Nvidia's compact system.
Clusters and Apple-based services are emerging
The interest in Macs for AI is not limited to single-machine use. The open-source software Exo lets users link multiple Macs into a cluster to run large models locally.
That clustering angle changes how the machines can be thought about. A single Mac mini may be a local AI computer, but several Macs linked together become a larger local setup for running large models.
There is also activity around Apple-based cloud infrastructure. Peter Voell, formerly on OpenAI's computing infrastructure team, is building an Apple-based cloud service called Mount Thor.
The source article also notes a broader business signal from Apple. Apple's Mac revenue jumped nearly 29 percent to $10.4 billion in the June quarter.
That revenue figure is not presented as being caused only by AI demand. But placed alongside the reported purchases by AI labs, the AWS rentals, local AI interest, Exo clustering, and Mount Thor, it shows why the Mac is now part of the AI hardware conversation in a more visible way.
What this signals for computer-use agents
The clearest takeaway is that computer-use agents need infrastructure that looks different from a narrow model-training story. OpenAI is using these machines to train agents that autonomously handle multi-step tasks, and other AI labs are buying the same class of hardware at large scale.
That makes the Mac mini more than a consumer desktop in this context. It is being used as part of the practical machinery behind agent training, local AI workloads, clustering experiments, and cloud services built around Apple hardware.
The trend is still constrained by supply. The most powerful models have been sold out for months because of a memory chip shortage, so wider adoption depends not only on interest from AI labs and local users, but also on whether enough suitable machines are available.