Glonce

OpenAI and rival labs buy tens of thousands of Mac minis for agents

infrastructureThe Glonce editorsThe Decoder🔥74Glonce rating74Impact68Novelty57Relevance98Surprise63Players87Credibility72
OpenAI and rival labs buy tens of thousands of Mac minis for agents

OpenAI and other AI labs have bought tens of thousands of Mac minis and Mac Studios, according to reporting by The Information cited by The Decoder. OpenAI uses the machines to train computer-use agents, systems meant to handle multi-step tasks on their own. The company wants to buy still more, but the most powerful configurations have been sold out for months because of a memory chip shortage. Anthropic is taking a different route into the same hardware: rather than buying, it is renting Mac minis through AWS.

The interest is not confined to the big labs. The Mac mini is also gaining traction as a machine for running AI locally, a trend the article ties partly to the hype around OpenClaw. Its selling points for long AI workloads are strong chips, memory shared across CPU and GPU, and cooling that holds up under sustained load. Nvidia's DGX Spark is pitched as a compact rival, but it takes a different technical approach, relying on dedicated GPU power through CUDA and Tensor cores instead of Apple's unified memory design.

A piece of open-source software called Exo lets users chain several Macs into a cluster so they can run large models locally rather than on a single machine. Peter Voell, who previously worked on OpenAI's computing infrastructure team, is going further and building an Apple-hardware cloud service called Mount Thor. Apple, meanwhile, is a direct beneficiary of the demand: its Mac revenue climbed nearly 29 percent to $10.4 billion in the June quarter.

Key facts

  • OpenAI and other AI labs have bought tens of thousands of Mac minis and Mac Studios to train computer-use agents that handle multi-step tasks autonomously.
  • A memory chip shortage has left the most powerful Mac configurations sold out for months, even as OpenAI wants to buy more.
  • Anthropic is renting Mac minis through AWS rather than buying them outright.
  • Open-source software called Exo lets users cluster multiple Macs together to run large models locally, and Peter Voell, a former OpenAI infrastructure engineer, is building an Apple-hardware cloud service called Mount Thor.
  • Apple's Mac revenue rose nearly 29 percent to $10.4 billion in the June quarter, with AI-driven demand as one contributing factor.

Why it matters

Training computer-use agents at scale takes a lot of compute, and the standard path runs through data center GPUs. This story shows labs reaching for a second option: Apple's Mac minis and Mac Studios, bought in bulk rather than as developer workstations. That a memory chip shortage can leave the top configurations sold out for months, with OpenAI still wanting more, says the demand for agent-training hardware is currently outrunning even a supplier as large as Apple. It also marks Apple silicon's unified memory architecture as a real alternative to GPU clusters for this specific workload, not just a curiosity for hobbyists.

Who it affects

OpenAI and unnamed rival AI labs are the buyers, spending on hardware rather than only on cloud GPU capacity. Anthropic sits alongside them but takes a lighter-commitment path, renting Mac minis through AWS instead of purchasing. Apple benefits directly through Mac sales, which the article ties to a near 29 percent jump in Mac revenue for the June quarter. Nvidia faces an unusual comparison here, since its DGX Spark is framed as the closest rival product despite using a different architecture. Peter Voell, formerly on OpenAI's infrastructure team, is now building a business, Mount Thor, around renting out Apple hardware in the cloud, and local AI users benefit from the open-source Exo software that lets them cluster Macs for their own model runs.

How to use it

For labs and developers who want to try the same approach, the Mac mini or Mac Studio's strong chips, shared memory, and solid cooling are what the article credits for the boxes' fit with long-running AI workloads. Exo is the specific tool named for linking several Macs into a cluster to run large models locally instead of relying on a single device or a cloud GPU. Buyers should expect availability limits: the article states the most powerful Mac configurations have been sold out for months due to the chip shortage, so this is not a purchase to plan around short lead times right now.

How solid is it

The account rests on reporting from The Information, relayed here by The Decoder rather than reported firsthand. The scale of the buying is given only as "tens of thousands" of units, with no exact count, no named labs beyond OpenAI and Anthropic, no purchase dates, and no prices disclosed. Apple's own revenue figure, the near 29 percent rise to $10.4 billion in the June quarter, is a company-reported number and the more concrete data point in the piece, though the article does not say whether that growth is measured year over year or quarter over quarter.

Risks and caveats

The core claim about lab purchasing volume is an approximation, not an audited figure, and it is not clear how much of Apple's Mac revenue growth this AI buying specifically accounts for versus other demand. The memory chip shortage that has sold out the most powerful configurations could keep constraining how much labs can actually acquire, regardless of how much they want. The article also does not indicate how established Exo or Voell's Mount Thor service are, so their reach beyond early adopters is unclear.