OpenAI and other AI laboratories have bought tens of thousands of Mac minis and Mac Studios to train agents that perform multi-step computer tasks, according to The Information. OpenAI reportedly wants additional machines, but high-end configurations have been sold out for months amid a memory-chip shortage. Anthropic also rents Mac minis through Amazon Web Services.
Apple hardware is useful for this work because the agents are learning to operate desktop environments, while Apple’s unified memory lets processors and graphics cores share a large pool of data. The compact systems also offer cooling suited to long-running workloads. Nvidia’s DGX Spark targets a similar desktop-AI market with dedicated GPU features, CUDA and Tensor cores instead of Apple’s unified architecture.
Demand extends beyond frontier labs. The open-source Exo project can link several Macs into a cluster to run models locally, and former OpenAI infrastructure engineer Peter Voell is building an Apple-based cloud service called Mount Thor. Apple’s Mac revenue rose nearly 29 percent to $10.4 billion in the June quarter.
The purchases show that AI infrastructure is not limited to data-center GPUs. Training software to use ordinary graphical applications also creates demand for fleets that closely match the computers and operating systems those agents will encounter.