OpenAI, Anthropic turn to Apple's Mac minis to train AI agents

Beyond Nvidia's GPU clusters, leading AI labs are deploying thousands of Mac mini and Mac Studio machines for reinforcement learning workloads, driving a surge in Apple's Mac revenue amid ongoing high-end memory shortages.

AI developers have traditionally leaned on massive clusters of Nvidia GPUs to train large language models, given the raw computing power needed to process enormous datasets and run demanding AI workloads. But a new trend is emerging: companies are increasingly turning to Apple's Mac mini and Mac Studio desktops for a specific category of AI work.

According to a report by The Information, OpenAI has purchased tens of thousands of these Mac units to help train AI agents capable of performing tasks autonomously and interacting with software on their own. Anthropic, meanwhile, is taking a different route to the same hardware, reportedly renting Mac minis through Amazon Web Services (AWS) for comparable workloads.

Built for reinforcement learning, not a GPU replacement

The Macs are said to be primarily used for reinforcement learning (RL), a training method where AI systems improve through repeated trial and error — attempting actions, evaluating outcomes, and adjusting behaviour accordingly. This approach is central to building AI agents that can operate computers the way a human user would.

Industry watchers note this shift doesn't signal Nvidia GPUs being phased out of data centres. Instead, companies appear to be matching different types of AI workloads to the hardware best suited for them, with Mac hardware filling a specific niche that traditional GPU servers aren't optimised for.

Unified memory design gives Macs an edge

Central to the appeal of Apple's M-series chips is their Unified Memory Architecture (UMA), which allows the CPU, GPU and Neural Engine to draw from a single shared memory pool rather than maintaining separate memory banks. This cuts down on unnecessary data transfer between components and lets larger models run more efficiently on a single machine.

Beyond performance, the Mac mini and Mac Studio offer practical benefits for sustained AI workloads: they are compact, energy-efficient, and engineered to handle heat effectively during extended use — qualities that make them well suited to reinforcement learning, agent training and testing.

The growing demand from AI companies appears to be feeding directly into Apple's hardware business. Mac revenue reportedly climbed nearly 29% year-on-year in the most recent quarter, touching $10.3 billion, while high-end configurations have remained out of stock for months amid ongoing shortages of memory components.