The Information reports OpenAI has quietly bought tens of thousands of Mac minis and Mac Studios over the past few months. Not for employees — for training.
Why Macs, not GPUs
Computer-use agents learn by doing: run inside a real OS, look at the screen, click, get feedback, repeat millions of times. That reinforcement learning loop is memory-bound and barely parallel — the opposite of transformer pre-training. Apple’s unified memory, where CPU, GPU and neural engine share one pool, fits this workload better than a rented GPU cluster, and costs less.
The side effect is real: Mac Studio delivery times reportedly stretched from two weeks to nearly two months, and Apple pushed out refreshed models earlier than usual.
Buy vs rent
Anthropic runs the same playbook — but rents Mac minis through AWS. OpenAI owns the metal. One bets asset-heavy, one stays light, both agree Apple Silicon is now agent-RL infrastructure.
The best part: Apple doesn’t even have an enterprise AI sales team. It stumbled into this market by accident — the two hottest AI labs showed up writing checks, and nobody at Cupertino was assigned to answer the door.
You Might Also Like
- Anthropic Gives Claude Computer use on mac and you can Control it From Your Phone
- Ollama mlx on Apple Silicon 1810 Tokens sec Prefill and the end of Llama cpp on mac
- Tinygpu George Hotz got Apple to Sign an Nvidia gpu Driver for mac
- Openai Codex Record and Replay Demo a mac Task Once the Agent Repeats it Forever
- Openai gpt Realtime 2 1 gpt Realtime 2 1 Mini cut Voice Agent Latency by 25
Top AI Product 
Leave a comment