Top artificial intelligence (AI) research laboratories are snapping up tens of thousands of Apple Inc. desktop computers to train next-generation autonomous software.
OpenAI has acquired an estimated tens of thousands of Mac mini and Mac Studio units in recent months, according to reports from The Information.
Rather than relying exclusively on massive NVIDIA Corp. GPU clusters, the AI firm is using the display-less Apple desktops as dedicated workhorses for reinforcement learning and training AI designed to navigate digital interfaces, write code, and manage multi-step workflows like a human user.
Meanwhile, rival Anthropic is reportedly taking a similar approach by renting Mac mini capacity through Amazon Web Services.
The sudden enterprise pivot centers on Apple’s Unified Memory Architecture (UMA).
Traditional server setups divide system RAM and graphics memory, creating data transfer bottlenecks. Apple’s M-series architecture allows the CPU, GPU, and Neural Engine to access a single pool of high-speed memory on the same chip.
For AI agents that constantly interact with operating systems — observing screens, taking actions, and analyzing feedback in rapid feedback loops — this shared memory pool offers significant performance advantages over standard parallel computing setups.
The buying spree is altering both the hardware market and consumer availability.
High-RAM configurations of the Mac Studio and Mac mini have seen delivery estimates stretch from two weeks to nearly two months amid an industry-wide RAM shortage.
In response to unprecedented demand, Apple broke its traditional autumn release cadence to unveil updated Mac models in late August.
The refreshed Mac mini features Apple’s M6 chip, while the new Mac Studio offers M5 Max and M5 Ultra configurations, with enhanced support for linking multiple units to handle larger localized AI workloads.
Apple’s Mac segment posted $10.4 billion in quarterly revenue, reflecting a 29% year-over-year increase driven in part by unexpected enterprise adoption.
The trend underscores a fundamental shift in AI development. While massive GPU clusters remain essential for pre-training large language models from scratch, training autonomous agents requires environments optimized for sustained, memory-heavy operating system interactions.
The surge in demand presents a unique challenge for Apple. Despite selling machines originally designed for creative professionals and everyday consumers, the company now finds itself serving as critical infrastructure for leading AI labs without a dedicated enterprise division.
Whether Apple adapts its long-term strategy to support large-scale enterprise clusters or remains a passive beneficiary of the AI boom, consumer desktops have unexpectedly become a primary battlefield for the future of agentic AI.
