The world of artificial intelligence is experiencing an unexpected shift. According to a report by The Information, OpenAI has purchased tens of thousands of Mac minis and Mac Studios in recent months. The goal is to train its AI agents to operate a computer like a human: clicking, navigating menus, filling out forms, and performing complex tasks. This strategy, which has surprised many, demonstrates that Apple hardware is no longer just for consumers and creatives, but has become a key component in the infrastructure of major AI labs.
The news, also reported by outlets such as Cult of Mac, Moneycontrol, and 24/7 Wall St., indicates that these computers are being used in a very specific phase of training: reinforcement learning. In this process, AI models interact with real graphical interfaces, testing actions and receiving rewards or penalties based on the outcome. This method demands a significant amount of memory and the ability to run thousands of simulations in parallel, something for which Macs with Apple Silicon chips appear to be particularly well-suited.
Why Apple Silicon is key to reinforcement learning
Training an AI agent that must use a computer is very different from training a traditional language model. Instead of absorbing large amounts of text, the agent has to interpret what you see on the screen, decide what action to take, and check if it worked.Each attempt is scored, and the model adjusts its strategy based on that feedback. For this to be possible, thousands of independent sessions are needed, each running simultaneously with its own state and context.
This is where Apple's Unified Memory Architecture (UMA) comes into play. Unlike traditional GPU-based systems, where video memory and RAM are separate and data has to travel back and forth, in Macs the CPU, GPU, and Neural Engine share the same memory block. This reduces bottlenecks and allows the model to quickly access the information it needs at any given time. As several analyses point out, this feature is crucial for the agent to maintain a consistent representation of the screen, instructions, and the results of its actions.
Another advantage of the Mac mini and Mac Studio is their cooling system. Unlike MacBooks, which are thinner and lighter, these desktop computers have dedicated fans that allow them to maintain stable performance for hours or even days of intensive training. This is crucial for reinforcement learning, which can require very long sessions without the computer overheating and slowing down.
OpenAI buys, Anthropic rents: two strategies for the same end
The news also reveals that OpenAI is not the only company interested in this approach. Anthropic, backed by Google and Amazon, has reportedly opted for a different strategy: instead of buying the equipment, is leasing Mac mini capacity through Amazon Web Services (AWS)This difference is important because it demonstrates that you don't need to invest a fortune in hardware to test whether this architecture works. You can rent the capacity, measure the performance, and then decide whether it's worth buying or continuing to rent.
The interest of these labs in Macs doesn't mean they will replace Nvidia's large GPU clusters. Large-scale language model training still relies on those specialized accelerators. However, for intensive inference tasks and for training agents that interact with interfaces, Macs offer a more economical and efficient alternative. As analyst Shay Boloor explains, Apple's computers are used for the reinforcement learning phase, while GPUs remain essential for pre-training.
Demand puts strain on the supply chain and Apple reacts
This Mac craze hasn't gone unnoticed in the supply chain. According to several reports, Mac mini and Mac Studio configurations with more memory have been sold out for months. And delivery times for custom versions have stretched for weeks. The global shortage of DRAM and NAND memory, driven by demand from data centers, has exacerbated the situation. Suppliers are prioritizing large hyperscalers, and Apple's orders are competing in the same queue.
To address this situation, Apple accelerated its refresh schedule. On August 25, 2026, the company introduced the new Mac mini with M6 and M5 Pro chips, as well as the Mac Studio with M5 Max and M5 Ultra. This unusually early move responds to the need to replenish the market with more powerful and efficient chips. The new models include significant improvements in performance and memory, with options reaching up to 512GB of unified storage in the Mac Studio M5 Ultra.
Apple's decision to move up the launch has also been interpreted as a response to demand pressure. The company hasn't officially confirmed the acquisition of OpenAI, but delivery times and stock shortages suggest that institutional demand is real. In fact, revenue from the Mac division grew 29% year-over-year in the last quarter, reaching $10.400 billion, making it Apple's fastest-growing business line.
Nvidia is watching closely as Apple positions itself in the enterprise market
The rise of Macs in the field of local AI has not gone unnoticed by Nvidia. According to a source close to the company, Nvidia already considers Apple its biggest competitor in local AI processingTo counter this threat, Nvidia launched the DGX Spark late last year, a desktop computer with a design similar to the Mac mini, equipped with a Grace Blackwell chip and up to 128 GB of unified memory. This product is aimed directly at the desktop AI inference market.
Apple, for its part, is trying to capitalize on this moment to solidify its position in the enterprise market. In June, the company held a private event at Apple Park called "Business at the Park," attended by executives from Disney, Ford, and Anthropic co-founder Jared Kaplan. At this event, Apple highlighted the advantages of its hardware for processing AI tasks locally, with the Mac mini taking center stage. However, the company still lacks a dedicated sales team for the enterprise AI segment, something some former executives have pointed to as a missed opportunity.
Supply shortages are also giving rise to new business models. Peter Voell, a former OpenAI employee, has founded Mount Thor, a cloud computing company based on Apple hardware that is still operating in stealth mode. Other companies, such as Namespace Labs, have resorted to racking MacBook Pros to meet customer demand. This emerging ecosystem demonstrates that interest in Macs as AI infrastructure extends far beyond large laboratories.
In short, the news that OpenAI has bought tens of thousands of Mac minis and Mac Studios to train its AI agents marks a turning point in the industry. Apple Silicon's unified memory, energy efficiency, and ability to run thousands of sessions in parallel These devices have become a valuable tool for reinforcement learning. Although neither Apple nor OpenAI has officially confirmed the deal, sales figures, stock shortages, and early releases suggest this trend is real. It remains to be seen how Nvidia will respond and whether Apple will capitalize on this opportunity to solidify its position in the enterprise AI market.
