The Apple Mac mini has long been the unassuming, easily overlooked desktop computer tucked away in a corner of Apple’s store. Practical and priced in line with Apple’s standards, it has rarely attracted attention from the artificial intelligence community—until OpenClaw arrived.
On Thursday, Tim Cook told analysts that the Mac mini and Mac Studio are sold out and that this situation may last for several months. “These two products are excellent platforms for AI and intelligent agent tools,” he said. Apple's Q2 2026 Earnings Call “And customers are recognizing this faster than we expected.”
Apple underestimated the demand from developers for these machines, especially as scarcity disrupted the market.
Mac revenue was $8.4 billion this quarter, up 6% year-over-year. While the growth isn’t dramatic, the constraint is supply, not demand. Higher-memory configurations of the Mac mini and Mac Studio have been delayed, and some models have even been removed from the Apple Store.
The base Mac mini priced at $599 is out of stock in the United States. Currently, neither delivery nor in-store pickup is available. The upgraded configuration with 64GB of memory has a lead time of 16 to 18 weeks. Mac Studio models with 512GB of unified memory have been completely removed from the store. Scalpers on eBay quickly capitalized on this, raising the price of the base model to nearly double the retail price.
What is the catalyst for all of this? The rise of OpenClaw and memory-intensive agents.
An open-source AI agent framework—built by Peter Sternberg—is now backed by OpenAI. Following a competition with Meta, the project has garnered over 323,000 stars on GitHub, becoming the fastest way for individuals and small teams to run persistent AI agents locally. The unofficial reference hardware for running this project almost instantly became the Mac mini.
But this is not the result of marketing efforts.
Most reports about Mac shortages overlook one point: for years, Apple has had almost no influence in the realm of serious AI workloads. Before AI agents became mainstream, people complained that running LLMs, Stable Diffusion, or any other type of home AI software was extremely slow and nearly unusable. At the time, the performance of M2 Macs was comparable only to GPUs from 2019. Apple rejected CUDA and Nvidia’s technologies, instead pushing its own MLX technology, making it as irrelevant in AI as it is in gaming.
NVIDIA dominates the industry because its proprietary GPU programming framework, CUDA, is the foundation of model training and inference. The entire AI technology stack is built around it. Apple had no product at the time that could compete with it. No one would think of using a Mac for local inference.
But CUDA has a little-known secret: memory limitations.
Even the best consumer-grade GPU from NVIDIA, the RTX 5090, has only 32 GB of VRAM. This is a hard limitation—models exceeding 32 GB of VRAM cannot run at full speed on this GPU; part of the data must be stored in slower system memory and transferred over the PCIe bus, significantly reducing performance. To run a complex model with up to 70 billion parameters on NVIDIA hardware, you need multiple GPUs, a server rack, substantial power consumption, and an investment of thousands of dollars.
Apple's Unified Memory Architecture solves this problem in ways CUDA cannot. On Apple Silicon chips, the CPU, GPU, and neural engine share the same physical memory pool—there is no separate video memory and no PCIe bus to cross. A Mac mini with 64GB of memory can load a 70-billion-parameter model, while an RTX 5090 graphics card priced at $1,800 cannot even handle it.
The M4 Ultra chip—the core of the high-end Mac Studio—supports up to 192GB of unified memory. This is enough to locally run a 100-billion-parameter model on a single machine, without needing a server or paying monthly cloud fees.
OpenClaw makes this trade-off clear. Since it runs agents locally—connecting to your files, applications, and messaging—you need a machine capable of handling inference loads without renting cloud computing resources. A Mac mini with 32GB of unified memory can easily run a 30-billion-parameter model, while a Mac Studio with 128GB of memory can handle models that most developers couldn’t have processed a year ago without enterprise-grade GPU clusters.
A Mac computer that runs slowly but can load powerful AI models is far better than a high-performance Nvidia GPU that cannot load the model at all.
As a result, developers are beginning to buy Mac Minis the way they once bought Raspberry Pis—purchasing multiple units at once, treating them as infrastructure rather than personal computers. Apple’s supply chain was never designed for this model.
In addition, a broader memory shortage has exacerbated this issue. IDC expects global PC shipments to decline by 11.3% in 2026 due to memory chip shortages driven by demand for AI servers. Apple is currently competing with hyperscale data center builders for the same memory supply.
Cook indicated that the supply and demand balance for the Mac mini and Studio may take “several months” to normalize. The M5 chip is expected to be updated by late 2026, which may alleviate supply pressures—but current buyers can either wait or pay premium prices from scalpers.
In 2026, the release of the Mac mini is more urgent than at any point in its 20-year history—and all it needs is help from an open-source project completely unrelated to Apple to achieve this.