OpenClaw, the open-source AI agent framework created by developer Peter Steinberger and now backed by OpenAI after a bidding war with Meta, has accumulated over 323,000 GitHub stars.
This lets individuals and small teams continuously run AI agents locally, connecting directly to files, apps, and messaging without routing data through the cloud. The unofficial reference hardware to run it, almost by accident, became the Mac Mini.
Cook accepted the change directly. “Both of these are amazing platforms for AI and agentic tools,” he said of the Mac mini and Mac Studio. “Customer recognition is happening faster than we predicted.” Apple’s supply chain, designed for typical consumer upgrade cycles, was not built for developers purchasing multiple units at a time as an infrastructure.
The technical reason for the sudden rise in importance of the Mac Mini is only one limitation, the Video Random Access Memory (VRAM). Even Nvidia’s best consumer GPU, the RTX 5090, which costs $1,800, can provide a maximum of 32GB of video memory. A language model with more than 32 billion parameters will not work optimally on such a card.
When crossing over to PCIe, spilling over into less efficient system memory, performance degrades. The 70B-parameter language model on Nvidia architecture will require multiple cards.
On Apple silicon, the CPU, GPU, and Neural Engine share the same physical memory space with no VRAM and no PCI Express bottleneck.
A Mac mini with 64GB of integrated memory can accommodate 70 billion parameter models while the RTX 5090 can never manage such a feat. The more powerful M4 Ultra found in the Mac Studio has a capacity of up to 192GB, allowing a 100B-parameter model using a single consumer computer.
The entry-level Mac mini, priced at $599, is not currently available in the United States. eBay scalpers sold base-model machines for double the price shortly after release. The problem arises amid a global shortage of memory chips, with IDC estimating that global PC shipment volumes will decline by 11.3% in 2026 due to competition for RAM resources by hyperscaler AI servers with Apple Computer.
The upcoming M5 processor update in 2026 may relieve some pressure on memory sources but does nothing for today’s customers who face two choices: longer waits or scalper prices.
