Mac mini supply crunch shows agents need dedicated hardware

Apple’s Mac mini has started to look less like a consumer desktop and more like infrastructure for always-on AI agents. That shift showed up on April 30, when Apple CEO Tim Cook used the company’s second-quarter 2026 earnings call to talk about Mac mini and Mac Studio supply in unusual detail.

Cook told analysts that both products were sold out across multiple configurations and said the supply-demand balance was still “several months” away. He said the demand was being driven by “agentic AI tools and workflows,” according to Apple’s earnings call. CFO Kevan Parekh went further, citing Perplexity by name as a developer choosing Mac as the platform for enterprise-grade AI assistants.

A week later, on May 7, Perplexity released Personal Computer as a new Mac app available to all Mac users. The company kept the full agent experience behind paid Pro, Max, and Enterprise tiers. In its own documentation, Perplexity said Mac mini is the recommended machine for running the agent 24/7.

That matters because Perplexity is not the only project converging on the same hardware. OpenClaw recommends Mac mini in its official hardware guide and through its community. Hermes Agent is less specific about Apple hardware, but its local-first Ollama path makes Apple silicon a natural fit for people who want to run models locally.

The pattern is bigger than one product launch. Over roughly two weeks, three different agent runtimes ended up pointing developers to the same answer to a question that has been circulating for months: where should an always-on agent live?

The appeal of Mac mini is practical. A persistent agent is not a normal chat session. It runs while the user is away from the keyboard, receives messages, drafts code, checks inboxes, and executes scheduled tasks. That kind of workload needs a machine that stays on, runs quietly, integrates with the user’s existing operating system, and does not cost more than renting cloud infrastructure over time.

Apple’s 2024 M4 Mac mini is listed at 4 watts at idle, which makes it cheap to keep on compared with a typical server or desktop. For that reason, the machine that Apple originally positioned for small offices and home entertainment is being repurposed by software builders as a dedicated agent host.

OpenClaw is the clearest example of that bottom-up shift. The project passed 300,000 GitHub stars by early April 2026 and later moved to an independent foundation backed by OpenAI after Peter Steinberger joined the company. Its own documentation calls the Mac mini “quietly the best hardware for running OpenClaw” and highlights macOS integrations such as iMessage, Shortcuts, Apple Notes, Reminders, and Keychain.

That integration is the key point. An OpenClaw agent on a Mac mini is not just running on Apple silicon; it is connected to the same identity, calendar, and messaging surface the user already uses. Community guides have already standardized around a headless Mac mini running under a non-admin account, accessible remotely through Tailscale, with FileVault enabled and startup settings configured to keep the agent alive.

Hermes Agent shows the same hardware story from a different angle. Nous Research says the project crossed 100,000 GitHub stars within weeks of release and now has more than 800 contributors. Hermes is designed to remember across sessions, create and modify its own skills, and support more than 200 models through providers including OpenRouter, OpenAI, Anthropic, Nous Portal, and Google.

The project is not Mac-specific, but its Ollama integration makes Apple silicon an attractive local-first option. Users who want to avoid recurring API spend often gravitate toward Mac mini because the machine’s unified memory architecture makes local inference more practical. According to the source material, a Mac mini with 32 GB can run quantized 30B-parameter models for inference at acceptable token rates, depending on context window and runtime.

Perplexity’s release makes the shift visible outside open source. The company said Personal Computer takes “Computer out of the cloud-only world and onto the device where most of your real work already takes place.” It also described an always-on use case in which a task starts on an iPhone, runs on a Mac mini, and notifies the user when approval is needed.

Apple appears to be treating the demand as durable. The company said in February that future Mac mini production would move to a new factory in Houston, which suggests a longer-term supply plan. Decrypt reported that higher-RAM Mac mini and Mac Studio configurations were seeing wait times of 16 to 18 weeks, while some listings had gone missing from Apple’s store and resellers were listing base models at much higher prices.

For now, the signal is clear: developers building persistent AI agents are settling on Mac mini as a shared host, and Apple is already seeing that choice show up in supply data.