The cheapest way to run AI at home just turned into a scalper’s market. According to TechCrunch AI, the $599 M4 Mac mini base model is sold out on Apple’s retail site with no delivery or pickup options, and resellers on eBay are charging up to $979 for refurbished units that cost $599 new last week. TechCrunch AI reports this is the first time the base model has gone fully out of stock, with higher-storage configurations now backlogged until June.
The driver isn’t gaming, isn’t crypto, and isn’t a holiday rush. It’s people running local AI models on their kitchen counter.
Why a tiny desktop became the AI machine
Mac minis hit a sweet spot that PCs and laptops can’t match. They’re power-efficient, run quietly enough to leave on 24/7, and the unified memory architecture handles language models far better than the price tag suggests. TechCrunch AI notes the demand started with the OpenClaw craze and has since spread to alternatives like ZeroClaw, plus tools from Anthropic, OpenAI, and Perplexity Computer.
What stands out here is the substitution pattern. The moment Mac minis sold out, demand jumped to the Mac Studio, which is now also sold out across several configurations. Meanwhile, MacBook Pros with 128GB RAM ship in a few weeks, and the new MacBook Neo is moving in two to three. Buyers aren’t picking the most powerful Apple Silicon they can get. They’re picking the cheapest box that runs a local model well.
The market signal underneath
This is one of the clearest consumer-side signals yet that on-device AI has crossed from hobby into mainstream demand. A few data points worth watching:
- eBay listings show $715 to $795 for “open box” base models and around $700 for “lightly used” units, more than $100 over Apple’s retail price.
- A single “last one” brand-new listing went up at $925.
- Bloomberg reports the shortage overlaps with an industry-wide memory crunch and a planned Mac mini refresh, though refreshes alone haven’t caused shortages historically.
The memory crunch matters because it’s not Apple-specific. DDR5 and HBM pricing has been climbing across the industry as hyperscalers buy aggressively for AI training and inference. When the same memory chips that power data center GPUs also live inside a $599 desktop, scarcity at the top eventually bleeds into the consumer tier.
What this means going forward
Three shifts are worth tracking over the next 12 to 24 months:
- Local AI is becoming a hardware category. Vendors will start marketing devices specifically for on-device inference, the way “gaming PC” became a category in the 2000s. Apple, Framework, and Asus are already positioning here.
- Memory, not compute, is the bottleneck. Model weights live in RAM. Buyers are choosing 16GB to 128GB configurations because that’s what determines which models fit. Expect a price war on unified memory and high-bandwidth RAM in consumer desktops.
- Privacy and cost are pushing workloads off the cloud. Running Claude or similar models locally avoids API fees and keeps data on the machine. For freelancers, small agencies, and developers, the math increasingly favors a one-time hardware purchase over recurring tokens.
Practical takeaways
If you’re a practitioner thinking about local AI infrastructure, don’t chase eBay markups. The premium over retail erases the cost advantage of running models locally. Wait for the refresh, watch the Mac Studio inventory, or look at Framework Desktop and AMD Strix Halo machines that are starting to compete on memory bandwidth for similar prices.
If you’re a business, the signal is bigger than one product. On-device AI is shifting from “interesting demo” to “shipping workload,” and the supply chain isn’t ready. Lock in hardware budgets now, before the next memory squeeze pushes prices another step up.
Full details and the eBay screenshots are in the original TechCrunch AI report.