🖥️Apple Is Building Harbor to Turn Macs Into Private AI Servers
Your Mac stack might become a private AI cluster
TL;DR
Apple is testing Harbor, a Mac app that runs open-weight models locally and pools memory across multiple Macs over Thunderbolt 5. Nothing leaves the hardware. If you already own a few Macs, you may soon have a private inference rig without renting cloud GPUs.
Apple is quietly building Harbor, a Mac app that turns your machines into a private AI inference server. Your data never leaves the hardware, and nothing touches the cloud. For anyone who's been wincing at API bills, that's a pretty good pitch. Harbor runs open-weight models directly on Apple silicon, and it can link two or more Thunderbolt 5 Macs into one cluster that pools their unified memory. You can load models bigger than any single machine can hold, or serve more concurrent requests from a desk full of Minis. Setup takes minutes, not hours, and the app detects Thunderbolt 5 peers automatically. Model browsing, downloading, and loading all happen inside the app, so you're not juggling CLI tools. Harbor is in private preview under Apple's Claris brand. Apple hasn't given a public launch date, but a broader release is likely in the coming weeks and months. Many developers have bought Macs this year for local AI work, and Apple is clearly leaning into that use case on purpose. Pricing, supported model formats, and whether clustering holds up beyond two machines are the open questions.

Key Points
Harbor pools unified memory across two or more Thunderbolt 5-capable Macs into a single inference resource, letting you run models larger than any one machine's memory allows.
Keep reading — it's free
Enter your email to keep reading — plus the best of AI & tech, daily. Free, forever.
Already a member? Sign in
Comments
Be the first to comment