Apple Develops Harbor to Cluster 2-Plus Macs for Local AI as Memory Scales to 512 GB
Updated
Updated · Trending Topics SEE · Nov 11
Apple Develops Harbor to Cluster 2-Plus Macs for Local AI as Memory Scales to 512 GB
3 articles · Updated · Trending Topics SEE · Nov 11
Summary
Harbor is described as a Claris private-preview app that turns Apple silicon Macs into private AI inference servers, letting users run open-weight models locally and link multiple machines into one cluster.
Thunderbolt 5 connections and macOS Tahoe 26.2 RDMA support would let Harbor pool memory across two or more Macs, aiming to run models too large for a single system and simplify setup without command-line tools.
Apple has been laying the hardware groundwork with M6 and M5 Ultra Macs offering up to 512 GB of unified memory, a design that has already made Macs popular for local AI workloads.
Meta showed the concept in October by running its 816 billion-parameter Muse Spark Open on two connected Mac Studios, highlighting the kind of large-model use Harbor could make easier.
The app appears aimed at businesses, but pricing, performance and launch timing remain unclear; multi-Mac setups can cost about $38,000, and Apple has not confirmed the product.